i

Journal of Geomatics Vol 10 No. 1 April 2016

Journal of Geomatics
(A publication of the Indian Society of Geomatics)

Editorial Board
Chief Editor: Prof. Ajai
(Address for Correspondence: Space Applications Centre, ISRO, Ahmedabad - 380 015)
Phone: +91-79-26913305 (O), 91-02717-235441 (R), Email: [email protected]
Associate Editors:
R. P. Dubey Ahmedabad, Email: [email protected]
Markand P. Oza Ahmedabad, Email: [email protected]

Members:
V. Balaji Hyderabad, Email: [email protected]
Mahesh Chandra New Delhi, Email: [email protected]
A.R. Dasgupta Ahmedabad, Email: [email protected]
P.K. Garg Dehradun, Email: [email protected]
A.K. Gosain New Delhi, Email: [email protected]
Ashok Kaushal Pune, Email: [email protected]
I.V. Murali Krishna Hyderabad, Email: [email protected]
V.N. Patkar Mumbai, Email: [email protected]
S.M. Ramasamy Tiruchirapalli, Email: [email protected]
Aniruddha Roy New Delhi, Email: [email protected]
P.S. Roy Hyderabad, Email: [email protected]
Milap Chand Sharma New Delhi, Email: [email protected]
P. Venkatachalam Mumbai, Email: [email protected]

Advisory Board
Paul J. Curran Vice-Chancellor, Bournemouth University, Poole, UK
V. Jayaraman Bengaluru, India
R. Krishnan Thiruvananthpuram, India
Sugata Mitra NIIT GIS Ltd, New Delhi, India
P. Nag Varanasi, India
M.P. Narayanan President, CSDMS, NOIDA, U.P., India
R.R. Navalgund ISRO H.Q., Bengaluru India
Y.S. Rajan ISRO H.Q., Bengaluru, India
R. Siva Kumar New Delhi, India
Josef Strobl Interfaculty Dept. of Geoinformatics, University of Salzburg, Austria
 ii

Journal of Geomatics Vol 10 No. 1 April 2016

Indian Society of Geomatics

Executive Council 2014 – 2017

President A.S. Kiran Kumar, ISRO, Bengaluru - 560231

Vice-President A.S. Rajawat, Space Applications Centre, Ahmedabad-380015
Pramod K. Verma, MPCOST, Bhopal-462003
Secretary N.S. Mehta, Ahmedabad-380015
Joint Secretary R. Nagaraja, National Remote Sensing Centre, Hyderbad-500037
Treasurer K.P. Bharucha, Space Applications Centre, Ahmedabad-380015
Members K.L.N. Sastry, Space Applications Centre, Ahmedabad-380015
G. Parthasarathy, NGRI, Hyderabad-500007
Shakil Ahmed Romshoo, Kashmir University, Srinagar-190006
P.M. Udani, Inst.of S&T for Adv. Studies & Res., VVNagar- 388120
S. Palria, M.D.S. University, Ajmer-305009

Ex-Officio (Immediate Past President) Shailesh Nayak, MoES, New Delhi – 110003

Secretary: (address for correspondence)

39, Basant Bahar-II, Bopal, Ahmedabad – 380058, India
Email : [email protected] or [email protected]
 iii

Journal of Geomatics Vol 10 No. 1 April 2016

Journal of Geomatics
(A publication of the Indian Society of Geomatics)
Vol. 10 No. 1 April 2016

Research articles
1
1 Flood hazard zoning using analytic hierarchy process: A case study for
Pampa river basin, Kerala, India
N.A. Mayaja and C.V. Srinivasa

2 Performance evaluation of dimensionality reduction techniques on CHRIS hyperspectral 7

data for surface discrimination
Veeramallu Satya Sahithi and Iyyanki V Murali Krishna

3 Accuracy evaluation for online Precise Point Positioning services 12

Ashraf Farah

4 REIS: A spatial decision system for land valuation 19

Himanshu Deshmukh, Rupali Khare, Kamal Pandey and Dheeraj Kumar

5 Unfolding the time relationship of structural events through Landsat data: A case study 24
from Khandia formation, Champaner group, Gujarat
M.A. Limaye

6 Runoff estimation from a tributary of lower Tapi basin using SCS-CN method integrated 29
with remote sensing and GIS data
Sudhakar B. Sharma, Anupam K. Singh and Ajay S. Rajawat

7 Change detection on the Volta river due to the construction of the Kpong dam using 36
remote sensing techniques
E.M. Osei Jnr, A.S. Amoah, B.E.K. Prah, F.K. Anyah and K.A. Addo-Gyan

8 An objective method for detecting night time fog using MODIS data over northern India 40
Sasmita Chaurasia and B.S. Gohil

9 Morphometric changes of the Varuna river basin, Varanasi district, Uttar Pradesh 48
K. Prakash, S. Singh and U.K. Shukla

10 Malaria-risk assessment using geographical information system and remote sensing in 55

Mecha district, West Gojjam, Ethiopia
Emebet Dessalegne, K.V. Suryabhagavan and M. Balakrishnan

11 Object based classification techniques for citrus orchards 65

Kadapala Bharath Kumar Reddy, J. Mohammed Ahamed, R. Hebbara, Uday Raj and Rama
Rao

12 Evaluation of predictive ability of support vector machines and naive Bayes trees methods 71
for spatial prediction of landslides in Uttarakhand state (India) using GIS
Binh Thai Pham, DieuTien Bui, Indra Prakash and M.B. Dholakia
 iv

Journal of Geomatics Vol 10 No. 1 April 2016

13 Early estimation of crop sown area by integrating multi-source data 80

D. Ram Rajak, Rajesh K. Jain and Shibendu S. Ray

14 Formosat-2 with Landsat-8 temporal - multispectral data for wheat crop identification 89
using Hypertangent Kernel based Possibilistic classifier
Rohit Nandan, Ankit Kamboj, Anil Kumar, A. Senthil Kumar and K. Venkata Reddy

15 Crime mapping analysis of Ajmer city -A GIS approach 96

Ravi Sharma, Sarvesh Palria and Parul Bhalla

16 On the quality of orthometric correction determination 102

Raaed Mohamed Kamel Hassouna

Guidelines for "National Geomatics Award for Excellence” v

National Geomatics Awards vii
Fellows and Patron Members x
Instructions for Authors xi
Journal of Geomatics: Advertisement Rates xii
Indian Society of Geomatics: ISG Membership Form xiii

Published biannually by the Secretary, Indian Society of Geomatics on behalf of the Society
Copyright Indian Society of Geomatics
ISG Website: www.isgindia.org
 1

Journal of Geomatics Vol 10 No. 1 April 2016

Flood hazard zoning using analytic hierarchy process: A case study for
Pampa river basin, Kerala, India
N.A. Mayaja1 and C.V. Srinivasa2
1
Department of Civil Engineering, Atria Institute of Technology, Anand Nagar, Bangalore, 560098, India
2Department of Civil Engineering, Global Academy of Technology, Bangalore, 560024, India

Email: [email protected], [email protected]

(Received: Jan 06, 2016; in final form: Mar 06, 2016)

Abstract: The river Pampa is the third largest river of Kerala with a catchment area of 2235 km2. This river almost every
year causes substantial damage to human life, properties and the cropland during monsoon. In this study an attempt is
made to classify the regions in the river basin in order of risk and severity due to floods. The severity of flood hazard in
these locations varies due to various geospatial factors. The hazard due to flood in any particular location and its impact
can be assessed in relative terms by using an analytical approach as applied to a set of geospatial factors ranging from
qualitative to quantitative type. This paper evolves appropriate risk indices for the entire Pampa river basin and classifies
them according to the severity of flood risk using a popular Analytic Hierarchy Process (AHP). The study brought out
that two regions in the river basin fall under very high flood risk category whereas four villages come under high risk
category. It was revealed that highly populated and urbanized regions located in the downstream of this river basin are
more vulnerable to flood hazard.

Key Words: Flood Hazard Index (FHI), Analytic Hierarchy Process (AHP), Pampa river basin, Kerala floods

1. Introduction September) contributing about 60% of the rainfall and

North-East monsoon (October to December), providing
The holy river Pampa (also referred as Pamba) in Kerala about 30% of the rainfall. With a relative humidity of
state is the third largest river (about 176 km) with a 70% to 90%, the study area experiences a temperature
catchment area of 2235 km2. It originates on the in the range of 210 to 360 C. The peak altitude of the
Western Ghats and flows through Kuttanadu, the rice basin is about 1677 m (at the origin of the river) and
bowl of the state and drains into the Vembanadu lake. while flowing through a distance of about 176 km the
The severity of floods caused by this river and river reaches the sea level and finally joins the
consequent disasters are increasing annually. Some Vembanad lake and Arabian sea.
studies have revealed that the recurring incidents of
flood are mainly on account of the human interventions
like deforestation, reclamation, sand mining beside
indiscriminate developmental activities. It has thereby
caused severe damages to the physical and biological
environment of this river system. A GIS based study of
flood-prone areas of Pampa river basin has been carried
out using the ground parameters and satellite imagery
(Mayaja and Srinivasa, 2012). In this paper, appropriate
flood hazard indices for the river basin have been
generated based on the severity due to floods. The study
has made use of the well known Analytic Hierarchy
Process (AHP) (Saaty and Alexander, 1989).

2. Study area and data used

This study focuses the basin of Pampa river

(approximately 2235 km2) which is shown in Figure 1.
The river basin stretches over four districts of Kerala,
viz., Idukki, Kottayam, Pathanamthitta and Alappuzha. Figure 1: Study area
The area extends over dense tropical monsoon forests,
semi-urbanized settlements, one famous pilgrim center The population density of the regions (number of
– Sabarimala and also a rich agricultural (rice) bowl of persons per km2) was obtained from the latest National
Kerala, called Kuttanad. The study area lies between census data. Rainfall data (in mm) was obtained from
76020’ to 76059’ East in longitude and 9019’ to 9039’ the India Meteorological Department and Department
North in latitude. With humid tropical monsoon climate of Irrigation, Government of Kerala. Land use data and
(average annual rainfall 3000 mm with summer rains basin slope were availed from the National Remote
constituting about 10%), the basin experiences two Sensing Centre of Indian Space Research Organization.
distinct rainy seasons, South-West monsoon (June to Data related to types of soil were taken from the official

© Indian Society of Geomatics

 2

Journal of Geomatics Vol 10 No. 1 April 2016

website of the Department of soil survey and soil The factors considered at level 1, 2 and 3 are illustrated
conservation, Government of Kerala. The geomorphic in Appendix 1. The values of the sub-classification at
data and the details of road networks were collected level 3 for level 2 factors is shown in Table 1.
from the Kerala State Remote Sensing Centre,
Thiruvananthapuram. The basin elevation was available Table 1: Level 3 sub-classifications of decision
in the Survey of India toposheets. factors

3. Methodology Level-2 Level 3 sub-factors

factor
3.1 Analytic Hierarchy Process (AHP) Population <1000 1000 2000 >3000
The Analytic Hierarchy Process (AHP) is a multi- density (Low) to 2000 to 3000 (very
criteria decision making statistical technique, which (number of (medium) (high) high)
provides a systematic approach for assessing and persons /
integrating the impacts of various factors, involving km2)
several levels of dependent and independent variables. Annual 2500 to 3000 to 3500 to >4000
It is a statistical tool popularly used in assessing the average 3000 3500 4000 (very
impact of various conflicting factors and computing risk rainfall (Less) (medium) (high) high)
indices. AHP attempts to resolve conflicts and analyze (mm)
judgments through a process of determining the relative Land use- Agriculture Forest Built Waste
importance of a set of activities or criteria by ‘pair-wise’ land cover up land
comparison (Saaty and Alexander, 1989, Saaty 1994, Soil type Hill soil Clayey Laterite Sandy
Saaty and Vargas 2002). This technique has been loam soil soil
effectively used to identify and rank the factors
affecting flood in Kosi river basin (Venkata Bapalu and Slope 3°- 5° 5°-15° 15° - >30°
Sinha, 2014). ( Gentle) (Moderate) 30° (Very
(Steep) steep)
In order to perform AHP analysis, a complex problem Geomorphic Denudati- Lower Coastal Alluvial
is first divided into a number of simpler problems in the factors onal Hills Plateau Plain Plain
form of a decision hierarchy. Once the hierarchy is built, Road High Medium Low Very
the decision makers systematically evaluate its various quality low
elements by comparing them to one another, two at a Elevation 0 - 10 10 to 30 30 to >150
time, with respect to their impact on an element above (metres) (Low) (Medium) 150 (very
it in the hierarchy. In making the comparisons, the (High) High)
decision makers can use concrete data on the elements,
or they can use their judgments about the element’s Appendix 1 shows the hierarchy and the relative
relative meaning and importance. It is the essence of the importance weight of level 2-decision factors (RIWi2)
AHP that human judgments, and not just the underlying arrived at by pair-wise comparison of the decision
information, can be used in performing the evaluations. factors. This was followed by pair wise comparison
within each level 3-decision factor to get the
In AHP computation, the decision vectors are corresponding relative importance weight (RIWi3).
constructed at each level of the hierarchy by pair-wise
comparison of the elements (decision factors). The 4. Flood Hazard Index (FHI)
eigen vectors so formed are then normalised. The
Relative Importance Weights (RIWs) of each decision 4.1 Algorithm
factor is obtained as the sum of the values in the The FHI for each location was determined by
corresponding row of the normalised eigen vector. aggregating RIWs of decision factors at each level of
Similar computation of RIWs is performed at each level the hierarchy. FHI was calculated by multiplying the
of the hierarchy. The RIWs thus computed are assigned RIWs of level 3-decision factor by the associated RIWs
to specific part of the study area. The final solution is of the level 2 factors at each level and summing the
evolved by aggregating the product of RIWs at each values of all grouped elements. As the problem is
level. (Saaty and Alexander, 1989). defined in three level hierarchies, the simplified generic
equation used is as follows:
3.2 Primary decision factors
𝑁2
FHI = ∑𝑖=1[(RIW𝑖 2 ) ∗ (RIW𝑖𝑗 3 )] (1)
The first level of the analysis is the generation of the
Flood Hazard Index (FHI). In level 2 analysis, this study where, FHI = Flood Hazard Index;
considered eight primary decision factors viz. N2 = the number of level-2 decision factor;
population density, rainfall, land use, soil type, basin 2
slope, geomorphic factors, quality of roads and RIWi = Relative importance weight of level2 decision
3
elevation. Once the level 2 decision factors are selected, factor i; RIW ij = Relative Importance Weight of level
they are further sub-divided into level 3 sub-factors of 3 sub-factor j of level-2 decision factor i.
smaller class for finer evaluation (Table 1).
 3

Journal of Geomatics Vol 10 No. 1 April 2016

The level 2 normalised relative importance weight Table 3: Flood prone regions in Pampa river basin
matrix computed for various decision factors are given
in Table 2.
No. Zone / region FHI Category
Table 2: Normalised relative importance weights of 1 Thiruvanvandoor 1.87 VERY HIGH
level 2 2 Veeyapuram 1.87 VERY HIGH
3 Chengannur 1.32 HIGH
Sl.No Decision factor Level–2 4 Pandanadu 1.32 HIGH
Relative 5 Niranam 1.32 HIGH
Importance 6 Kuttoor 1.31 HIGH
Weight 7 Chenneerkkara 1.20 MEDIUM
1 Population density 0.35 8 Omalloor 1.20 MEDIUM
2 Rainfall 0.091 9 Aranmula 1.19 MEDIUM
3 Land Use Land cover 0.076 10 Mulakkuzha 1.19 MEDIUM
4 Soil 0.096 11 Mezhuveli 1.19 MEDIUM
5 Slope 0.129 12 Kulanada 1.19 MEDIUM
6 Geomorphic 0.04 13 Mannar 1.19 MEDIUM
7 Road quality 0.038 14 Kozhencherry 1.16 MEDIUM
8 Elevation 0.18 15 Iraviperoor 1.15 MEDIUM
16 Naranganam 1.15 MEDIUM
The level 3 normalised RIW matrices computed for 17 Puliyoor 1.14 MEDIUM
each of the sub-decision factor are given in Appendix 2. 18 Cherukole 1.11 MEDIUM
FHI and consistency ratios at levels 2 & 3 are given in 19 Koipram 1.11 MEDIUM
appendix 3 and 4, respectively. 20 Mylappra 1.09 MEDIUM
21 Mallappuzhassery 1.09 MEDIUM
4.2 Region specific FHIs 22 Ilanthoor 1.08 MEDIUM
The river basin consists of 52 regions (Panchayaths and 23 Ala 1.06 MEDIUM
Municipalities) as per the local administrative 24 Ayiroor 1.05 MEDIUM
classification of Government of Kerala. All the 52 25 Ezhamattoor 1.05 MEDIUM
regions were considered for the purpose of analysis in 26 Thottapuzhassery 1.05 MEDIUM
this study. The FHI in respect of all these regions were 27 Manimala 1.04 MEDIUM
computed as per the algorithm given under 4.1. The
28 Naranammoozhi 1.02 MEDIUM
frequency distribution of FHI is shown in Fig. 2.
29 Ranni-Perunnadu 1.02 MEDIUM
30 Malayalappuzha 1.02 MEDIUM
31 Kottanadu 1.01 MEDIUM
32 Kottangal 1.01 MEDIUM
33 Ranni-Angadi 1.01 MEDIUM
34 Vadaserikkara 1.00 MEDIUM
35 Ranni 0.99 MEDIUM
36 Ranni-Pazhavangadi 0.99 MEDIUM
37 Kadapra 0.67 LOW
38 Peruvanthanam 0.62 LOW
39 Vechuchira 0.61 LOW
40 Bhudhanoor 0.61 LOW
41 Chennithala-Thripperunth 0.61 LOW
42 Thannithode 0.61 LOW
Figure 2: Histogram distribution of FHI 43 Erumely 0.60 LOW
44 Seethathode 0.60 LOW
It can be seen that the 52 FHIs so computed are 45 Elappara 0.59 LOW
predominantly falling in four frequency bandwidths 46 Aurvappulam 0.59 LOW
(Fig. 2). Based on the histogram distribution, the regions 47 Konni 0.58 LOW
of Pampa river basin have been grouped into low, 48 Chittar 0.58 LOW
moderate, high and very high flood-risk category. The 49 Vandiperiyar 0.57 LOW
risk category of these regions along with their respective 50 Peerumade 0.56 LOW
FHI values are shown in Table 3. A schematically 51 Kumily 0.56 LOW
classified risk map of the river basin is shown in Fig. 3. 52 Mundakkayam 0.56 LOW
 4

Journal of Geomatics Vol 10 No. 1 April 2016

in the basin and also implementing regulatory
mechanisms to check uncontrolled and haphazard rate
of urbanisation, which is detrimental to both humans as
well as the river itself. The land use and land cover also
is identified as an influencing factor. Hence proper
environmental and ecological regulations and auditing
can reduce the vulnerability of the region to flood. A
flood mitigation policy based on the above suggestions
can effectively help in reducing the flood risk.

6. Conclusions
Figure 3: Flood risk map of Pampa river basin
Ranking the villages in the flood plain is of utmost
5. Results and discussions importance in flood management planning. In this
study, 52 villages in the Pampa river basin are classified
In this study, a set of composite flood hazard indices for according to their Flood Hazard using Analytic
the Pampa river basin has been worked out by adopting Hierarchy Process. The indices have been derived from
the AHP methodology. The indices have been derived a variety of parameters (factors) ranging from geospatial
from primary decision factors viz: population density, data to population density, rainfall, land use, type of
annual average rainfall, land use, type of soil, slope, soil, slope, geomorphic factors, quality of road and
geomorphic features, quality of roads and elevation. The elevation. The flood prone areas of Pampa river basin
Panchayaths/ municipalities of Pampa river basin have have been classified into four categories viz. low,
been accordingly classified into low, medium, high and medium, high and very high. The analysis reveals that
very high risk categories based on histogram human activities, which result in increased population
distribution of FHI. density, land use land cover changes etc. make the
region more vulnerable to flood hazards. Hence a
The analysis revealed two regions - Thiruvanvandoor, comprehensive basin planning, considering the above
and Veeyapuram - covering about 7.5 km2 of the basin factors only will be effective in mitigating flood hazard
as areas prone to ‘very high’ levels of flood risk. Four
regions, covering about 35 km2 come under the ‘high References
risk’ category of flood where as another 30 regions (395
km2) fall under the ‘medium risk’ category. Remaining Mayaja, N.A. and C.V. Srinivasa (2012). A GIS based
portions of the basin are relatively under ‘low risk’. It flood mapping study for Pampa river basin, Kerala.
is observed that though the regions falling under very Research dissertation of Visweswaraya Technological
high and high level of flood hazard constitute only 0.3% University, Karnataka.
and 2 % respectively of the basin area, these are densely
populated and highly urbanised regions (with density of Saaty, T.L. and J. Alexander (1989). Conflict
population more than 3000 persons / km2), located at the resolution: The analytic hierarchy process. Praeger,
downstream of the river. Further, the land use pattern of New York.
these regions reveals high level of built up area and they
have good network of paved highways. Thus, it can be Saaty, T.L. (1994). Fundamentals of decision making
inferred that the prime reasons of flood hazard are high and priority theory with the analytic hierarchy process.
rate of urbanization and human interventions in this ISBN 0-9620317-6-3.
region. The extensive road networks recently developed
in the river basin also testify this finding. Owing to the Saaty, T.L. and L.G. Vargas (2002). Models, methods,
same reasons the flood occurrence at these regions concepts & applications of the analytic hierarchy
causes more damages to both humans as well as process. Book. ISBN 0-7923-7267-0, Kluwer
infrastructure. Academic.
Amongst the eight primary decision factors influential Venkata Bapalu, G. and R. Sinha (2014). GIS in flood
in causing flood hazard, the most prominent hazard mapping: A case study of Kosi river basin,India.
anthropogenic factor identified is population density. download from internet, http://GISdevelopment.net
This calls for urgent need of an effective urban planning
 5

Journal of Geomatics Vol 10 No. 1 April 2016

Appendix-1
(FHI computation: Decision Hierarchy at levels 1, 2 and 3)

Appendix 2
(The comparison matrix for level- 2)

Factor Population Rainfall LULC Soil Slope Geomor- Roads Elevation

Density phic factors
Population Density 1 3 4 5 5 7 4 5
Rain-fall 0.333 1 0.5 0.333 0.2 4 7 0.25
LULC 0.25 2 1 0.333 0.2 3 4 0.2
Soil 0.2 3 3 1 0.5 2 2 0.5
Slope 0.2 5 5 0.5 1 3 3 0.333
Geomorphic factors 0.143 0.25 0.333 0.5 0.333 1 2 0.25
Roads 0.25 0.148 0.25 0.5 0.333 0.5 1 0.333
Elevation 0.2 4 5 2 3 4 3 1

Appendix 3
(FHI computation)

The FHI for each location was determined by aggregating RIWs at each level of the hierarchy. FHI was calculated by
multiplying the RIWs of level 3-decision factor by the associated RIWs of the level 2 factors at each level and summing
the values of all grouped elements. The level 2 and level 3 Relative Importance Weight matrices computed are shown
below:

Level 2 Population RIW = RIW = RIW =

Density RIW = 0.35 Rainfall 0.091 LULC 0.076 Soil 0.096
Low 0.048 Very High 0.466 Agriculture 0.238 Hill soil 0.145
Medium 0.108 High 0.277 Forest 0.116 Clayey loam 0.462
Level 3
High 0.259 Medium 0.161 BuiltUp 0.584 Laterite 0.282
very High 0.586 Less 0.096 WasteLand 0.062 Sandy 0.111

Level 2 Geomorphic RIW = Road RIW = RIW =

Slope RIW = 0.129 factors 0.04 quality 0.038 Elevation 0.18
Denudational
Gentle 0.586 Hills 0.222 very Low 0.468 Low 0.554
Lower
Level 3 Moderate 0.259 Plateau 0.237 Low 0.279 Medium 0.219
Steep 0.108 Coastal Plain 0.122 Medium 0.149 High 0.133
Very steep 0.048 Alluvial Plain 0.419 High 0.103 very High 0.094
 6

Journal of Geomatics Vol 10 No. 1 April 2016

Appendix 4
(The consistency ratios at levels 2 and 3)

Consistency Level 3 Overall

Ratio CR Level 2 Population Rainfall LULC Soil Slope Geomorphic Roads Elevation CR
Density factors
0.176 0.048 0.014 0.071 0.129 0.048 0.027 0.008 0.029 3.106

ISG Website
(http://www/isgindia.org)

The web site of Indian Society of Geomatics contains all pertinent information about
ISG and its activities. The latest announcements can be found on homepage itself.
“About ISG” link gives information about the constitution of ISG and its role in
Geomatics, both the technology and its applications in the Indian context. The site also
furnishes information about the members in different categories, like – Patron
Members, Sustaining Members, Life Members and Annual Members. One can
download Membership form from this section or through the Downloads link. The
website also has full information about the Executive Council Meetings’ Agenda of
past and present along with Executive Agenda and Minutes. The details of local
Chapters’ office bearers are also provided. The Annual General-body Meeting (AGM)
Agenda, their minutes for particular year can also be seen in the “AGM” section. The
list of Events organized by the society can be found through the “Events” link.

Visit ISG Web-site

http://www.isgindia.org

Website related queries, suggestions and feedback to improve the website can be sent to
the webmaster at e-mail: [email protected] or [email protected]
 7

Journal of Geomatics Vol 10 No. 1 April 2016

Performance evaluation of dimensionality reduction techniques on CHRIS
hyperspectral data for surface discrimination
Veeramallu Satya Sahithi and Iyyanki V Murali Krishna
Research Centre Imarat (RCI-DRDO), Vigyana Kancha, Hyderabad – 500069
Email: [email protected], [email protected]

(Received: Jun 19, 2015; in final form: Dec 31, 2015)

Abstract: Dimensionality reduction (DR) techniques help in reducing the volume of the hyperspectral data with
minimum loss of information. The three most commonly used DR techniques – PCA, MNF and ICA have their own
advantages and limitations in transforming the redundant hyperspectral data to non-redundant data thus aiding in an
improved feature extraction. In the present work, an attempt was made to analyze the performance of hybrid
dimensionality reduction method which uses a combination of three non – linear DR techniques for extracting the
concrete materials from the CHRIS hyperspectral data. SAM and SID classifiers were used for classifying six different
surface materials (concrete, paved and unpaved) in the study area along with four different vegetation types. Analysis
has shown that the hybrid method gave satisfactory results for classifying the surface materials in CHRIS data. The SAM
classifier gave the best results with an accuracy improvement of 10% after adapting the hybrid method. The classification
accuracies have increased from 79.54% to 85.14% for SAM classification and 80.24% to 84.90% for SID classification.

Keywords: Principle component analysis(PCA), Minimum noise function (MNF), Independent component analysis
(ICA), Hybrid method, Classification

1. Introduction unsupervised DR techniques is used to extract the features

of interest. Initially an MNF transform was applied on
Hyperspectral data contain high spectral information due AVIRIS hyperspectral data and the bands containing the
to the continuous spectral bands and narrow band width. highest signal to noise ratio (SNR) were considered for the
Such high voluminous data may lead to redundant next step. In the next step, the first 10 MNF bands
information. To overcome the problem of redundancy and containing high SNR were used as inputs for PCA and ICA
for flexible feature extraction from the voluminous and the outputs of these transformations were stacked
hyperspectral datasets, certain dimensionality reduction together to form a new vector for each pixel in the image.
(DR) techniques are used. The dimensionality reduction Each of these vectors was then classified using a Support
techniques transform the data into a new domain where the Vector Machine. The method proposed in Galal and Hasan
data in each band is made uncorrelated to the other band (2012) improved the overall performance of the SVM
based on certain criteria. There are many kinds of DR classifier and gave good results statistically. In the present
techniques that are broadly classified as linear and non – work, the performance of this method was tested on a
linear DR techniques. The three well known feature hyperspectral CHRIS image for extracting various
extraction/dimensionality reduction techniques include concrete structures. The objective of the present study is to
the principal component analysis (PCA), minimum noise extract different kinds of concrete materials from
fraction (MNF) and independent component analysis dimensionally reduced the CHRIS hyperspectral image.
(ICA) techniques. Each of these techniques works on a
unique principle and has its own advantages and 1.1. Principal Component Analysis (PCA)
disadvantages. Besides, each technique extracts unique PCA, also called the Karhunen-Love transform (KLT) or
features that are totally different to that others extract. The the Hotelling transform, is a classical statistical technique
PCA techniques works on the data variance, MNF sorts used to reduce the dimensionality of the multi-dimensional
the information based on the SNR data. PCA finds a new set of orthogonal axes that have
and the ICA assumes each band to be a linear mixture of their origin at the data mean and that are rotated to
some independent hidden components and thus applies a maximize the data variance. The covariance matrix which
linear unmixing procedure to extract the independent is used as a transformation matrix is defined as:
features. PCA and MNF measure the second order
statistics of the data with a gaussian assumption, while ∑ 𝑐𝑜𝑣 = ∑𝑛𝑖=1(𝑥̅𝑖 − 𝑚 ̅)𝑇
̅) (𝑥̅𝑖 − 𝑚 (1)
ICA uses higher order statistics and gives statistically
independent components with non – gaussian assumption where 𝑥̅𝑖 is the i th spectral value, 𝑚̅ is the mean spectral
(Wang et.al., 2014). The second order statistics used in values, n is the number of pixels in the image. The eigen
PCA and MNF cannot uncover the subtle material decomposition of the covariance matrix is performed in
substances that are uncovered by hyperspectral images order to calculate the new orthogonal axes, which can be
(Wang et.al., 2014). Hence, to make the most out of each given as:
of these techniques and to overcome their disadvantages, ∑ 𝑎̂𝑘 = 𝜆𝑘 𝑎̂𝑘 ; k= 1,2,……N (2)
Galal and Hasan, 2012, in his paper “Learning Flexible where 𝜆𝑘 is the kth eigen value, 𝑎̂𝑘 is the corresponding
Hyperspectral Features” proposed an improved method of eigenvector and N being the number of bands. The eigen
feature extraction, where a combination all the three vector forms the axes of PCA space while the eigen values

© Indian Society of Geomatics

 8

Journal of Geomatics Vol 10 No. 1 April 2016

are the measure of variance of the corresponding where U is an unknown matrix called the unmixing matrix.
eigenvector. The information content of the band increases This is a blind source separation algorithm since we do not
with the increasing value of variance (Panwar et al., 2014). have any prior information about unmixing matrix or even
The PCA bands are arranged based on the variance value on the source themselves.
– first band contains highest information with a largest
eigen value, the second PC band contains the second 2. Datasets and study area
largest amount of information and is orthogonal to the first
PC, the third PC has the third largest variance value and is The dataset used for the present study is the hyperspectral
orthogonal to both first and second PC’s and so on. Compact High Resolution Imaging Spectrometer (CHRIS)
on board the Proba mission of the European Space
1.2. Minimum Noise Fraction (MNF) Agency. CHRIS is an experimental satellite which
MNF is one of the most commonly adopted unsupervised monitors the earth in five modes and in five different look
DR techniques for the hyperspectral data. The MNF angles - +55º, +36º, 0º,-36º and -55º (Sahithi and Agrawal,
transform is specifically designed as a linear 2014). The image considered is a nadir (0º) looking image
transformation that maximizes the signal-to-noise ratio, obtained in mode 5 (land use/land cover applications). It
thus ordering images in terms of decreasing image quality has a spatial resolution of 17m with 18 spectral bands in
in lower order components. The foundations of the MNF the range of 400 – 1500nm. The present study area is one
transform were developed by Green et al. (1988) and Lee of the test sites of CHRIS experimental sensor. The study
et al. (1990). The former explained it based on signal to area is a part of Suratgarh airbase station located in Sri
noise ratio and demonstrated noise filtering via complex Ganganagar district, Rajasthan. It falls within the latitude
matrix inversion. Lee et al. (1990) simplified this as a two- and longitude of 29⁰22’24”N to 29⁰24’49”N and
cascaded PCA transform. The first phase of the MNF 73⁰52’44”E to 73⁰55’54”E. It has a total area of the
transform starts with image noise determination and noise airbase and is mostly composed of different concrete
covariance matrix calculation of the image which is materials and two to three kinds of vegetation. A location
subsequently followed by eigen value decomposition. The map of the study area is given in Fig 1.
third step includes image mean correction, noise
decorrelation and finally normalization of the linear noise
in the data which is called as noise whitening (Mundt et
al., 2007). This noise whitened data is decorrelated using
a PCA transform which is the second step of MNF
transform. The higher order images will have high SNR
which gradually reduces towards the lower order images
(which are noise dominated). Transformed MNF data are
highly decorrelated and have zero mean and a unit noise
variance. The covariance matrix of an MNF transformed
dataset is a diagonal matrix with elements equal to the
MNF eigen values.

1.3. Independent component Analysis (ICA)

ICA on multispectral or hyperspectral datasets transform a
set of mixed random signals into components that are
mutually independent. The major advantage of this ICA
transform over PCA and MNF methods is that it is based Figure 1: Study area used - LISS IV image of
on the non-Gaussian assumption of the independent Suratgarh airbase, Rajasthan, India
sources which is a typical characteristic of hyperspectral
datasets. It uses higher-order statistics to discover some 3. Methodology
interesting features in non-Gaussian hyperspectral datasets
(Yusuf and He, 2011). IC transformation can distinguish The radiometrically and atmospherically corrected CHRIS
features of interest even when they occupy only a small data is first reduced using the MNF transformation. The
portion of the pixels in an image while in PCA these small MNF components having high information are considered
features are buried in the noisy bands (ENVI 2010). Hence for further analysis. A principal component transformation
ICA analysis is very much helpful in spectral unmixing, and independent component analysis is performed over the
anomaly and target detections. selected MNF components which gave a set of
ICA is a blind source separation technique. It assumes that components for each of these transformations. All these
each band is a linear mixture of independent hidden components are then stacked to form a new vector for each
component and extracts the independent feature using a pixel in the image.
linear unmixing operation (Panwar et al., 2014).
Suppose we have N statistically independent signals, si(t), A simple SAM (Spectral Angle Mapper) classifier and
i=1,…….,N. let X(t) denote the original source signal. Spectral Informed Divergence (SID) classifier are used
ICA estimates X(t) by, over this stacked image to classify various surface
materials in the study area. Total of 10 classes were
s(t)= U X(t) (3) considered - new concrete, old concrete, tar, white paint,
sand, cemented area, croplands, shrubs, thorny trees,
 9

Journal of Geomatics Vol 10 No. 1 April 2016

bushy trees. The average spectra of various materials transformations of the hyperspectral image are shown in
considered for classification are shown in Fig 3. A table 1. The components having the highest eigen values
constant SAM angle of 0.15 was used for classifying the are considered as the informative bands and hence the first
final hybrid DR reduced image and the original CHRIS 8 bands are considered in this work. The huge difference
image. in the eigen values of first and second PCA component
shows the diversity in information content and non-
Observations were made to trace some inherent features correlation between the adjacent bands. Visual inspections
which are not observed in CHRIS original hyperspectral have shown that independent component analysis had
image. The extracted features after classification are more noise. An observation was made on the eigen values
validated using the temporal google earth images. of PCA, MNF and ICA transformations which showed that
PCA and ICA had similar eigen values. This was due to
Radiometrically corrected CHRIS data the fact that ICA is a two step process which starts with an
initial PCA transform of the image. PCA and ICA
transforms performed on the MNF components produced
better results than the direct PCA and ICA results.
MNF
Table 1: The eigen values of PCA, MNF and ICA
transformations
PCA ICA
Band PCA ICA after MNF
number after MNF Eigen
Stacking of PCA and ICA components
MNF values
1 79.764 79.764 79.764
2 10.259 10.259 10.259
3 7.670 7.670 7.670
End member extraction and
4 5.995 5.995 5.995
classification 5 3.626 3.626 3.626
6 3.284 3.284 3.284
7 2.519 2.519 2.519
Analysis and accuracy 8 2.309 2.309 2.309
assessments
However, the eigen values of the PCA post MNF and ICA
Figure 2: Methodology post MNF were same (Table 2) over the original CHRIS
image.

Table 2: The eigen values of PCA after MNF, MNF and

ICA after MNF transformations
MNF
Band PCA Eigen ICA Eigen
Eigen
number Values Values
values
1 23439699.36 79.765 23439699.36
2 174555.788 10.260 174555.788
3 8415.922 7.671 8415.923
4 5898.305 5.995 5898.305
5 625.206 3.626 625.206
6 551.119 3.284 551.119
7 340.441 2.519 340.441
8 208.063 2.3094 208.063
Figure 3: Average spectra of various materials
collected for classification The stacked image obtained by laying the PCA post MNF
components over the ICA post MNF components was used
4. Results and analysis for classification of various surface materials. Supervised
SAM and SID classifications were used for surface
On performing the MNF transform over the CHRIS image, material classification over the stacked image and the
the first 8 bands are considered for further processing due original atmospherically corrected CHRIS nadir image. A
to their high signal to noise ratio. The components were comparative study of classified stacked image and the
chosen based on their eigen value and visual information classified original image for six different concrete (paved
content. Apart from the MNF transform, a simple PCA and and unpaved) surfaces have shown that the considered
ICA are also applied over the hyperspectral image to test method yielded better classification results than the
the performance of various dimensionality reduction original image. The DR techniques not only helped in
techniques. The eigen values of PCA, MNF and ICA reducing the size of the input image but also removed the
 10

Journal of Geomatics Vol 10 No. 1 April 2016

unwanted and redundant noise from the image. This methodology. Some of the positive visual observations
technique highlighted certain hidden man made concrete noticed using the classification results are shown in the
features like roads, houses etc. following figures. The classification of the stacked image
The removal of unwanted noise from the bands improved extracted the roads in the airbase more distinctly than the
the results of SAM classification. Thin structures like road classification of the original CHRIS image. Improvements
path, smaller concrete structures etc which were poorly in the extraction of road structures and curved paths are
classified in the original hyperspectral image were well shown in Fig 6.
classified after using the stacked vector. The overall
accuracies improved by 5 to 6% after using this

New Concrete Old Concrete

Tar Scrubs
White Paint Thorny trees
Cemented area Bushy trees
Sandy surface Croplands
Figure 4: Classification of Stacked image (PCA after MNF and ICA after MNF) and the CHRIS atmospherically
corrected image using SAM and SID classifiers

Table 3: Classification accuracies of various surface SAM and SID classifiers had equal performance in
materials considered in classification classification. A SAM angle of 0.15 was used for
classification of both the original and transformed images.
Spectral Angle Spectral The overall average accuracies (Fig. 5) have increased
Mapper Informed from 79.54% to 85.14% for SAM classification and
Divergence 80.24% to 84.90% for SID classification.
Class Original Stacked Original Stacked
name CHRIS Vector CHRIS Vector 86.00% 85.14% 84.90%
New 78.56% 85.25% 79.56% 87.67% 84.00%
Concrete 82.00% 80.24%
79.54%
Tar 85.45% 87.63% 84.68% 88.65% 80.00%
White 88.56% 91.58% 89.25% 91.44% 78.00%
Paint 76.00%
Cemented 72.66% 85.86% 73.58% 80.25%
area
Sandy 74.56% 84.78% 81.21% 82.56%
surface
Old 80.45% 87.45% 81.56% 87.24%
Concrete Figure 5: Overall classification accuracies
Scrubs 81.25% 83.15% 79.54% 81.45%
Thomy 75.45% 77.63% 74.68% 78.65% The spatial resolution of the hyperspectral image plays a
trees major role in the image classification, apart from its high
Bushy 79.21% 83.47% 80.12% 84.65% spectral resolution. In the present work, the spatial
trees resolution of 17m of the CHRIS hyperspectral image is
Croplands 79.25% 84.56% 78.21% 86.46% still coarse to misclassify the pixel into a different class
Average 79.54% 85.14% 80.24% 84.90% (mixed pixel effect).
accuracy
Accuracy assessment was performed over the four Also it is accepted that the transformed image could not
classified images - SAM classified original and completely restore the spectral resolution of the original
transformed stacked images, SID classified original and CHRIS image after using the DR techniques. Hence, a
transformed stacked images. The accuracies for each class compromise in the overall accuracy rate of the classifiers
in all the four classified maps are shown in table 3. Both is accepted.
 11

Journal of Geomatics Vol 10 No. 1 April 2016

Figure 6: A - Improvements in the extraction of road paths up on classification; B - Improvements in the extraction
of urban and concrete buildings in a township; C- Extraction of curved path; D - Extraction of building roof coated
with artificial paints; E - Google Earth images of figures – A, B, C, D respectively

5. Conclusions terms of image quality with implications for noise

removal. Transactions on Geoscience and Remote
In this paper, an existing methodology was tested in Sensing, 26(1), 65-74.
extracting the features from dimensionally reduced
CHRIS hyperspectral data. The MNF technique helped in Lee, J.B., S. Woodyatt and M. Berman (1990).
extracting the data with high signal to noise ratio and in Enhancement of high spectral resolution remote sensing
removing the noisy bands from the voluminous data by a noise-adjusted principal components transform.
hyperspectral data. PCA and ICA transformation of the IEEE Transactions on Geoscience and Remote Sensing,
transformed MNF components helped in bringing out 28(3), 295-304.
some features which could not be directly observed in the
atmospherically corrected untransformed hyperspectral Mundt, J.T., D.R. Streutker and F.G. Nancy (2007). Partial
data. ICA technique disclosed the spectral differences in a unmixing of hyperspectral imagery: Theory and
patch of mixed urban and vegetation. The PCA technique methods. Proceedings of the American Society of
de-correlated the MNF transformed data and some Photogrammetry and Remote Sensing.
features like roads and other structures hidden within the
various bands of the CHRIS image were revealed after this Panwar, A., A. Singh and H.S Bhaduria (2014).
transform. On the whole a combined method of MNF, International Journal of Emerging Technology and
PCA and ICA techniques seemed to be of great use in Advanced Engineering. ISSN 2250-2459, ISO 9001:2008
detecting the smallest information present in the Certified Journal, Volume 4, Issue 5, 701 - 705.
hyperspectral data. The performance of SAM classifier
ameliorated on using the transformed non-redundant and Sahithi, V.S. and S. Agrawal (2014). Sub pixel location
decorrelated data showing a significant improvement in identification using super resolved multilooking CHRIS
the classification accuracies. data. ISPRS-International Archives of the
Photogrammetry, Remote Sensing and Spatial
Acknowledgement Information Sciences, 1, 463-468.

The authors are thankful to Dr Sateesh Reddy, the Tutorial, E.Z. (2010). ENVI user guide. Colorado Springs,
outstanding scientist and Director, Dr BV Rao, CO: ITT.
Technology Director, Dr Arindan Biswas, Scientist from
IIRS and Rama Sarma, DOMS, RCI, DRDO for their Wang, Y., G. Wu and L. Ding (2014). Plant Species
interest and for providing the necessary infrastructure and Identification Based on Independent Component Analysis
extending all support whenever required. for Hyperspectral Data. Journal of Software 9.6: 1532-
1537.
References
Yusuf, B.L. and Y. He (2011). Application of
Galal, A. and H. Hasan (2012). Learning Flexible hyperspectral imaging sensor to differentiate between the
Hyperspectral Features. International Journal of Remote moisture and reflectance of healthy and infected tobacco
Sensing Applications. IJRSA Vol.2 Iss. 2, 44-48 leaves. African Journal of Agricultural Research, 6(29),
6267-6280.
Green, A.A., M. Berman, P. Switzer and M.D. Craig
(1988). A transformation for ordering multispectral data in
 12

Journal of Geomatics Vol 10 No. 1 April 2016

Accuracy evaluation for online Precise Point Positioning services
Ashraf Farah
College of Engineering, Aswan University, Aswan, Egypt
Email: [email protected]

(Received: Aug 11, 2015; in final form: Jan 01, 2016)

Abstract: Precise Point Positioning (PPP) has been proved by many researchers in the last decade as a cost-effective
alternative for Differential GPS (DGPS) with an estimated precision sufficient for many applications. PPP
implementation needs state-of-art software to correct GNSS observations from different types of errors. Several online
PPP services have been developed recently by government agencies, universities, industries and individuals. The PPP
software centre managed by University of New Brunswick (UNB), Canada is offering the user four online PPP services
(CSRS-PPP, GAPS, APPS and magicGNSS) (UNB-PPP, 2015). This research presents an accuracy assessment
evaluation study for those services by processing 3h 52min. dual frequency-static GPS observations which were divided
into 10 sessions with different observation duration (10min, 20min, 30min, 45min., 1 hr, 1.5 hr, 2 hr, 2.5 hr, 3 hr and 3hr
52min.)

Keywords: GPS, Static, Precise point positioning (PPP), Observation duration

1. Introduction 2. Precise Point Positioning (PPP) Approach

Precise point positioning (PPP) is an enhanced single The basic observable for PPP is the ionosphere-delay-
point positioning technique for code or phase free (up to first order effects) pseudo-range and carrier-
measurements using precise orbits and clocks instead of phase observations. The simplified observation
broadcast data. PPP became viable with the existence of equations are given as:
the extremely precise ephemerides and clock
corrections, offered by different organizations.To
compensate for ionospheric effects (the largest source
of error for GPS observations), dual frequency and
measurements are used for an ionosphere free
combination (Rizos et al., 2012).

PPP can provide positioning accuracy of centimeters or

millimeters using un-differenced carrier phase where Pifis the ionosphere-delay-free pseudorange
observations where ambiguities are usually estimated as observation, Φif is the ionosphere-delay-free carrier-
float values. PPP precision varies based on observation phase observation, in units of meters, ρ is the geometric
type (single or dual frequency) and the duration of distance between the satellite and the receiver antenna
observations among other factors (Farah, 2013). phase centers, c is the vacuum speed of light, dTand
dtare the receiver and satellite clock errors, T is the
The PPP software centre managed by University of New tropospheric slant delay, λif is the ionosphere-delay-free
Brunswick (UNB), Canada is offering the user four wavelength and Nifis the carrierphase ambiguity. In the
online PPP services (CSRS-PPP, GAPS, APPS and standard PPP model, Nifis not an integer as it is
magicGNSS) (UNB-PPP, 2015) which allows for: contaminated by instrumental delays. It is possible to
solve this problem by estimating separate pseudorange
 An easy comparison of PPP solutions and carrier-phase clock offsets. However, for this
provided by different online PPP services approach, both the pseudorange and carrier phase
require a satellite-specific clock bias parameter which
 An increased reliability for the user by giving must be provided externally, in a similar manner to the
access to independent PPP solutions clock and orbit products which are already used in PPP,
to prevent the normal matrix from being singular
 An insight at the performance of different (Collins, 2008).
implementation strategies
The tropospheric slant delay is normally separated into
the zenith delayand a mapping function which is
 A means of validation for potential PPP
required to make the zenith delay parameter common
software developers
over all satellites. Table 1 presents the PPP biases and
errors that have to be considered (Rizos et al., 2012).

© Indian Society of Geomatics

 13

Journal of Geomatics Vol 10 No. 1 April 2016

Table 1: The PPP biases and errors that have to be 3.2 GAPS-PPP Service
considered (Rizos et al., 2012) The GPS Analysis and Positioning Software (GAPS) is
a GPS PPP application developed at the UNB (Leandro
Satellite specific errors et al., 2007). GAPS exists in two forms: a web-based
Precise satellite clock corrections positioning service, to which users can upload GPS
Satellite antenna phase centre offset observations to be processed, and a command-line
Satellite antenna phase centre variations executable version, which can be used to process large
Precise satellite orbits amounts of GPS data in a fast and convenient
Satellite antenna phase wind-up error manner.The algorithms and code structure used in
Receiver specific errors GAPS follows standard GPS PPP approaches but with
Receiver antenna phase centre offset some important and unique differences.
Receiver antenna phase centre variations
Receiver antenna phase wind-up error The ionospheric delay estimation uses a spherical
Atmospheric modelling ionospheric shell model, in which the vertical delays are
Troposphere delay described by means of a zenith delay at the station
Ionosphere delay (L1 only) position and two horizontal gradients. This estimation
Geophysical models makes use of carrier-phase measurements only.
Solid earth tide displacements
Ocean loading The code multipath estimation is based on the
Polar tides assumption that the several effects present in code
Plate tectonic motion measurements are dealt with within PPP, but strongly
based on carrier-phase measurements. Based on this,
3. UNB-PPP Software Centre these effects can be removed from pseudorange
measurements, and the leftover effect is essentially the
PPP implementation needs state-of-art software to code multipath plus receiver noise. Another effect
correct GNSS observations for different types of errors. which afflicts pseudorange measurements is the code
Several online PPP services have been developed bias. The code biases are important because satellite
recently by government agencies, universities, clock data products are computed using a certain
industries and individuals. The University of New arbitrary convention of observation type, such as P1
Brunswick (UNB), Canada has ongoing effort to code measurements (from semicodeless P(Y) tracking)
promote PPP technique. The PPP software centre rather than the C1 code (from C/A-code tracking). If the
managed by UNB, Canada is providing the user four user´s receiver uses a different observation type than the
online PPP services CSRS-PPP, GAPS, APPS and one which was used to generate the satellite clock error
magicGNSS (UNB-PPP, 2015). Each service is corrections, one has to apply an offset to the correction,
implemented and managed by an independent equivalent to the bias between the observations, to be
organization, however submitting Rinex observation able to use these clock products. One of GAPS’ analysis
file through UNB-PPP centre will recover solutions tools produces values of the satellite code biases, based
from the four PPP services. A simple presentation for on a positioning observation model, as opposed to being
the four PPP services is followed. Table 2 presents based on a satellite clock estimation observation model
processing parameters for the used four PPP online as is usually the case when bias values are provided to
services. users. Regarding satellite clock error estimates, GAPS
was enhanced in order to provide estimates of satellite
3.1 CSRS-PPP Service clock offsets. This tool was created aiming at a suitable
The Canadian Spatial Reference System (CSRS) PPP approach for real-time carrier-phase based satellite
service provides post-processed position estimates over clock estimation (Leandro et al., 2007).
the internet from GPS observation files submitted by the
user. Precise position estimates are referred to the CSRS 3.3 APPS-PPP Service
standard North American Datum of 1983 (NAD83) as The Automatic Precise Positioning (APPS) service of
well as the International Terrestrial Reference Frame the Global DGPS System is an online service from the
(ITRF). Single station position estimates are computed JPL (California Institute of Technology, CA, USA). The
for users operating in static or kinematic modes using APPS online service is currently based on JPL’s
precise GPS orbits and clocks (IGS, 2015; Kouba, GIPSY-OASIS software, v6.3. It provides a PPP
2001). The online PPP positioning service is designed solution for the uploaded RINEX files in static or
to minimize user interaction while providing the best kinematic techniques. Currently, it deals only with dual
possible solution for a given observation availability. frequency data (Table 2). The result is shown on the
Currently, users need only to specify the mode of webpage interface in static technique and in case of
processing (static or kinematic) and the reference frame kinematic technique, the result is sent to the user’s email
for position output (NAD83 (CSRS) or ITRF). CSRS- address. Moreover, the elevation angle and the solution
PPP service is processing both single and dual output interval can be controlled by the user (JPL,
frequency observations from GPS and GLONASS 2015).
(Farah, 2014; CSRS-PPP, 2015).
 14

Journal of Geomatics Vol 10 No. 1 April 2016

3.4 magicGNSS-PPP Service orbits, satellite and station clock offsets, phase
The magicGNSS, v2.5, operated by GMV Aerospace ambiguities, tropospheric zenith delays, and station
and Defence, Spain, is based on software developed for coordinates. The magicGNSS software is processing
GALILEO orbit determination and time only dual-frequency observations and does not accept
synchronization. A batch least-squares algorithm is used single-frequency observations. (Piriz et al., 2009).
to minimize measurement residuals and to determine

Table 2: Processing parameters for PPP online services

PPP Service CSRS-PPP GAPS-PPP APPS-PPP magicGNSS-PPP

Reference ITRF2008
system
Coordinate LLH/XYZ
format
Satellite orbit IGS IGS JPL IGS
and clock
ephemeris
Satellite phase IGS ANTEX
centre offsets
Receiver phase IGS ANTEX
centre offsets
Tropospheric
model
Dry model Davis (GPT) UNB-VMF1 Standard A batch least squares
(CMC) formula algorithm that minimizes
Wet model Hopfield model Gradient Chen Standard(0.10m) measurement residuals
(GPT) and Herring Gradient model solving for orbits, satellite
model and station clock offsets,
phase ambiguities and
station tropospheric zenith
delays.

Mapping GMF VMF1 GMF -

function
Ionospheric Second-order Linear Second-order Linear ionospheric free
model linear ionospheric ionospheric free linear combination
combination combination ionospheric
combination
Min. Elevation 10o 10o 7.5o 10o
angle
GNSS System GPS/GLONASS GPS GPS GPS/GLONASS/Galileo
Software CSRS-PPP GAPS v5.2.0 GIPSY-OASIS magicGNSS v5.3
v6.3
Observation Single/dual dual frequency dual frequency dual frequency
Data frequency
Static/kinematic Static/kinematic Static/kinematic
Static/kinematic
Ocean tide FES 2004 FES 2004 Desai Onsala Space Observatory
loading parameters

APPS: Automatic Precise Point Service; CSRS: Canadian Spatial Reference System; GAPS: GPS Analysis and
Positioning Software; PPP: precise point positioning; IGS: International GNSS Service; JPL: Jet Propulsion Laboratory;
GPT: global pressure and temperature data; GMF: global mapping function; FES: Finite Element Solution;
(CMC):Canadian Meteorological Centre ;(GPS)Global Positioning System
 15

Journal of Geomatics Vol 10 No. 1 April 2016

Table 3: The estimated coordinates from the four PPP-services for the 10 observation batches
10 min. observation batch
Coordinate Latitude longitude Height (m)
Reference 24º 43/ 25.8626// 46º 37/ 02.7899// 645.290
CSRS-PPP 24º 43/ 25.8640// 46º 37/ 02.7916// 645.361
GAPS 24o 43/ 25.8636// 46o 37/ 02.7906// 645.721
magicGNSS No data No data No data
APPS No data No data No data
20 min. observation batch
Reference 24º 43’ 25.8626’’ 46º 37/ 02.7899// 645.290
/ //
CSRS-PPP 24º 43 25.8637 46º 37/ 02.7868// 645.035
GAPS 24o 43/ 25.8642// 46o 37/ 02.7866// 645.321
magicGNSS 24o 43/ 25.8748// 46o 37/ 2.8024// 645.944
APPS 24o 43/ 25.8655// 46o 37/ 2.7807// 645.169
30 min. observation batch
Reference 24º 43/ 25.8626// 46º 37/ 02.7899// 645.290
/ //
CSRS-PPP 24º 43 25.8627 46º 37/ 02.7876// 645.146
GAPS 24o 43/ 25.8641// 46o 37/ 02.7869// 645.291
magicGNSS 24o 43/ 25.8685// 46o 37/ 2.7941// 646.882
APPS 24o 43/ 25.8651// 46o 37/ 2.7822// 645.324
45 min. observation batch
Reference 24º 43/ 25.8626// 46º 37/ 02.7899// 645.290
/ //
CSRS-PPP 24º 43 25.8625 46º 37/ 02.7895// 645.292
GAPS 24o 43/ 25.8632// 46o 37/ 02.78946// 645.443
magicGNSS 24o 43/ 25.8662// 46o 37/ 2.7935// 647.217
APPS 24o 43/ 25.8638// 46o 37/ 2.7871// 645.331
1 hr. observation batch
Reference 24º 43/ 25.8626// 46º 37/ 02.7899// 645.290
/ //
CSRS-PPP 24º 43 25.8626 46º 37/ 02.7907// 645.318
GAPS 24o 43/ 25.8631// 46o 37/ 02.7906// 645.501
magicGNSS 24o 43/ 25.8621// 46o 37/ 02.7927// 647.557
APPS 24o 43/ 25.8628// 46o 37/ 02.7913// 645.454
1.5 hr. observation batch
Reference 24º 43/ 25.8626// 46º 37/ 02.7899// 645.290
CSRS-PPP 24º 43/ 25.8629// 46º 37/ 02.7899// 645.282
GAPS No data No data No data
magicGNSS 24o 43/ 25.8639// 46o 37/ 2.7887// 647.169
APPS 24o 43/ 25.8626// 46o 37/ 2.7896// 645.542
2 hr. observation batch
Reference 24º 43/ 25.8626// 46º 37/ 02.7899// 645.290
CSRS-PPP 24º 43/ 25.8630// 46º 37/ 02.7898// 645.281
GAPS 24o 43/ 25.8634// 46o 37/ 02.7896// 645.517
magicGNSS 24o 43/ 25.8634// 46o 37/ 02.7895// 647.320
APPS 24o 43/25.8626// 46o 37/ 02.7895// 645.541
2.5 hr. observation batch
Reference 24º 43/ 25.8626// 46º 37/ 02.7899// 645.290
/ //
CSRS-PPP 24º 43 25.8628 46º 37/ 02.7898// 645.303
GAPS none none none
magicGNSS 24o 43/ 25.8625// 46o 37/ 02.7891// 647.371
APPS 24o 43/ 25.8628// 46o 37/ 02.7895// 645.558
3 hr. observation batch
Reference 24º 43/ 25.8626// 46º 37/ 02.7899// 645.290
/ //
CSRS-PPP 24º 43 25.8627 46º 37/ 02.7899// 645.301
GAPS 24o 43/ 25.8631// 46 37/ 02.7896// 645.554
magicGNSS 24o 43/ 25.8629// 46o 37/ 02.7898// 647.372
APPS 24o 43/ 25.8627// 46o 37/ 02.7895// 645.535
3 hr. 52min observation batch
Reference 24º 43/ 25.8626// 46º 37/ 02.7899// 645.290
/ //
CSRS-PPP 24º 43 25.8626 46º 37/ 02.7899// 645.290
GAPS No data No data No data
magicGNSS 24o 43/ 25.8629// 46o 37/ 02.7896// 647.328
APPS 24o 43/ 25.8627// 46o 37/ 02.7897// 645.519

© Indian Society of Geomatics

 16

Journal of Geomatics Vol 10 No. 1 April 2016

Table 4: The (estimated coordinates – reference coordinates) from the four PPP-services for the 10
observation batches
10 min. observation batch
Estimated coordinate – Reference coordinate Latitude (m) Longitude (m) Height (m)
CSRS-PPP 0.042 0.051 0.071
GAPS 0.029 0.021 0.431
magicGNSS No data No data No data
APPS No data No data No data
20 min. observation batch
CSRS-PPP 0.033 -0.093 -0.255
GAPS 0.049 -0.099 0.031
magicGNSS 0.366 0.375 0.654
APPS 0.087 -0.275 -0.121
30 min. observation batch
CSRS-PPP 0.003 -0.069 -0.144
GAPS 0.044 -0.090 0.001
magicGNSS 0.177 0.126 1.592
APPS 0.075 -0.232 0.034
45 min. observation batch
CSRS-PPP -0.003 -0.012 0.002
GAPS 0.017 -0.013 0.153
magicGNSS 0.108 0.108 1.927
APPS 0.037 -0.085 0.041
1 hr. observation batch
CSRS-PPP 0 0.024 0.028
GAPS 0.014 0.021 0.211
magicGNSS -0.015 0.084 2.267
APPS 0.007 0.042 0.164
1.5 hr. observation batch
CSRS-PPP 0.009 0 0.008
GAPS No data No data No data
magicGNSS 0.039 -0.036 1.879
APPS 0.001 -0.008 0.252
2 hr. observation batch
CSRS-PPP 0.012 -0.003 -0.009
GAPS 0.023 -0.010 0.227
magicGNSS 0.024 -0.012 2.030
APPS 0.001 -0.011 0.251
2.5 hr. observation batch
CSRS-PPP 0.006 -0.003 0.013
GAPS No data No data No data
magicGNSS -0.003 -0.024 2.081
APPS 0.006 -0.011 0.268
3 hr. observation batch
CSRS-PPP 0.003 0 0.011
GAPS 0.014 -0.008 0.264
magicGNSS 0.009 -0.003 2.082
APPS 0.004 -0.012 0.245
3 hr. 52min observation batch
CSRS-PPP 0 0 0
GAPS No data No data No data
magicGNSS 0.009 -0.009 2.038
APPS 0.004 -0.006 0.229

© Indian Society of Geomatics

 17

Journal of Geomatics Vol 10 No. 1 April 2016

Figure 1: The (estimated coordinates – reference coordinates) in meters from the four PPP-services for the 10
observation batches
The quality of observation data was checked using
4. Test Study UNAVCO’s translation, editing, and quality check
utility (TEQC) software (TEQC, 2012). This set of
To compare the accuracy of these four online PPP observations was divided using TEQC software into 10
services, an observation set of 3h 52 min. dual batches with different observation duration (10min,
frequency-static GPS observations was collected (GPS 20min, 30min, 45min., 1 hr, 1.5 hr, 2 hr, 2.5 hr, 3 hr and
day 17191) with Topcon GR-3 dual frequency receiver 3hr 52min.). The PPP coordinates for each session were
(Topcon GR-3, 2013) using 15 sec observation interval estimated using the four PPP services and compared
and 10o cut-off elevation angle. with the reference coordinates for the tested station. The
reference coordinates is obtained from processing the (3
hr 52 min.) batch using CSRS-PPP service.
 18

Journal of Geomatics Vol 10 No. 1 April 2016

5. Results and discussion CSRS-PPP (2015). Canadian Spatial Reference System
(CSRS) Precise Point Positioning (PPP) service.
Table 3 presents the estimated coordinates from the four http://webapp.geod.nrcan.gc.ca/geod/tools-outils/.
PPP-services for the 10 observation batches (10min, (Accessed 8/8/2015).
20min, 30min, 45min., 1 hr, 1.5 hr, 2 hr, 2.5 hr, 3 hr and
3hr 52min.).Table 4 and Figure 1 present the difference Farah, A. (2013). Effect analysis of GPS observation
between the estimated coordinates and the reference type and duration on convergence behavior of static
coordinates for the tested station for the tested batches. PPP. Journal of Geomatics, vol.7, no.2, October 2013.
It can be seen from these tables that each PPP-service
provide different coordinate estimates comparing with Farah, A. (2014). Assessment study of static-PPP
the other services. GAPS service provides no estimates convergence behavior using GPS, GLONASS and
for some observation batches such as (1.5 hr, 2.5 hr and mixed GPS/GLONASS observations. Artificial
3hr 52 min.). APPS service did not process 10 min. Satellites, Vol. 49, No. 1 – 2014 DOI: 10.2478/arsa-
batch for low number of observations where observation 2014-0005
interval was 15 seconds, so the 10min. Rinex file has 40
epochs. IGS (2015). International GNSS service.
https://igscb.jpl.nasa.gov/components/prods_cb.html
CSRS-PPP service provides the most accurate (Accessed 8/8/ 2015).
coordinate estimates for the (10min. up to 1hr)
observation batches with centimeter accuracy for JPL (2015). Website for JPL products, Available at:
latitude, longitude and height coordinates. CSRS https://gipsyoasis. jpl.nasa.gov/index.php (Accessed
provides millimeter accuracy for latitude, longitude and 8/8/2015).
height coordinates from 1.5hr observation batch and up
to 3hr 52min. batch. Kouba, J. and P. He´roux (2001). Precise point
positioning using IGS orbit and clock products. GPS
APPS-service provides the second best accuracy for Solutions. 5 (2), 12–28, 2001
coordinate estimates after the CSRS service with
slightly worse values for the height coordinate. GAPS Leandro, R.F., M.C.Santos and R.B. Langley (2007).
service provides similar accuracy for APPS service GAPS: The GPS analysis and positioning software – a
when it returns solutions. magicGNSS service did not brief overview. In: Proceedings of the 20th international
provide a solution for 10min. observation batch. technical meeting of the satellite division of the institute
magicGNSS service provides the worst coordinate of navigation — ION GNSS 2007. The Institute of
estimates for the total tested 10 batches comparing with Navigation, Fort Worth, 25–28 September, pp 1807–
other services. magicGNSS provides better results with 1811.
long observation batches such as 1 hr and more for the
latitude and longitude coordinates, However, the height Pı´riz, R., D. Calle, A. Mozo, P. Navarro, D. Rodrı´guez
coordinate has the worst accuracy ever with an average and G. Tobı´as (2009). Orbits and clocks for GLONASS
error of 2 m. precise-point-positioning. In: Proc. ION GNSS 2009.
Savannah, Georgia, pp. 2415–2424, September 22– 25,
6. Conclusions 2009.

It can be concluded from this research that different PPP Rizos, C., V. Janssen, C. Roberts and T. Grinter (2012).
services provide different solutions according to Precise point positioning: is the era of differential GNSS
processing strategy. The user should compare his positioning drawing to an end?. FIG Working Week
solutions from number of services and not depend on 2012. Rome, Italy, 6–10 May 2012.
one service. The difference between PPP estimates from
different services could reach 2 meters for height TEQC (2015). TEQC-UNAVCO tutorial.
coordinate and at a decimeter level for latitude and http://facility.unavco.org/software/teqc/doc/UNAVCO
longitude coordinates. The study recommends the _Teqc_ Tutorial.pdf. (Accessed 8/8/2015).
CSRS and APPS services to be used by PPP-users.
Topcon GR-3 (2015). Topcon Positioning Systems
References (GR-3). http://www.topcon.co.jp/en/positioning/
products/pdf/gr-3_e.pdf.(Accessed 8/8/2015).
Collins, P. (2008). Isolating and estimating
undifferenced GPS integer ambiguities. In: Proceedings UNB-PPP (2015). The precise point positioning service.
of ION NTM 2008, San Diego, 28–30 January 2008, pp University of New Brunswick (UNB)
720–732 http://www2.unb.ca/gge/Resources/PPP/. (Accessed
8/8/2015.
 19

Journal of Geomatics Vol 10 No. 1 April 2016

REIS: A spatial decision system for land valuation
Himanshu Deshmukh1, Rupali Khare2, Kamal Pandey1 and Dheeraj Kumar2
1
Indian Institute of Remote Sensing, Dehradun
2
Indian School of Mines, Dhanbad
Email: [email protected], [email protected], [email protected], [email protected]

(Received: Mar 01, 2016; in final form: April 13, 2016)

Abstract: Land is a vital asset for any federal government. Valuation of land is a very important activity as it has high
impact on revenue generation for any country. Present valuation technique is time consuming and cannot quantify the
spatial importance of the land for decision making related to real estate. GIS provides a technical platform for
management of geographic data and inherent location information to support application of spatial statistical and location
econometric tools. Spatial database helps to take decisions for their projects and act as a base data for further functions
like tax calculation, land purchase etc. The main objective of this study is to develop a Real Estate Information System
(REIS) for the valuation purpose especially for the buying, selling and taxation of land properties. This computerized
standalone application, as a repository of land value data is capable to providing easy access to the user/customer about
the land information. The technology used for development is open source and hence offers easy modification and
customization at user end also. The current version is developed for windows environment and can be used and installed
in any windows based system with very less effort. The system can be utilized by the users interested in real estate market
for the land procurement process. The system can also assist government agency for taxation purpose.

Keywords: Geographical information system (GIS), Real Estate Information System (REIS), Open source, Land
valuation

1. Introduction take full advantage of location information contained in

these databases to support the application of spatial
Development activity in India is increasing at a high rate statistical and spatial econometric tools
and is only expected to increase further in the future.
Rising urbanization offers opportunities for the The main objective of the study is to develop a GIS
development of real estate market. The growth in real based system which facilitates decision making related
estate industry is visible in every city. Land plays a to land valuation information. The sub objectives of the
crucial role in the life of people. Difference in position, study are as follows:
fertility or natural resources make some locations and  To create a spatial database of unit rate for real
land parcels more desirable and valuable than the others. estate properties.
System of valuation provides control of real estate  To generate parameter based valuation of
market. Property valuation is a process of identifying properties using GIS functionalities.
and assigning those factors that affect the value  To develop interactive application which enables
(Horsley, 1992). For the purpose of valuation, the prospective user in decision making related to land
valuation methods are classified as comparative or real estate properties.
methods, cost methods and interactive analysis method
(Dale, et. al 1999; Rangawala et al., 2011). But for this 2. Study area and data used
study, valuation purpose is mainly for buying and
selling the real properties. The valuation is done on the 2.1 Study area
basis of parameters and rules set by the local governing The study area is a part of Bhopal city, the capital of
body i.e. in the case of present study area District Madhya Pradesh, which is situated in Central India.The
Collector, Bhopal & District Valuation Association. total population of Madhya Pradesh is about 7.26 crores.
These parameters are used for the valuation of land According to census of India, population of Bhopal has
properties for payment of registration charge of increased from 10,62,771 in 1991 to 14, 37,351 in year
ownership. These parameters are per unit area called 2001 at 35.24% increase and up to 17,98,218 in 2011 at
circle rate. Thisis defined for the different locations. a 25.10% increment. Population wise, it is the second
These circle rates are taken as bench mark rates for this largest city of Madhya Pradesh. Study area is north-east
study. These rates are revised yearly by the District part of city having coordinates 77°26´12´´E,
Collector in India. 23°16´10´´N at an average altitude of 427m from mean
sea level. The study area includes Raisen road, Vidisha
GIS provides a technical platform on which market road and Ring road. The study is carried out along these
analysis as well as spatial representation of property roads for various land parcels for valuation purpose.
information can be shown in the form of maps (Waytt,
1997). Maps improve the decision making capabilities 2.2 Data used
of human being. Capabilities of GIS facilitate the The study is based on data that are collected from
management of geographic data, as well as it enables to various sources including internet. The primary

© Indian Society of Geomatics

 20

Journal of Geomatics Vol 10 No. 1 April 2016

requirement is the detail about land records in spatial
format, which is to be valued. A land use map of the Table 1: Data sets used for the study
study area is also required for classifying the
properties as residential, commercial, etc. The S. Data Type Source
development plan of the study area viz. Bhopal city is No.
also a crucial requirement which contains the data 1 Khasra Vector Wardmap
provided by the Town and Country Planning Boundry
Department, Bhopal. For valuation purpose unit rates 2 Commercial Vector Landuse map
and rules of valuation of properties are required. For area
each year circle rates and guidelines for valuation of 3 Industrial Vector Landuse map
immovable properties under Bhopal district are area
announced by District collector, Bhopal, taken from 4 PSP area Vector Town and Country
website of Collector office Bhopal (www.mp.nic.in). Planning Bhopal
The circle rates are taken as benchmark in the place of 5 Existing Vector Town and Country
market rates and the guidelines are helpful in making PUF area Planning Bhopal
parameters for the valuation purpose. But the rates 6 Recreational Vector Town and Country
vary from place to place in different wards, so a ward area Planning Bhopal
map is also needed because neither development plan 7 Residential Vector Landuse map
nor land records are based on wards. The complete area
dataset used in the study is listed in Table 1. 8 Existing Vector Landuse map
Transport
3. Tools and technologies 9 Wardmap Vector Wardmap
10 Circle rate of Text file Wardmap
The tools and technologies used in the development of
Bhopal City
the software solution are listed in Table 2.
11 Satellite Raster http://earthexplorer.
image usgs.gov/
4. Methodology
(ORBVIEW-3)
The overall methodology adopted for the study is given
in Figure 1.

Figure-1: Methodology of the study

 21

Journal of Geomatics Vol 10 No. 1 April 2016

Table-2: Tools and Technologies used in the study help of the scanned and georeferenced wardmap. After
S. No. Component Specification digitization, an attribute table was prepared for the
Operating Windows (32 Bit wardmap layer with the attributes ward_no,
1 system &64 Bit) shape_length, shape_area and ward_name.
2 GIS package Map window
Programming
3 language VB.net
4 Database File based(.shp)
.
Map window active
5 Libraries control(.ocx)
(a) (b)
The data available for ward boundaries and road
network was available in image form. These data are
converted in .shp files by digitization. After base map,
attribute coding was done which includes ward name,
ward no., area for wards and road width, road length,
road name for roads. This work was done in open source
GIS and final base map was prepared. The available data
(c) (d)
was cadastral map which contains Khasra no. area, P. H.
No. and village name. This map was further improvised
Figure 2: (a) Digitized ward map; (b) attribute table
by adding unit circle land rates applicable for various
of ward map; (c) digitized roads map; and (d)
areas as per the guidelines for circle rates issued by
attribute table of road map
administration of that area. Multiplicative factors are
worked out for each of the parameter which affect the
5.2 Preparation of ward wise land value map
land unit rate and hence added to the base map file. The
Ward map so prepared is used to assign the individual
parameters and multiplicative factors as listed below:
cadastral units to different ward the land values as per
the Govt. guidelines and other attributes are coded in the
 Landuse
attribute table with the help of Field Calculator tool.
Land Comme Indus Agricul Reside Field names in the value map attribute table are
use rcial trial tural ntial Khasra_no, type, ward_name, area, land_value and
MF 1.50 1.50 1.00 1.00 PH_HALKA.
 Depth from road
Depth from Less than 20 More than 20 5.3 Finalizing parameters for valuation and
road Meter meter valuation of sample properties
MF 1.00 0.75 The parameters affecting the value of any property are
 Type of road taken from the “Valuation Guidelines of Immovable
Road National State Other Properties” issued by the District Collector, Bhopal.
type Highway Highway Roads These guidelines describe various parameters such as
MF 2.00 1.50 1.20 depth from road, roadtype, land use etc, which are inputs
for working out valuation. Every parameter has been
5. Results and discussion given a weight and these weights are filled across the
polygons in the attribute table of cadastral maps.
The development of Real Estate Information System i.e.
REIS using geospatial techniques has been carried out
in the following stages-
 Creation of geospatial database 
 Valuation of properties 
 Development of software solution.

5.1 Creation of geospatial database

A variety of database has been utilized in the study. First
of all the wardmap was obtained from municipal
corporation Bhopal. The map was scanned and geo-
referenced with the help of ground control points. The
geo-referenced wardmap was converted to vector layer
by manual digitizing. The vector ward map layer was
attribute coded with ward name and population
collected from census of India 2011. Figure 3: Ward wise value map

Figure 2 shows the ward map and road map with the
 22

Journal of Geomatics Vol 10 No. 1 April 2016

Figure 4: A sample land value map Figure 6: Graphical User Interface (GUI) of REIS

5.5 Sample use case of REIS application

CASE 1: land type=commercial, Locality = Indrapuri,
Budget (Rs.) = Above 50 Crores, Area (m2) = Above
50,000 m2, Result: One area selected.

5.6. Discussion
The application (REIS) is developed for realizing an
analytical system to study the real estate market by the
user. It provides results based on the input and also can
be used for finding costliest or cheapest properties in
any particular area or in the city. Any kind of analysis
can be done like comparison of values in same area or
comparison of same extent of land in different wards.
Figure 5: Valuation of properties This study has provided the GIS based system to work
out the valuation of open land parcels but further
5.4 GUI designing and coding of basic structure of parameters and multiplication factors can be added for
REIS built up parcels owned by individuals so that entire
To fulfil the third objective, a software application has valuation will become more scientific and customised.
been developed using open source technology that Creation of spatial database is very useful for valuation
facilitates prospective users to search a property purpose and it can be recommended to prepare spatial
according to their needs. In the present study a simple databases for the government departments related to the
graphical user interface (GUI) is created wherein land. It can help them to take decisions for their projects
dynamic map are displayed and it allows to perform and can act as a base data for the further functions like
basic GIS operation and user-based queries (Figure 6). Tax calculation, etc. by adding appropriate attributes.
The GUI have been developed using visual basic System integration or a development of centralized
programming language. system for all the departments related to the land records
is important so that benefit could be obtained from
The query window is the main part of the application business as well as technical point of view. REIS can
which has been customized to meet the needs of the also be linked to other databases of different
user. The query window is linked with the database to departments so that user can access the other types of
fetch the requested results and to display it on the map data. This would save the resources, time and effort and
window. The user interface contains basic display can help in the fast decision making. This system can be
components like zoom-in, zoom-out, pan etc. It also improved by adding the market rate, which can help user
provides the facility of spatial query and it provides the a better understanding of the real estate market.
attribute data of any land parcel where the user has
clicked. On click by the user on any part of the map, a
new window pops up and the attribute information is
displayed in that window.
 23

Journal of Geomatics Vol 10 No. 1 April 2016

Horsley, G.J. (1992). Market value: The sacred cow.

Journal of Property Valuation and Investment, 10, 694-
700.

http://www.mapwindow.org/. Accessed on 22.06.2013

http://www.microsoft.com/visualstudio/ Accessed on
22.06.2013

http://msdn.microsoft.com/en-us/library/hh127540.
Accessed on 22.06.2013

Rangawala,S.C., K.S. Rangawala and P.S. Rangawala

(2011). Valuation of Real Properties. Charotar
Publishing House pvt. Ltd., Surat.

Tomić, H., H. Matijević, S. Mastelić Ivić and A.

Rončević (2006): Development of Land Valuation
Figure 7: Result of the sample use case query (a) System, FIG XXIII Congress, Munich 8-13 October
overlaid on ward map; and (b) overlaid on satellite 2006.
image
Wyatt, P,J. (1997). The development of a GIS-based
6. Conclusion property information system for real estate valuation.
International Journal of Geographical Information
A Real Estate Information System (REIS), which is Science, 11, 435-450.
useful for the valuation purpose, has been developed for
standalone windows platform using open GIS. This can Wofford, L.E. and Thrall Grant (1997). Real estate
also be linked to other databases of different problem solving and geographic information systems: A
departments so that user can access the other types of stage model of reasoning. Journal of Real Estate
data. This system can be improved by adding the market Literature, 5,177-201.
rate, which can help user a better understanding of the
real estate market. www.mp.nic.in Accessed on 28.06.2013

References www.99acres.com Accessed on 18.05.2013

Dale, P. and J. McLaughlin (1999). Land www.olx.in/real-estate-cat-16 Accessed on 18.05.2013

Administration. Oxford University Press, New York.
Yomralioglu, T. and R. Nisanci (2004). Nominal asset
Directorate of Town & Country Planning, (2001). land valuation technique by GIS. TS27.4, FIG Working
Development plan Bhopal 2031. Week May 22-27, Athens, Greece.

Goodall, B. (1972). The Economics of Urban Areas.

Oxford: Pergamon Press, New York.
 24

Journal of Geomatics Vol 10 No. 1 April 2016

Unfolding the time relationship of structural events through Landsat data: A case
study from Khandia formation, Champaner group, Gujarat
M.A. Limaye
Department of Geology, Faculty of Science
The Maharaja Sayajirao University of Baroda, Vadodara 390002
Email: [email protected]

(Received: Feb 23, 2016; in final form: Apr 13, 2016)

Abstract: Imprints of multistructural events recorded within the rocks are visualised through satellite data. The events
of superposed folding and shearing at Koba-Rustampura area belonging to Khandia Formation of Champaner group has
been studied. Classification of such multiple events becomes simple with the help of their respective trends. These
structural events have been delineated by using visual image interpretation techniques to study the spatial pattern and
textures on the Landsat image. By deciphering axial traces and directions of displacement, one can build the chronology
of the structural events revealing the deformational history.

Keywords: Remote sensing, Time relationship, Koba, Rustampura, Champaner group

1. Introduction 2. Data used and methodology

Applications of remote sensing in geosciences are well The true colour composite of Landsat image of 2016
established and widely accepted for identifying and with 30m spatial resolution has been used to identify
correlating structures, regionally. Identification of different structural events present within the study area.
structures on regional scale have been a necessity to Deformational events from later to former have been
understand overall deformational pattern of any terrain. interpreted by studying the spatial pattern on the image
Such attempts includes mapping of large scale features, through visual image interpretation techniques. The
extracting lineament patterns, identifying regional fold information collected during ground truth carried out in
trends, quantifying the fault directions, etc and its 2016, have been used to build the time relationship and
correlation up to plate dynamics (Nama, 2004; Kenea, to understand overall deformational history pertaining
1997; Heddi et al., 1999; Semere and Ghebread, 2006; to the Koba-Rustampura area.
Marghany et al., 2009; Maged and Mazlan, 2010;
Stefouli and Osmaston, 1986; Shuichi, 2002; Stamouslis 3. Regional geological setup
and Rogers, 2003; Yamaguchi and Naito, 2003; Rowan
and Mars, 2003; Gomez et al., 2005; Harding and The Koba-Rustampura area belongs to a part of
Berghoff, 2000; Misra et al., 2014; Joshi et al., 2014). Champaner group, which is well known for its low-
grade meta-sedimentary sequence. The group consists
In order to establish the correlation, in terms of regional of lithological entities such as quartzite, phyllite, meta-
structures, it is prerequisite to appreciate the continuity conglomerate, schist, impure dolomitic limestone and
from meso to micro scale. The present work reports a intermixed variety of granites and gneisses (Gupta et al.,
study on the time relation of structural events in Koba- 1992, 1995; Joshi et al., 2014). Geographically
Rustampura area which is situated 24 Km east of Champaner group is surrounded on three sides (i.e.
Vadodara district, Gujarat (falls under latitude and north, east and south) by younger plutonic intrusive
(Godhra granite) and one side (i.e. west) by Deccan trap
longitude 22o16’38.61” – 22o21’21.18” N and
rocks. Geologically the Champaner group represent an
73o28’25.63” – 73o38’51.38” E respectively). The
example of inlier due to the presence of younger rocks
study area belongs to Khandia formation of Champaner
neighbouring from all sides (Gupta et al., 1997).
group, Aravalli Supergroup having meso-proterozoic
age. The region has experienced polyphase
Structurally, rocks of Champaner group display two
deformational history and are characterised by
significant trends of axial traces. D1 phase of
lithological entities such as phyllite, quartzite, meta- deformation has resulted F1 folds of E-W trend where as
conglomerate. Based on the Landsat image of 2016, D2 phase of deformation has resulted F2 folds of N-S
acquired from the google earth portal, trend. The proximity of F2 folds decreases from eastern
characterisation of rocks holding different structural end to western end of Champaner group (Jambusariya
trends have been attempted. The same have been and Merh, 1967; Gopinath et al., 1977; Merh, 1995;
delineated in a chronological order and supplemented by Shah et al., 1984).
field as well as micro-structural studies.

© Indian Society of Geomatics

 25

Journal of Geomatics Vol 10 No. 1 April 2016

Interpretation of satellite imagery reveals that the Koba- direction of axial plane N-S. Based on the overall
Rustampura area manifests two major tonal and textural structural pattern suggested by folding it can be said that
variations with a distinct elevation difference between there are two sets of folds superposed on one another.
them on the image (Fig. 1). The high land region The first phase (F1) having E-W axial trace has found to
consisting of quartzite and meta-conglomerate display be superimposed by (F2) having N-S axial trace (Fig. 2).
light green tone with medium to rough texture, whereas
the low lying areas consist softer rocks such as phyllites
representing light brown tone and medium texture. The
south-western part of the study area represents crescent
shaped outcrop pattern whereas there has been
development of broad sinuous curve over the linear
ridge in the northern part. Furthermore, evidences of top
to NW, top to NE and down to SE displacement can be
appreciated, between the linear and crescent shaped
ridge and within the ‘C’ shape outcrop pattern
respectively.
Figure 2: Geological map of the study area

With the same connection there are good evidences of

shear present throughout the study area. The northern
limb, which shows the displacement with the major fold
present in the south-western part, shows the shear (i.e.)
top to NW. Also, within the major fold morphology
there has been generation of several shears resulting into
intrafolial fold at the southernmost margin of the study
Figure 1: Location map and Landsat data of the area. The direction of these shear include top to NE
study area shear and down to SE shear. The main litho-units
affected by shears are meta-conglomerate and quartzite.
In order to appreciate the overall deformational pattern The signatures include brittle fracturing and crushing of
of the study area, chronologically, restoration of quartzite, elongation of clasts in meta-conglomerate,
deformational events needs to be applied by unfolding formation of breccia and dragging of quartzite ridge
the terrain sequentially. With the help of image parallel to the shear plane. In addition to that
interpretation, it can be seen that the latest event, which microstructural analysis suggest dominant S-C fabric of
has occurred in the study area is shearing. Such idea can oriented mica grains, quartz fish, group 2 mica fish and
be profound due to its cross-cutting relationship ‘V’ pull apart mechanism with domino like
embracing on later deformational patterns (i.e) folding. microstructure in quartz clast of meta-conglomerate. In
quartzite evidences of shearing are supported by
The former structural event occurred in the study area is oriented mica flakes having inclusions of quartz aliened
folding, which is represented in the form of ‘C’ shaped in the direction of shearing. Furthermore, breccia
outcrop and broad sinuous curve over the linear ridge. consists of medium to coarse grain angular clasts of
The major fold event occurred in the study area has E- quartz cemented by fine grained quartz and Fe rich
W trend whereas minor one suggest N-S trend of axial matrix. Sweeping undulose extinction is observed
trace. Moreover, the N-S trending folds are developed within the coarser angular quartz clasts (fig. 3).
on Km long limb of E-W trending fold.
5. Discussion and conclusion
4. Ground truth verification and inferences
On the basis of satellite data interpretation and field
The Koba-Rustampura area represents the part of observation/ ground truth, similarity exists between the
Khandia Formation and located in the south-western fold morphology and shearing events. As per satellite
part of Champaner group. The main rock type includes image and field evidences the superimposed pattern of
meta-conglomerates, quartzites, phyllite and breccia. one fold event over the other has been confirmed. The
The study area constitutes a mega scopic westerly N-S trending open folds are found to be superimposed
plunging anticlinal fold. The northern limb is long in over E-W trending gentle fold. Considering regional
comparison to the southern limb and strikes E-W, structural setup, it can be observed that the two
having dip direction due north and due south significant fold events, F1 and F2 are occurred
respectively. Based on the attributes through throughout the group. Furthermore, the proximity of F 2
stereographic projection the fold has its axial trace E-W folds in the Champaner group increases from W-E.
with a plunge of 15o in the direction of N 270o. The axial Hence, in order to establish a time relationship between
plane is vertical, which strikes along the direction of fold fold events, E-W represent the first fold event, whereas
axes (i.e. N270o). Apart from that there has been second fold event is characterised by N-S axial trace.
generation of open folds on the northern limb having
 26

Journal of Geomatics Vol 10 No. 1 April 2016

Figure 3: Field Photograph showing: (a) Closely spaced factures in quartzites. Loc. Koba village; (b) Breccia
containing angular fragments of quartz embedded in ferruginous matrix. Loc. Rustampura village; (c) Elongated
clasts of quatz in meta-conglomerate, ball pen signifies the stretching direction. Loc. Koba village; (d) Group II
mica fish in meta-conglomerate (10XCN); (e) Quartz fish in meta-conglomerate (10XCN); (f) Development of S-C
fabric of mica grains in meta-conglomerate (4XCN); (g) Breccia containing medium to coarse grained angular
quartz clasts in ferruginous matrix (4XCN); (h) ‘V’ pull apart microstructure with domino like arrangement of
quartz clasts, arrows indicated shear direction (4XCN)
 27

Journal of Geomatics Vol 10 No. 1 April 2016

These fold events have undergone post deformational island, Washington state. Proceedings of the American
shearing along the weak planes. Directions of these Society of Photogrammetry and Remote Sensing,
shears includes top to NW, top to NE and down to SE Annual Conference, Washington, DC, 9 pp.
displacement. Field evidences, such as brittle fracturing
in quartzites and elongation of clasts in meta- Heddi, M., D.J. Eastaff and J Petch (1999).
conglomerate, gives sustainable sense of shear. In Relationships between tectonic and geomorphological
addition to field evidences, supportive microstructural linear features in the Guadix-Basa basin, Southern
evidences are also envisaged. These include dominant Spain. Earth Surface Process and Land-forms, 24,
S-C fabric of oriented mica grains, quartz fish, group 2 pp.931-942.
mica fish and ‘V’ pull apart mechanism with ‘domino’
like microstructure. Jambusaria, B.B. and S.S. Merh (1967). Deformed
greywacke conglomerates of Jaban near Sivrajpur,
Based on the above facts the time relationship of Panchmahals district, Gujarat. Ind. Minerals. 8, pp. 6-
deformational events is established. Chronologically it 10.
can be represented from older to younger as: 1. E-W
trending folds; 2. N-S trending open folds on limbs of Joshi, Aditya and M.A. Limaye (2014). Evidence of
earlier folds; 3. Shearing. syn-deformational granitoid emplacement within
champaner group, Gujarat. Journal of, The Maharaja
Acknowledgement Sayajirao University of Baroda, 49(1), pp. 45-54.

The author is thankful to Prof. L.S. Chamyal, Head, Joshi, Aditya, M.A. Limaye and B.S. Deota (2014).
Dept of Geology, M.S. University, Baroda for providing Structural footprints extraction from rocks of Lunavada
necessary facilities. The author is sincerely thankful to region,Gujarat through IRS LISS III. Journal of,
Mr. Aditya Joshi for academic discussion. Author is also Geomatica, 8(2), pp. 170-173.
grateful to the Forest Department of Vadodara and
Panchmahal District, Gujarat for granting the Kenea, N.H. (1997). Improved geological mapping
permission to carry out fieldwork in the Sanctuary area. using Landsat TM data, Southern Red Sea Hills, Sudan:
Author will never forget the help rendered by Mr. H.J. PC and his decorrelation stretching. International
Bhatt during the fieldwork. Journal of Remote Sensing, 18, pp. 1233-1244.

References Maged, M. and H. Mazlan (2010). Lineament mapping

using multispectral remote sensing satellite data.
Gomez, C., C. Delacourt, P Allemand, P Ledru and R Research Journal of Applied Sciences, 5(2), pp. 126-
Wackerle (2005). Using ASTER remote sensing data set 130.
for geological mapping, in Namibia. Physics and
Chemistry of the Earth, 30, pp. 97–108. Marghany, M., M. Hashim and S. Mansor (2009).
Geologic mapping of United Arab Emirates using
Gopinath,K., A.D. Prasad Rao, Y.G.K. Murty and K.K. multispectral remotely sensed data. American Journal of
Krishnaunni (1977). Precambrian of Baroda and Engineering and Applied Sciences, 2(2), pp. 476-480.
Panchmahals, Gujarat. Elucidation of stratigraphy and
structure. Rec. Geol. Surv. India.108, pp. 60-68. Merh, S. (1995). Geology of Gujarat. Geol. Soc. India,
Publ. Bangalore, p.244.
Gupta, S.N., R.K. Mathur and Y.K. Arora (1992).
Lithostratigraphy of Proterozoic rocks of Rajasthan and Misra, A.A., G. Bhattacharya, S. Mukherjee and N.
Gujarat - A review: Records of Geological Survey of Bose (2014). Near N-S Paleo-Extension in the western
India, 115, pp. 367-379. Deccan region, India: Does it link strike-slip tectonics
with India-Seychelles rifting?, 103,pp. 1645-1680.
Gupta, S.N., Y.K. Arora, R.K. Mathur, B.P. Iqbaluddin,
T.N. Sahai and S.B. Sharma (1995). Geological map of Nama, E.E. (2004). Lineament detection on Mount
the Precambrian of the Aravalli region, Southern Cameroon during the 1999 volcanic eruptions using
Rajasthan and northeastern Gujarat, India. Geological Landsat ETM. International Journal of Remote Sensing,
Survey of India Publication, Hyderabad. 25, pp. 501-510.

Gupta, S.N., Y.K. Arora, R.K. Mathur, B.P. Iqbaluddin, Rowan, L.C. and J.C. Mars (2003). Lithologic mapping
T.N. Sahai and S.B. Sharma (1997). The Precambrian in the Mountain Pass, California, area using Advanced
geology of the Aravalli region, Southern Rajasthan and Spaceborne Emission and Reflection Radiometer
NE Gujarat, India. Mem. Geological Survey of India (ASTER) data. Remote Sensing of Environment, 82,
123, pp. 262. pp.350-366.

Harding, D.J. and G.S. Berghoff (2000). Fault scarp Semere, S. and W. Ghebread (2006). Lineament
detection beneath dense vegetation cover: airborne characterization and their tectonic significance using
Lidar mapping of the Seattle fault zone, Bainbridge Landsat TM data and field studies in the central
 28

Journal of Geomatics Vol 10 No. 1 April 2016

highlands of Eritrea. Journal of African Earth Sciences, Stamoulis, V. and P. Rogers (2003). Geological
46, pp. 371-378. mapping for mineral exploration using ASTER data.
MESA Journal, 30, pp. 16-19.
Shah, A.M., R.V. Karanth and S.A. Barot (1984).
Geology of the area around Khandia with special Stefouli, M. and H. Osmaston (1986). The analysis of
reference to the lead mineralization, Dist., Baroda, linear geologic features on Landsat images of Crete.
Gujarat, Proc. Ind. Geol. Congress., pp. 127-134. Journal of the British Interplanetary Society, 39, pp.
546-551.
Shuichi, M. (2002). Regional lineament analysis and
alteration mineral mapping for intrusive-related copper Yamaguchi, Y. and C. Naito (2003). Spectral indices for
exploration in the Myanmar central volcanic belt. Asian lithologic discrimination and mapping by using the
Conference on Remote Sensing 2002 (Kathmandu, ASTER SWIR bands. International Journal of Remote
Nepal) Sensing, 24(22), pp. 4311-4323.

ISG Newsletter

Indian Society of Geomatics (ISG) brings out a newsletter which is very popular because
of its content on geomatics. The newsletter has featured special themes like desertification,
mountain ecosystem, watershed development, climate change etc.

The forth coming issue of ISG Newsletter will feature popular geomatics articles of current
interest.

ISG invites articles of general interest on current topics related to geomatics. The articles
may be sent to:

Shri R.P. Dubey, Editor, ISG Newsletter

E-mail: [email protected]
Phone: 02717-235434
 29

Journal of Geomatics Vol 10 No. 1 April 2016

Runoff estimation from a tributary of lower Tapi basin using SCS-CN method
integrated with remote sensing and GIS data
Sudhakar B. Sharma1, Anupam K. Singh2 and Ajay S. Rajawat3
1
Institute of Science, Nirma University, Ahmedabad
2
Institute of Engineering & Technology, J.K. Lakshmipat University, Jaipur
3
Geoscience Division, Space Applications Centre (SAC/ISRO), Ahmedabad
Email: [email protected], [email protected], [email protected]

(Received: Sep 28, 2015; in final form: Jan 27, 2016)

Abstract: The purpose of this research paper is to identify watersheds with high flood potential based on their
characteristics for formation of surface runoff. The SCS-CN method relies on remote sensing and GIS data for obtaining
watershed characteristics. A 30m raster grid size digital elevation model (DEM) has been generated from field survey
using Global Positioning System (GPS) of 3m accuracy integrating with Survey of India topographical maps of 1:50,000
scale having 10m contour interval. The undisturbed soil samples from field have been collected and laboratory analysis
was carried out using modified proctor compaction test as per ASTM D1557 and sieve analysis as per ASTM C136. This
has helped in establishing hydrological soil map while land use map has been prepared using Landsat 7 ETM+ image
band 2, 3, 4 (30m) merged with PAN band 8 (15m) for classification. The supervised maximum likelihood classification
approach has been employed for preparation of land use map for Varekhadi catchment having 442 km 2 of geographical
coverage. The major land use categories classified on 10 Nov 2001 Landsat 7 ETM+ image were agriculture (32%),
forest (29%), wasteland (20%), fallow land (14%), built-up (4%) and water bodies (2%). The hydrological soil groups
generated in GIS environment have identified two soil groups viz. group B and group C that exist under study area. The
Varekhadi catchment has been delineated into five watersheds viz. Amli, Zankhwaw, Visdaliya, Godsambha and Wareli
delineated using DEM and stream network. The SCS-CN model was applied for estimating of daily run-off for each sub-
watershed. The results obtained on the flood potential analysis shows that Wareli watershed has highest flood potential
while the Amli watershed lowest. It should be noted that highest value of flood potential belongs to lowest part of
watershed, where high population density is found. This analysis reflects an increased vulnerability and risks to floods
and inundations for Wareli watershed. Stream gauge data has been used for result validation with a common event of
2010 and it shows good agreement with the model. The flood potential analysis within the lower Tapi basin tributary
suggests that the SCS-CN method with hydrological parameters derived using remote sensing and GIS data can be applied
to predict run-off in poorly gauged watersheds.

Keywords: SCS-CN method, Remote sensing, GIS, Landsat 7 ETM+, Runoff, Ungauged catchment

1. Introduction wide range of design situations by the practicing

engineers and hydrologists. Sindhu et al. (2013) have
Surface runoff generation is dependent on climatic, geo- carried out estimation of surface runoff in Nallur
morphological, topographical and landuse Amanikere Watershed using SCS-CN method. They
characteristics of a catchment or watershed. Of above, found the variation in runoff potential with different
the topographical characteristics and soil types with land use/land cover and with different soil conditions.
land use as hydrological soil group have been of
immense importance. A combination of characteristics This method allows the identification and zoning of
favourable for runoff generation and runoff watersheds with a high risk of generating floods and of
concentration increases the flood potential in a those exposed to runoff generation processes. Soil and
watershed. Most of the watersheds in India have been landuse parameters, which control surface runoff, can
poorly gauged or un-gauged, as they do not have be evaluated and mapped through remote sensing
adequate records of runoff generation for a rainfall satellite images. Sharma and Singh (1992) in their
event to understand the hydrological response. In the research work for Luni river catchment have
flood prone catchments, it is required to calculate peak successfully used Landsat TM and SCS-CN model to
flood discharge or flood potential from each watershed. estimate runoff potential. Katimon et al. (2003)
A number of discharge estimation methods available in estimated flood potential of two small watersheds of
the literature namely Rational method, Soil Salengor and Pontian in Malaysia using SCS-CN
Conservation Service- Curve Number (SCS-CN) method and GIS based empirical approach to predict
method, Cook’s method and Unit hydrograph method. daily event storm runoff. They have pointed out that
However, the SCS-CN method for predicting direct run- SCS-CN and GIS have limitations in flood estimation in
off or discharge from rainfall excess of ungauged absence of accurate hydrologic soil group data. They
watershed is extremely important. Balvanshi and Tiwari have noted large variations in the surface runoff if
(2014) have reviewed SCS CN method for runoff hydrological soil group changes. Later, Behzad et al.
estimation and recommended that. It is being used in a (2012) used SCS-CN method for estimating flood

© Indian Society of Geomatics

 30

Journal of Geomatics Vol 10 No. 1 April 2016

potential for different return periods. They emphasised Gal khadi. The flood during August 2006 in LTB caused
on geomorphologic characteristics of Tarik flood basin huge damage to personal and property resulting into 300
in Iran. Several other methods on peak flood discharge people being killed and US$ 4.5 billion value property
and associated parameters estimation have been damage (Singh et al., 2009).
suggested in the literature for ungauged basins. Zhang
and Haung (2004) have developed Arc-CN tool Varekhadi catchment is a tributary of LTB having a
integrated with ArcGIS and applied SCS-CN method river length 50 km covering a geographical area is
for estimating run-off and preparing CN and run-off 442km2 (figure 1) which confluences near Mandvi
maps. However, the parameter reliability between town. The Varekhadi catchment has been divided into
various methods varies to a large extent and none is 5-watersheds consisting of lone urban centre Zankhwaw
found to be suitable universally. In spite of few along with almost 150 rural settlements. It has 2 major
limitations, SCS-CN along with hydrological soil group surface water reservoirs viz. Issar and Amli dams which
and land use remain to be a popular method for are located in the study area. The dam storage is mainly
estimation of flood potential under poorly gauged or used for flood control during monsoon season and for
ungauged catchments due to its performance and irrigation during the post-monsoon through gravity
reliability. canal system. The right bank canal from Kakrapar weir
located 30km upstream of Varekhadi confluence passes
SCS-CN has been applied in the present study for the through watershed and is being predominantly used for
estimation of run-off from five watersheds of Varekhadi irrigation purpose.
catchment- a tributary of lower Tapi basin. This SCS-
CN requires information on catchment characteristics
related to DEM, land use and hydrologic soil group for
estimation of catchment runoff. The purpose of this
method is to determine the curve number (CN) of the
catchment accurately that assesses the estimated runoff
potential. Hydrologic soil group, land use type,
vegetation cover are important physical characteristics
of a watershed used for the calculation of CN. Thus, the
most important step in estimation of surface runoff or
flood discharge is to calculate watershed characteristics
accurately. Remote sensing and GIS data together with Figure 1: Study area
field surveys and field measurements are input for
classification of watershed characteristics. Landsat 7 The geographic coordinates of the study area are
ETM+ image band 2, 3, 4 (30m resolution) have been 21014'N 73007'E to 21030'N 73030'E as lower left and
merged with PAN (15m resolution) data for supervised upper right corners. The study area receives an average
classification of land use classes using Gaussian yearly rainfall of 1376mm. The temperature in the
maximum likelihood classifier. Hence, the runoff CN catchment is variable which has a range of 22°C and
for different watersheds were determined using land use 40°C as minimum and maximum temperature. Major
and hydrological soil group map within the study area. land use categories are built-up area as settlements,
agriculture, forest, fallow land, water bodies and other
2. Study area uses. Considering the definition of land use CN and soil
types, the hydrological soil groups available in
Tapi river basin covers three states viz. Madhya Pradesh Varekhadi catchment are B and C. Major problem in
(9804 km2), Maharashtra (51504 km2) and Gujarat study area is flood in low laying areas near Wareli
(3837 km2) having a geographical area of 65145 km2 village at the confluence of Varekhadi catchment with
and is the India’s second largest inter-state westward main Tapi river.
draining river in Arabian sea. The Tapi river basin can
be classified in three zones, viz. upper Tapi basin, 3. Methodology
middle Tapi basin, and lower Tapi basin (LTB). The
area between Ukaidam to Arabian sea has been The proposed research methodology can be considered
considered as LTB, mainly occupying Surat and Hazira to have two parts. The part one is the modelling of the
twin city along with tens of small towns and villages spatial variability of topographical parameters using
along the river course. LTB has a geographical area of remote sensing and GIS while the part two involves
2920km2 which has been experiencing periodic floods analysis of the digital data base to derive hydrological
in urban settlements of Surat and Hazira. The Surat and model parameters. The research methodology used for
Hazira are about 106km and 122km downstream of estimation of surface run-off vis.a.vis flood potential
Ukai dam, respectively. Both these cities have been using SCS-CN method consist of five steps viz. Sub-
affected by recurrent floods during last 5-decades and watershed delineation, land use map and hydrological
flood frequency in the basin has been estimated to be soil group map generation, CN calculation and run-off
occurring once in 6-years. One among the major causes estimation. The flow chart of methodology is given in
of flood in LTB is attributed to formation of peak figure 2.
discharge early from various tributaries such as
Varekhadi, Anjanakhadi, Serulkhadi, Mau khadi and
 31

Journal of Geomatics Vol 10 No. 1 April 2016

Figure 2: Flow chart of methodology

In this paper, hydrological data related to runoff

estimation such as DEM, Landsat 7 ETM+ satellite
imagery, global positioning system (GPS) for level
points, soil map for grain size and soil moisture, land
use for CN, hydrological soil group as AMC (antecedent
moisture condition) and rainfall have been used. A
DEM of 30m raster grid size has been generated from
field survey using GPS of 3m accuracy integrating with
Survey of India topographical maps of 1:50,000 scale
having 10m contour interval.

3.1 Watershed delineation

A DEM of 30m raster cell size and 0.5m vertical
accuracy has been used to delineate 5-watersheds for
Varekhadi river basin using BASIN hydrological
model. The steps followed in a given sequence are (i) Figure 3: Digital Elevation Model (DEM)
creating a depressionless DEM; (ii) calculating flow
direction based on 3x3 cell neighbourhood algorithm; Image geometric correction has been done and land use
(iii) calculating flow accumulation and identify cell map were derived using Gaussian maximum likelihood
having given area; (iv) delineation of watershed outlet algorithm of supervised classification with field sample.
points leading to delineation of watersheds for a given Land use categories considered in the study area are
threshold area. Five such watersheds have been built-up land, agriculture, forest, fallow land, water
delineated named as Amli, Zankhwaw, Visdaliya, bodies and other as shown in figure 4 and their statistics
Godsambha and Wareli. Watershed parameters such as are given in table 1. SCS-CN method related land use
flow length, river length, watershed outlet point, description as an input for runoff generation process.
watershed area, river length and river slope were The CN can be empirically determined based on SCS
obtained from DEM (Figure 3). The validation of DEM (1985) land use description, hydrological soil group and
has been done using 24 ground control level points AMC conditions
using GPS.
3.3 Generating hydrological soil group map
3.2 Generating landuse map To create hydrological soil group map, soil survey of
The land use map of Varekhadi watershed has been study area was conducted. Thirty points were selected
generated by image analysis of satellite data. The image for soil sample in study area and soil samples have been
of Landsat 7 ETM+ (10 Nov 2001) bands 2, 3, 4 with analysed in the laboratory. Soil properties have been
(30m) spatial resolution and PAN with (15m) ground identified and a GIS map of hydrological soil group was
resolution were analysed. prepared. There are two type of soil in study area group
B and C as shown in figure 5.
 32

Journal of Geomatics Vol 10 No. 1 April 2016

initial losses (P- Ia) is equal to the cumulative
infiltration as given in equation 2 (Mishra and Singh,
2003).

P = Q + F + Ia ………………….. (1)

Q / (P-Ia) = F / S …………….. (2)

where the terms P is total precipitation (mm); Ia is the

initial abstraction (mm); F is cumulative infiltration
(mm); Q is direct runoff (mm); S is the potential
maximum retention or storage capacity of soil (mm). As
per USDA-SCS (1985) guidelines the initial abstraction
(mm) is assumed to be abstraction fraction (usually λ=
0.2) of the potential maximum retention as shown in
Figure 3: Land use / land cover map equation 3.

Ia = .S ……………………………….. (3)

The direct storm runoff Q (mm)can be related to the
effective rainfall and actual retention through the water
balance equation 4 (Yu, 1998). Equation 4 is valid only

Ia  0.2S
When P  S and generally , hence
equation 4 can be written as;

Qp = (P-Ia)2/(P-Ia+S)

= (P-0.2S)2/(P+0.8S) …….…. (4)

If effective rainfall P  S , then direct storm runoff Q

Figure 4: Hydrological soil map (mm) is taken as zero
3.4 Generating CN map
In practice, the potential maximum retention S (mm) of
The CN value has been used to estimate potential
the soil is determined using the CN given in equation 5.
maximum soil retention. The value of CN is 100 for
impervious surfaces and between 0<CN≤100 for other
S = (254400/CN) – 254 ………….. (5)
surfaces. The maximum potential storage which relates
to CN depends on land use, hydrological soil group,
The term CN is determined from a table based on land
hydrological condition and AMC (Ponce and Hawkins,
use, hydrological soil group and AMC. The
1996). Arc CN extension of ArcGIS software has been
hydrological soil group has four classes A, B, C and D
used to generate the CN map of Varekhadi watershed.
based on landuse, hydrological soil group and AMC.
The hydrologic soil group field from the soil map and
The hydrological soil group has four classes A, B, C and
the land use field from the land use map were selected
D based on infiltration rate of soil. AMC has three
for intersection under vector environment in GIS. After
classes I, II and III according to rainfall limits for
intersection, a map with new polygons representing the
sowing and growing season.
merged soil hydrologic group and land use (land-soil
map) was generated.
3.6 Stream gauge installation
Varekhadi has no discharge measurement gauging
CN map directly effect on surface runoff, all the surface
station and is classified as un-gauged watershed. This
water bodies gives the CN value zero and resulted run
being a remote location, it has been proposed to install
off is also zero. Average CN also have been calculated
automatic sensors with data logger capabilities. WL-
and the details are given in table 2. In shows that Wareli
16U stream gauge sensors of 25m cable from Global
watershed has highest CN value 86.07 and Zankhwaw
Water USA were procured and installed in field during
has lowest CN value 68.95.
June 2010. The sensor has 0.1mm measurement
accuracy and can record 10 reading per second. Three
3.5 Run-off estimation
discharge sites viz. Amli, Visdalia and Godsamba have
The SCS-CN method assumes that surface runoff will
been selected for installation of stream gauge. The
be generated once initial losses are satisfied. The SCS-
output data has been used for result validation for
CN method explaining the water balance can be given
deferent sub watershed.
by equation 1. The main hypothesis in this method is
that the ratio of direct runoff to the rainfall depth minus
 33

Journal of Geomatics Vol 10 No. 1 April 2016

4. Results and discussion sothat it gives low run-off. Geospatial distribution of
CN value in figure 6
The direct storm runoff depends on land use, soil type
and hydrological soil group and CN. Landsat 7 ETM+
data of 10 Nov 2001 has been used for supervised
classification of 5-land use classes. A merge product of
Landsat 7 ETM+ band 2, 3, 4 of 30m cell size with PAN
band of 15m spatial resolution was used for
classification. The land use distribution within
Varekhadi catchment shows that it is primarily an
agriculture catchment as 32% of study area is under
agriculture, 46% of area under agriculture and current
fallow; 66% of area under wasteland, fallow land and
agriculture (Table 1).

Table 1: Landuse Classification

Land use Area km2 % of total area

Agriculture 142 32
Fallow Land 60 14
Figure 5: Curve number (CN) map
Wasteland 87 20
Built-up land 17 4 Runoff depth value calculated for each grids from the
Water bodies 7 2 equation. Figure 7 depicts the distribution of runoff
Total 442 100 depth upon 262mm rainfall. Figure 8 (a), (b), (c), (d), (e)
depict the graphs of runoff vs. rainfall data will conform
The value of volume and surface runoff were calculated to a simple linear relationship with a good fit
for extracting percentage flood contribution for each polynomial trend line. The 10 years rainfall data from
sub-watershed which is shown in table 2. 1998-2008 run-off has been used to calculate runoff for
each sub-watersheds and graphs was drawn. The
Table 2: Estimation of runoff for each watershed (1999- correlation coefficient R2values are similar for all 5 sub-
2008) watershed 86.07 weighted CN so that Wareli is a high
flooding potential watershed and Amli has 68.95
Location Zankhw Amli Godsa Vishd Wareli weighted CN so that it gives low run-off. The details
aw mba alia run-off depth map is given in figure 7.
Area 100.52 91.85 78.85 71.05 99.73
(km2)
Retention 83.40 114.12 100.6 75.95 41.11
S (mm)
Weighted 75.28 68.95 71.63 76.98 86.07
CN (-)
Rainfall 277.0 267.5 251.0 251 252
(mm)
Runoff 197.15 166.68 160.81 178.77 201.73
(mm)

It can be seen that sub-watershed Wareli has high

flooding potential and Amli has low flooding potential.
Flooding potential depends on CN and CN depends on
land use and soil properties of watershed which is
clearly reflected in results.

It is required to estimate a coefficient that reduces the

total rainfall to runoff potential after losses in terms of
evaporation, absorption, transpiration and surface Figure 6: Runoff distribution
storage. It can be stated that the higher the CN value
leads the higher runoff generation. Wareli watershed Regression analysis has been carried out for each sub-
has 86.07 weighted CN so that Wareli is a high flooding watershed. This can be used for calibration purpose as
potential watershed and Amli has 68.95 weighted CN given in table 3.

Stream gauge data were used for result validation

(figure 9) for a common event of 2010. It is found that
model can give a good and accurate result for ungauged
catchment.
 34

Journal of Geomatics Vol 10 No. 1 April 2016

Table 3: Regression relationship between rainfall
and runoff for the sub watershed

Sub- Second order polynomial relationship

watershed between rainfall [X] and runoff [Y];
Y=AX2+Bx
A B R2
Zankhwaw 0.0024 0.1142 0.9830
Amli 0.0025 0.0229 0.9857
Visdalia 0.0028 0.1171 0.9827
Godsamba 0.0027 0.01419 0.9863
(a) Wareli 0.0022 0.3504 0.9847

(b)

Figure 9: Results of validation of predicted or

calculated with stream gauge data

5. Conclusion

Remote sensing and GIS data are of great use for surface
runoff estimation when conventional methods of runoff
estimation are inadequate. Both the techniques have
been used for generating model input for determination
(c) of physical characteristics of watershed such as land
use, hydrology soil group and CN number. Gaussian
maximum likelihood classifier has been used for
classification of land use and shows good field
acceptability. It has been integrated with SCS-CN
method for identification of watershed, estimation of
flood potential for a part of lower Tapi basin.

This analysis provides satisfactory results for rainfall-

runoff modelling. It will be useful for flood forecasting,
flood contribution of each watershed and flood
(d) discharge measurements. The method may be good tool
for runoff estimation for lower Tapi basin and un-
gauged catchment like Varekhadi catchment. It can also
state that higher the CN value of catchment leads to
more runoff and gives highest contribution for flood.

Acknowledgement

The authors express their sincere thanks to State Water

Data Centre officers Mr MG Golawala, Superintending
(e) Engineer and Mr SG Shah, Deputy Executive Engineer
for supplying daily rainfall data of 8-stations in and
Figure 8: Results of rainfall runoff relationship for around study area. We acknowledge assistance rendered
sub-watersheds (a) Amli; (b) Zamkhwaw; (c) by graduate students of Department of Civil
Visdalia; (d) Godsamba; and € Wareli Engineering, Nirma University for carrying out field
surveys and soil analysis in laboratory.
 35

Journal of Geomatics Vol 10 No. 1 April 2016

Reference

Balvanshi, A. and H.L. Tiwari (2014). A comprehensive

review of runoff estimation by the curve number
method. International Journal of Innovative Research in
Science, Engineering and Technology, Vol.3 (11),
pp17480-17485.

Behzad, A., R.S. Mohsmmsd and E. Moghimi (2012).

Estimating flood potential emphasizing on
geomorphologic characteristics in Tarikn basin using
the SCS method. European Journal of Experimental
Biology, Vol. 2 (5), pp 1928-1935.

Katimon, A., M. Zulkifli and M. Yunos (2003). Flood

potential estimation of two small vegetated watershed.
Journal Kejuruteraan Awam, vol 15 (1), pp 1-15.

Mishra, S.K. and V.P. Singh (2003). Soil conservation

service curve number methodology. Volume 43.
Kluwer Academic Publishers, Dordrecht, The
Netherlands.

Ponce, V.M. and R.H. Hawkins (1996). Runoff curve

number: Has it reached maturity? Journal of Hydrologic
Engineering, 1(1): 11-19.

SCS (1985). National engineering handbook, Section 4:

Hydrology, Soil Conservation Service, USDA,
Washington, DC.

Sharma, K.D. and S. Singh (1992). Runoff estimation

using Landsat Thematic Mapper data and the SCS
model. Hydrological Sciences 37 (1/2): 39-52.

Sindhu D, B.L. Shivakumar, A.S. Ravikumar (2013).

Estimation of surface runoff in Nallur Amanikere
watershed using SCS-CN method. International Journal
of Research in Engineering and Technology (2) pp 404-
409

Singh, A.K. and A.K. Sharma (2009). GIS and a remote

sensing based approach for urban flood-plain mapping
for the Tapi catchment, India. In IAHS Publ. 331, 389-
394.

Yu, B. (1998). Theoretical justification of SCS method

for runoff estimation. ASCE Journal of Irrigation and
Drainage Engineering, 124: 306-310.

Zhan, X. and M.L. Huang (2004). ArcCN-Runoff: An

ArcGIS tool for generation curve number and runoff
maps. Environmental Modelling and Software 19: 875-
879.
 36

Journal of Geomatics Vol 10 No. 1 April 2016

Change detection on the Volta river due to the construction of the Kpong dam using
remote sensing techniques
E.M. Osei Jnr1, A.S. Amoah2, B.E.K. Prah1, F.K. Anyah3 and K.A. Addo-Gyan3
1Department of Geomatic Engineering, Kwame Nkrumah University of Science and Technology, Ghana
2Department of Civil Engineering, Kumasi Polytechnic, Ghana
3Survey and Mapping Division, Ghana Lands Commission, Ghana

Email: [email protected]

(Received: July 08, 2015; in final form: Oct 09, 2015)

Abstract: Construction of dams brings about changes and this is the situation on the Volta River with the construction
of the Kpong dam. The aim of this study is to detect these changes and the trend of change after some years of construction
of the dam. Topographic map of the lake (1974), Landsat Thematic Mapper (TM) image (1990), Landsat Enhanced
Thematic image (2000) and Landsat (UTM) image (2010) were used. The images were georeferenced and classified into
six (6) classes namely, closed forest vegetation, open forest vegetation, dense shrub, grass/herbaceous cover, built up/bare
surfaces and water body for the purpose of our study, using supervised classification. Maps were produced to show the
changes in the land cover features. Careful observations of the produced maps showed that the most dominant change
happened after the construction of the dam. wherebythe river overflew its boundary submerging some islands and
communities along the river.This caused the Islands to reduce from 340.607ha in 1974 to 319.959ha in 1990 whilst the
lake increased from 2266.398ha in 1974 to 4007.07ha in 1990. After that the water increase has been gradual and all
other changes have been as a result of anthropogenic activities.

Keywords: Volta river, Kpong dam, Supervised classification, Land cover changes, Landsat images

1. Introduction on the Volta river, islands on it and communities along

it.
Construction of hydro electric dam comes as a joy to all
Ghanaians since it will increase the power generation Remote sensing provides synoptic view of the terrestrial
capacity and connect rural communities also to the landscape and is used for inventorying, monitoring, and
national grid which helps in the development of a change detection analysis of environmental and natural
nation. Although having far reaching benefits, they also resources (Narumalani et al., 1997). Although remotely
exert a number of adverse impacts as a result of sensed images seldom replace the usual sources of
potential negative impacts from the construction of information concerning water resources, they can
infrastructures (Tortajada, 2001; Ledec and Quintero, provide valuable supplements to field data by revealing
2003) on their immediate environment. broad scale patterns not recognizable at the surface,
The Kpong dam was constructed to purposely generate recording changes over time, and providing data for
electricity for the industrial and domestic uses to inaccessible regions (Campbell, 1996).
supplement that of Akosombo. The completion of the
Kpong dam with a capacity of 160MW in 1981 has Change detection analysis, which is the process of
raised the power generation capacity of the hydro identifying differences in the state of an object or
electric power projects on the Volta river to 1,072MW phenomenon by observing it at different times (Singh,
(Amankwaa, 2002). 1989), is employed with multi temporal data sets to
discriminate areas of land cover change between dates
Even though the main aim for the construction was of imaging of the Kpong dam and its environs from
attained, Girmay (2006) asserts that at the time of the 1974 to 1990 and 1990 to 2010.
construction Environmental Impact Assessment (EIA)
was not a planning and management tool available in The main aim of the work is to apply remote sensing
Ghana and in view of that, several issues of techniques to detect the extent of land cover changes
environmental impacts, have not been considered under that have occurred as a result of the construction of the
mitigation measures as should have been done. Kpong dam and the trend of change.
In most instances, changes along water bodies result in
environmental, social and economic impacts of greater 2. Materials and methodology
damage (i.e. flooding) than benefit to the area (Moshen,
1999). Also coastal zones are most vulnerable for land 2.1 Study area
use changes in this rapid industrialization and The Kpong dam is located in the Lower ManyaKrobo
urbanization epoch. It is necessary to evaluate land use district in the Eastern region of Ghana. Its coordinates
– land cover changes to develop efficient management are 6°4'60" N and 0°12'0" E. The administrative capital
strategies (Prabaharan et al., 2010). It is therefore of the district is Odumase. The district covers an area of
important to look at some of effects of the Kpong dam

© Indian Society of Geomatics

 37

Journal of Geomatics Vol 10 No. 1 April 2016

2
1,476 km , constituting about 8.1% of the total land area shows the size as at 1990 after the construction of the
within the region (18,310 km2). dam.

Obviously there has been an increase in the river size

after the Kpong dam was constructed.

This increase occurs especially at the upstream of the

river which caused the displacement of some
communities. The Islands decreased from 340.61 ha to
319.96 ha whilst the lake increased from 2266.40 ha to
4007.07 ha. The change can be clearly seen in figure 2.
According to the Volta river authority eight villages
were fully submerged and these include Vivokope,
Fremankope, Lomen, Nobotsukope, Fodzoku,
Ageteklekyi, Pokyenu and Gabrunya and the villages
that are partly submerged are four namely Glornu, Kasa,
Alabonu, Klamadaboe. The classified maps depict the
land cover/ use of the area for the three different years
at an interval of ten years, i.e. in 1990, 2000 and 2010.
Amount of change in area and percentages were
Figure 1: Google earth image showing the study area
generated to quantify the changes.
The major towns in the district include Odumase
township (which incorporates Atua, Agormanya and
Nuaso), Akuse and Kpong in the Lower Manya area The
district shares boundaries with Upper ManyaKrobo
district to the north, to the south with DangmeWest and
YiloKrobo respectively, to the west with YiloKrobo
municipal and to the east with Asuogyaman district.

2.2 Methodology
Landsat Thematic Mapper (TM) image of 1990,
Landsat Enhanced Thematic Mapper (ETM) 2000 and
Landsat Thematic Mapper (UTM) 2010 of the study
area on the Volta tiver were obtained. These images
were dry season images captured in the months of
March and April. The images underwent radiometric
and geometric corrections. Figure 2: Map showing the boundaries of the river
at 1974 and 1990 after construction of the dam
The topographic map of 1974 and the 1990 image were
used to produce another map showing the boundaries of
the lake for the two different years. The Landsat images
were classified into six (6) different land covers namely
closed forest, Open forest, dense shrub, grass, built
up/bare lands and water body using maximum
likelihood method of supervised classification.
Classifications for the various images were guided by
observations on the ground, local residents and other
techniques such as PCA and NDVI. Statistics were
generated for these classes on each image to know the
area covered by each land cover. Each classified image
was superimposed with the data of the boundary of the
river.

3. Results and analysis

A composite map of the river was produced from the

map of 1974 and Landsat image of 1990 which showed
the difference in size of the river before and after the
construction of the dam. From figure 1, two shades of
blue are used. A deep blue shows the river as at 1974
before the Kpong dam was constructed and light blue Figure 3: Land cover/use map of 1990
 38

Journal of Geomatics Vol 10 No. 1 April 2016

1974 and 1990. Land cover has also encountered
changes due to anthropogenic activities.

Table 1: Total acreages of the individual islands and

average area of the water body before and after the
construction
Distribution of Land cover/use classes for the years 1990,
2000 and 2010
Class 1990 2000 2010
Area % Area % Area %
(Ha) (Ha) (Ha)
Closed 11049.8 20.9 4613.1 8.71 3479.8 6.57
forest
Open 24918.5 47.1 21833 41.2 18937 35.8
forest
Dense 8082.13 15.3 18857 35.6 20691 39.1
shrub/her
baceous
Grass/he 4180.43 7.89 2563.9 4.84 4506 8.51
rbaceous
Built 718.47 1.36 982.8 1.86 1213.6 2.29
up/bare
surfaces
Water 4007.07 7.57 4106.7 7.75 4129.6 7.8
Figure 4: Land cover/use map the year 2000 Body
Total 52956.4 100 52956 100 52956 100

Table 2 shows the rate of change of the different classes

and this has been interpreted in chart 1. All the land
covers that reduced in area have their bars along the
negative axis whilst those that gained are on the positive
axis. Closed and open forest vegetation always
decreased. Changes in the built–up/bare and water body
are not that significant even though they have increased
along the years. Dense shrub has also increased along
the years. Grass/ herbaceous however has no consistent
change, from 1990-2000 there was a decrease then from
2000-2010 there was an increase in change. All these
changes are due to human activities.

Table 2: Percentage and area rate change of each

land cover

Land Cover 1990-2000 2000-2010

Area % Area %
Closed forest -6436.8 -12.2 -1133.3 -2.1
Open forest -3.85.8 -5.82 -2895.9 -5.5
Dense 10775.1 20.35 1833.47 3.46
Shrub/herb
aceous
Grass/herb -1616.5 -3.05 1942.06 3.67
Figure 5: land cover/use for the year 2010 aceous
Built up/bare 264.33 0.5 230.76 0.43
4. Conclusion surfaces
Water Body 99.65 0.18 22.91 0.05
All the results of this study have revealed the extent of
change in the Lake after the dam construction and According to Agbenyo, (2009) hydropower-projects
changes in land cover from 1990 to 2010. It is obvious which include the Kpong dam on the Volta river have
from the resultant thematic map that the size of the lake brought untold hardship onto the lives of the people,
has almost doubled and this happened between the years influencing human-induced environmental degradation
in the area. Activities like farming, firewood and
 39

Journal of Geomatics Vol 10 No. 1 April 2016

charcoal burning and out-migration among a number of Akosombo/Kpong dams in Ghana; MSc Thesis:
coping measures have been adopted by people in the Department of Land and Water Resources Engineering,
area to make livelihood comfortable. Royal Institute of Technology, Stockholm, Sweden.

The accuracy of the 1990 and 2000 images were not Ledec, G. and J.D. Quintero (2003). Good dams and bad
verified but that of 2010, 100 points were picked with dams: Environmental criteria for site selection of
the GPS and used to validate the classification. This hydroelectric projects. Sustainable Development
gave an overall accuracy of 84% and a Kappa of Working Paper16, November 2003
80.25%.
Moshen, A. (1999). Environmental land use change
Acknowledgement detection and assessment using with multi – temporal
satellite imagery. Zanjan University.
Sincere thanks to the Survey department of the Centre
for Remote Sensing and Geographic Information Narumalani, S., Y. Zhou and J.R. Jensen (1997).
System (CERSGIS) and Environmental department of Application of remote sensing and geographic
the Volta River Authority. information systems to the delineation and analysis of
riparian buffer zones. Aquatic Botany 58: 393-409.
References
Prabaharan, S., K. Srinivase Raju, C. Lakshumanan and
Agbenyo, P.K. (2009). Effects of the Akosombo and M. Ramalingam (2010). Remote sensing and GIS
Kpong power schemes on six selected mafi- applications on change detection study in coastal zone
communities in the Volta region. MSc Thesis: The using multi temporal satellite data. International Journal
Institute for Development Studies, Universsity of Cape of Geomatic and Geosciences, Volume 1, No. 2, 2010.
coast, Ghana
Singh, A. (1989). Digital change detection techniques
Amankwaa, O.P.J. (2002). Ghana: A human using remotely sensed data. International Journal of
geography for secondary schools. Takoradi, Ghana: St. Remote Sensing. Vol. 10, No. 6, p. 989-1003.
Francis Press Ltd.
Tortajada, C. (2001). Environmental sustainability of
Campbell, J.B. (1996). Introduction to remote sensing. water projects. Doctoral thesis: Division of Hydraulic
Second edition, The Guilford Press, USA. Engineering, Department of Civil and Environmental
Engineering, Royal Institute of Technology,
Girmay, Y. (2006). Assessing the environmental Stockholm.
impacts of a hydropower project: The case of
 40

Journal of Geomatics Vol 10 No. 1 April 2016

An objective method for detecting night time fog using MODIS data over northern
India
Sasmita Chaurasia and B.S. Gohil
Space Applications Centre, ISRO
Ahmedabad- 380015 (INDIA)
Email: [email protected], [email protected]

(Received: Feb 22, 2016; in final form: Mar 29, 2016)

Abstract: An objective method for detecting night time fog based on bi-spectral difference of 3.9µm and 11µm channel
brightness temperature has been developed. The thresholds used in this are dynamically derived, based on the data and
have been tested with MODIS AQUA/TERRA data for December 2012-January 2013 and December 2013-January 2014
winter season over northern India. The generated fog maps for December 2012-January 2013 have been validated
qualitatively with the fog map generated by IMD (India Meteorological Department) using MODIS data. Quantitative
validation has been carried out for January 2014, against visibility data at five locations of Northern India. Fog was
detected with 70% success rate using this objective method over this region for the validation period.

Keywords: Fog, MODIS, Bi-spectral, Brightness Temperature Difference (BTD), Indo Gangetic Plains (IGP)

1. Introduction (Ram Kripa et al., 2012). Emissions from agricultural

waste and biomass burning dominate during winter time
Fog is a meteorological weather phenomenon in which among other major source of aerosols.
the cloud has its base very close to ground and the
visibility reduces to 1000m or less. Apart from its Even though the effect of winter fog in the north Indian
impact on the transportation system as an important region is very high it has not been studied extensively.
socio-economic factor, it has great impact in the field of Specifically, its formation, spatial extent and evolution
meteorology, climate studies as well as on human health are required to be studied in detail. It is difficult to
and crops. Therefore, an improved understanding, monitor the spatial and temporal extent of fog over
monitoring, forecasting and nowcasting of fog will large-scale areas such as the IG plains using a limited
benefit the society as a whole (Gultepe et al., 2009). conventional and ground based observational network,
so satellite based fog monitoring becomes popular due
Over northern India, fog occurs for a considerable to its enhanced spatial, spectral and temporal resolution
period particularly during winter season (Singh et al., and offers new opportunities for near real-time fog
2004; Dutta et al., 2004; Tiwari et al., 2011). The Indo- detection and monitoring.
Gangetic (IG) plains along the Himalayan region form
a trough region, where cold air drainage flowing from The night time fog detection is very important because
the higher plateau gets collected leading to the fog at night causes difficulties in aviation, land and
enhancement of relative humidity. Under clear sky marine transport. Numbers of surface observations are
condition, fog forms basically due to radiative cooling also limited during night because many of the stations
of the earth’s surface during winter season. In this do not operate during this time. Because of the low
period a series of low and high pressure zones moves density of surface observation at night, and to get
from NW to NE along the Himalayas. During a low information about the complete coverage of fog, remote
pressure zone, it may rain to add more moisture to the sensing technology provides better opportunity. Many
atmosphere. This moisture content immediately after approaches have been developed to detect fog at night
the high pressure zone gets condensed due to radiative (Eyre et al., 1984; Turner et al., 1986; Bendix, 2002).
cooling of the surface, which may lead to the formation Detection of fog at night using satellite observation
of dense fog over a wide region (Singh et al., 2004). relies on the thermal emission of the surface. As it is
Apart from the meteorological condition the local levels not contaminated with solar radiation, the false alarm in
of surface as well as atmospheric conditions also leads night time fog detection is less. Most of these methods
to the intensification of fog. Once fog layer is formed it are based on the particular emissive properties of fog at
persists for the whole duration of the passage of the high 3.9µm and 10.8µm wavelengths (Bendix and
pressure zone and typically may extend from Punjab to Bachmann, 1991). The small droplets found in fog are
the Bay of Bengal region along the trough. In the IG less emissive at 3.9µm than at ~11µm, whereas for
plains, it is also observed that during winter season i.e. larger drops both emissivities are roughly same (Hunt,
from November to February the wheat fields are 1973). Thus the difference between brightness
irrigated which adds to the relative humidity over this temperatures at these wavelengths are useful and tested
region. The pollution over the metro city Delhi and for detecting fog against other clouds and land features.
surrounding is also responsible for fog formation due to The method has been successfully implemented on
availability of sufficiently large condensation nuclei Geostationary Operational Environmental satellite

© Indian Society of Geomatics

 41

Journal of Geomatics Vol 10 No. 1April 2016

which provides the required spectral bands (Ellord, radiation from land and fog/cloud is the main radiation
1995; Wetzel et al., 2004; Lee et al., 1997; Greenwarld source. The higher the temperature greater will be the
and Christopher 2000; Underwood et al., 2004). radiation from surface and fog will have higher
Recently Cermak and Bendix (2007, 2008) have applied temperature compared to medium and high level clouds.
the same algorithm to data of Spinning Enhanced Thus brightness temperature of fog will also be higher
Visible and Infrared Imager (SEVIRI) on board than that of medium and high clouds. However, in case
Meteosat Second Generation (MSG). As an extension to of fog there is always temperature inversion, so that the
earlier work (Chaurasia et al., 2011), in this study an temperature of fog is similar or even higher than that of
objective approach has been developed for the detection surface. Therefore, long wave IR channel can be used to
of fog during night time over the IG plains.Asthe eliminate medium and high clouds whereas it is very
brightness temperaturedifference depends on the optical difficult to differentiate between fog and the under lying
depth, particle size, type of fog etc. (Chaurasia et al., surface using thermal channel alone. The brightness
2011) a fix threshold leads to false detection or there is temperature of the medium/ high cloud is the lowest and
a probability of miss. In order to minimize this a almost below 270° K (Yang et al., 2008). In order to
dynamic thresholding approach has been developed and determine this threshold dynamically, the histogram of
the thresholds are dynamically set for each image under 10.8µm channel brightness temperature are examined.
consideration. It is smoothed to reduce spikes. The histogram of the
images corresponding to both the channels show two
2. Data used maxima around one minimum (shown in figure 2a, 2b).
The first maxima correspond to low brightness
Night time fog detection has been carried out using data temperature values, which corresponds to pixels with
from MODIS AQUA/TERRA. The MODIS sensor on high clouds and snow. Using the BT corresponding to
board the EOS TERRA and AQUA platforms shows the the minima in TIR channel, high clouds and snow are
best spectral resolution from operational Low Earth eliminated. If more than one minimum is found then a
Observation (LEO) systems with 36 spectral bands from threshold of 270°K for the TIR channel can be used to
visible to thermal infrared portion of the eliminate the high clouds and snow pixels considerably.
electromagnetic spectrum (Schueler and Barnes, 1998; The new image now contains only low clouds, fog as
Levy et al., 2007) (29 spectral bands with 1km, 5 well as pixels corresponding to land. For bi-spectral
spectral bands with 500m and 2 spectral bands with differencing brightness temperature difference image
250m nadir pixel dimensions). The MODIS Terra/Aqua (BTD=BT10.8µm – BT3.9µm) is generated.
Level 1B data for night time over north India were
obtained from http://ladsweb.nascom.nasa.gov for the
period December 2012 to January, 2013 and December MODIS
2013 to January 2014. The data of MODIS AQUA Terra/Aqua data
having passes over India around 2000UTC is more
appropriate for night time fog analysis, because the
formation of fog gets initiated at night and is more DN to BT conversion
intense towards early morning. However, the MODIS
Terra data at an earlier over pass time from 1600UTC to
1900UTC is also used, because it covers more area of
the IG plains of India. Gross cloud check Not a Cloud

3. Methodology Cloud

Elimination of high clouds

The flowchart for the objective dynamical thresholding
method as has been developed for the present study is
shown in figure 1. Besides using the thresholding
scheme in the brightness temperature difference image, Elimination of snow pixel
an attempt has also been made to eliminate the false
alarm due to other cloud types and snow by using
behavior of cloud and snow in individual channel.
Cloudy/Foggy Data

The Level 1B emissive channels were geo-referenced

and the data corresponding to 3.9µm and 10.8µm (Band
22 and 31 of MODIS channels) of MODIS were Bi-spectral Fog/ Low
ΔTB min <ΔTB<ΔTB max
Differencing Stratus
extracted. The radiance values of both channels were
converted to brightness temperature through Planck’s
function. As mentioned in previous paragraph for gross Figure 1: Flow chart for night time detection of fog
cloud check the long wave infrared or thermal infrared using MODIS data
(TIR) channel centered around 10.8 µm has been used.
In the long wave infrared (10.8 µm) channel the
 42

Journal of Geomatics Vol 10 No. 1April 2016

(a) (b)
Normalized Histogram of 10.8 µm BT Normalized Histogram of 3.9 µm BT

1 1

Normalized Histogram

Normalized Histogram
0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0
0
220 240 260 280 300
220 240 260 280 300

BT 10.8 μm BT 3.9 µm

Figure 2: Histogram of brightness temperature in two channels (a) 10.8 µm and (b) 3.9 µm

In order to separate out the fog pixels only, the scatter is also observed that for these values of BT the BTD is
plot between TIR brightness temperature and BTD having large positive value and appear as a separate
values of the image is studied. Figure 3 shows the scatter cluster. It is a remarkable feature (Marked as ellipse in
plot between TIR BT and BTD for a typical case having figure 3) which separates out the foggy pixels from the
fog, clear pixels, snow and cloud. From the scatter plot underlying land pixels. The information in this cluster is
it is seen that for brightness temperature less than 270° used to find out the minimum threshold value for BTD
K, the brightness temperature difference values are and the pixels having BTD value more than the
mostly negative or takes a very small positive value. minimum threshold value is classified as fog. The
This corresponds to the high level clouds and snow. ΔTBmin is determined automatically from the histogram,
of the BTD image. In a typical fog image, the histogram
of BTD image shows one primary maximum, a
secondary maximum and a minimum in between (figure
4a). In such a condition, BTD corresponding to the
BTD (BT 10.8 -BT 3.9 ) 0K

minima value can be considered as the ΔTBmin and

pixels havingΔTB more than ΔTBminis classified as fog.
For such type of scenes, the detection of fog is
completely objective as well as dynamic.

However, the MODIS data which has been used for this
analysis is in tile form and for each acquisition the frame
of the image changes and thus the proportion of
different features in the image also changes. Therefore,
for some cases in the histogram of BTD image of
BT 10.8 0K MODIS data, only single maximum is obtained (figure
Figure 3: Scatter plot of brightness temperature of 4b). In such cases the threshold for BTD (ΔTBmin) is set
10.8µm and brightness temperature difference (The to a point where the slope of the clear sky pixel drops
cluster marked within the ellipse indicates the considerably and ΔTB more than ΔTBminis classified as
presence of pixels with fog) fog. By applying this methodology fog region has been
identified with MODIS nighttime data from 01
However, for land, low clouds and fog regions where December 2012 to 31 January 2013 as well as for 01
the brightness temperature is more than 280°K, the December 2013 to 31 January 2014 and compared with
brightness temperature difference is positive. Due to MODIS fog image of IMD (India Meteorological
temperature inversion the brightness temperature of fog Department, www.imd.gov.in), whenever available as
is either similar or greater than that of the underlying well as with the early morning MODIS RGB data of
land. Thus the pixels with TIR brightness temperature next day. However, a quantitative validation of fog map
greater than 280 and less than 290° K, will correspond generated for January 2014 has been carried out with
to land pixels over laid by fog or low cloud pixels during visibility data at five different locations i.e. Delhi,
night (as observed from the figure 3). From the figure it Lucknow, Jaipur, Amritsar and Varanasi situated in
northern India.
 43

Journal of Geomatics Vol 10 No. 1April 2016

(a) (b)
Histogram of Difference Image Histogram of Difference Image
1 1
Normalized Histogram

Normalized Histogram
0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
-15 -10 -5 0 5 10 15 -10 -5 0 5 10

BTD=BT10.8µm -BT3.9µm BTD =BT10.8µm -BT3.9µm

Figure 4: Histogram of brightness temperature difference image (a) BTD image showing double peak (b) BTD
image showing single peak

4. Results and discussion shallow, dense and very dense depending on the change
in brightness temperature difference. The minimum
The evolution of fog over Indo-Gangetic plain in difference in brightness temperature ΔTBmin which is
December 2012 and January 2013, using MODIS obtained to detect fog varies from 3° to 8° K for the
Terra/AQUA data (Different colour shows the BTD period of analysis. For each image the threshold is found
values) is shown in figure 5. In order to show the to be different. This change in threshold is attributed to
evolution of fog to the maximum extend over the IG the change in fog optical depth and droplet size
plains, both the fog map generated using MODIS (Chaurasia et al., 2011). BTD value of 3° K corresponds
AQUA/TERRA has been shown in this figure. Care has to very thin fog. However, for BTD values between 4 to
been taken to incorporate maximum fog period to show 5° K corresponds to moderately dense fog. BTD value
the evolution. Therefore, few days with in December of 6 and >6 °K represents very thick fog. This gradation
2012 and January 2013 is missing in the figure 5. As is empirically made based on our earlier study
observed from the fog map generated by the above (Chaurasia et al., 2011) as well as from the visibility
method, the 2012-2013 fog periods started around early data.
December. On December 09, 2012, fog was initiated in
the eastern states like Bihar and gradually it extended The generated fog map has also been compared with the
spatially and covered Bihar and part of Uttar Pradesh by next day morning RGB image and fog map generated
December 15, 2012. The spatial extent continued to by IMD using MODIS data. The qualitative inter-
increase both from east to west as well as in the southern comparison of the two maps is shown in figures 6 and 7
direction starting from the foot hill of Himalaya. During indicating good agreement with each other. Bi-spectral
December 22-25, 2012 it covered the entire Indo- differencing technique has also been used in the
Gangetic plain starting from Bihar at the east to Punjab generation of fog map by IMD using MODIS data, and
at the west. This is the time when the wheat cultivation a pixel is classified as fog/low stratus when the BTD is
in the four major wheat growing states Bihar, UP, greater than 2.50 K (Product catalog, www.imd.gov.in ).
Haryana and Punjab are in full swing and the fields are This threshold is fixed for all images. In order to make
irrigated, which adds to the local moisture availability the two maps comparable instead of showing the
apart from the prevailing meteorological condition and variation of BTD over fog, only fog and no- fog regions
temperature inversion in this region which is conducive are generated. The blue colour in both the images shows
for fog formation. Similar situation continued till fog. It is observed from the figure 7 that, for December
January 07, 2013. 25, 2012, there is a systematic data loss in IMD fog map.
It is also observed that it has also picked up some of the
On January 10, 2013 part of north-western region of high cloud part in the southern part of the image, which
Uttar Pradesh was covered with fog which extended in is filtered by new developed dynamical thresholding
width till January 14, 2013. On January 19, 2013 there algorithm. Similarly, the cloud captured by IMD fog
was heavy fog condition over Punjab and parts of map on January 10, 2013 is also not captured by our
Haryana. On January 21, 2013 again Uttar Pradesh algorithm.
region was engulfed with fog which continued till
February 2013 (results shown till end of January). The A quantitative validation of the generated fog map has
gradient in the fog map shows the category of fog i.e. been carried against the in-situ visibility data over five
 44

Journal of Geomatics Vol 10 No. 1April 2016

different locations on north India i.e. Delhi, Jaipur, spectral behavior is similar.The brightness temperature
Varanasi, Lucknow and Amritsar. Table 1 shows the difference for low clouds is also very high like fog. The
error statistics of validation. It shows that the percent of algorithm is not able to detect fog when it is over laid
detection is 70%, the percent of false detection was 15% by high clouds due to advection. As this algorithm first
and percent of miss was 12.8%. The false detection is eliminates the high cloud region, the fog beneath
because of the presence of low cloud. For which the remains undetected.
horizontal visibility at surface level was high but

15 Dec 2012, 1700UTC 19 Dec 2012, 1640UTC 21 Dec 2012, 1625UTC 22 Dec 2012, 1710UTC

23 Dec 2012, 1750UTC 24 Dec 2012, 1655UTC 25 Dec 2012, 2020UTC 29 Dec 2012, 1715UTC

31 Dec 2012, 1715UTC 01 Jan 2013, 1745UTC 02 Jan 2013, 1650UTC 14 Jan 2013, 1715UTC

23 Jan 2013, 1710UTC 24 Jan 2013, 1615UTC 26 Jan 2013, 1600UTC 27 Jan 2013, 1645UTC

Figure 5: Evolution of fog over Indo-Gangetic plain in December 2012 and January 2013, using MODIS Terra/AQUA data
(Different colour shows the BTD values)
 45

Journal of Geomatics Vol 10 No. 1April 2016

07 Jan 2013, 2130 UTC 08 Jan 2013, 0515 UTC

30 Dec 2012, 2040 UTC 31 Dec 2012, 0605 UTC

28 Dec 2012, 0515 UTC

27 Dec 2012, 2010 UTC

(a) (b)
Figure 6: Inter-comparison of (a) generated fog map using MODIS Terra/AQUA of night time (Different colour shows the
BTDvalue) with (b) next day morning MODIS RGB image (recoloured)

25 Dec 2012, 2020UTC 10 Jan 2013, 2020UTC

13 Dec 2012, 1955UTC 15 Dec 2012, 1945UTC 20 Dec 2012, 1720UTC

(b)

13 Dec 2012, 1955UTC 15 Dec 2012, 1657UTC 20 Dec 2012, 1720UTC 25 Dec 2012, 2015UTC 10 Jan 2013, 2015UTC

Figure 7: Inter-comparison of generated fog map using (a) MODIS AQUA data with that of (b) IMD fog map using
MODIS data (recoloured)
 46

Journal of Geomatics Vol 10 No. 1April 2016

Dutta, H.N., B. Singh and A. Kaushik (2004). Role of

Table 1: Statistics of night time fog detection at five western disturbances in the development of fog over
locations (Delhi, Jaipur, Amritsar, Varanasi and Northern India. Presented in XIII National Space
Lucknow) in the IG plains from January 01-31, 2014 Science Symposium (NSSS), Koteyam.

Total no. of Observations 140 Ellord, G.P. (1995). Advances in the detection and
analysis of fog at night using GOES multi spectral
No. of non-concurrent data 25 infrared imagery. Weather Forecasting,10, 606-619.
Number of Hits 81
Eyre, J.R., J.L. Brownscombe and R.J. Allam (1984).
Number of Miss 16 Detection of fog at night using Advanced Very High
Number of false detection 18 Resolution Radiometer. Meteorological Magazine, 113,
266–271.
Percent of Detection (POD) 70%
Percent of False Detection 15% Greenwarld, T.J. and S.A. Christopher (2000). The
GOES I-M imagers: New tools for studying
Percent of Miss 12.8% microphysical properties of boundary layer stratiform
clouds. Bulletin of American Meteorological Society,
5. Conclusions 81(11), 2607-2619.

This technique yields a very good probability of night Gultepe, I., S.G. Cober, G. Pearson, J. A. Milbrandt, B.
time fog detection with acceptable false alarm Hansen, G.A. Isaac, S. Platnick, P. Taylor, M.Gordon
conditions. It has been observed that with the proposed and J.P., Oakley (2009). The Fog Remote Sensing and
dynamical thresholding, for 70% of the cases the Modelling (FRAM) field project and preliminary report.
generated fog map successfully detects fog using Bulletins of American Meteorological Society, 90(3),
satellite IR imagery during night time. In future, it may 341-359.
be required to develop algorithm to minimize mis-
classification between low clouds and fog. Hunt, G.E. (1973). Radiative properties of terrestrial
clouds at visible and infra-red thermal window
Acknowledgements wavelengths. Quarterly Journal of Royal
Meteorological Society, 99, 346-369.
The authors wish to thank the Director, Space
Applications Centre, for his keen interest in this work. Lee, T.F., F.J. Turk and K. Richardson (1997). Stratus
We are also thankful to www.ladsweb.nascom.nasa.gov and fog products using GOES-8-9 3.9µm data. Weather
for MODIS data and India Meteorological Department Forecasting, 2, 664-677.
for providing the visibility data.
Levy, W.F., L.A. Remer, S. Mattoo, E. Vermote and
References Y.J. Kaufman (2007). Second-generation operational
algorithm: Retrieval of aerosol properties over land
Bendix, J. (2002). A satellite-based climatology of fog from inversion of moderate resolution imaging
and low-level stratus in Germany and adjacent areas. specroradiometer spectral reflectance. Journal of
Atmospheric Research, 64, 3-18. Geophysical research, 112: D13211,
DOI:10.1029/2006JD007811.
Bendix, J. and M. Bachmann (1991). A method for
detection of fog using AVHRR-imagery of NOAA Ram Kirpa, M., M. Sarin, A. K.Sudheer and R.
satellites suitable for operational purposes. Rangaranjan (2012). Carbonaceous and secondary
MeteorologischeRundschau, 43, 178 (in German), inorganic aerosols during wintertime fog and haze over
1991. urban cities in Indo-Gangetic plain. Aerosol and Air
Quality Research, 12, 359-370.
Cermak, J. and J. Bendix (2007). Dynamical nighttime
fog/low stratus detection based on Meteosat SEVIRI Schueler, C.F. and W.F. Barnes (1998). Next generation
data – a feasibility study. Pure and Applied Geophysics, MODIS for polar operational environmental satellites.
64, 1179-1192. Journal of Atmospheric and Ocean Technology, 5, 430-
439.
Cermak, J. and J. Bendix (2008). A novel approach to
fog/low stratus detection using Meteosat 8 data. Singh, B., H.N. Dutta and A. Kaushik (2004).
Atmospheric research, 87, 279-292. Ecological restoration by fog water. Presented in XIII
National Space Science Symposium (NSSS) Koteyam.
Chaurasia, S., V. Sathiyamoorthy, B. Paul Shukla, B.
Simon, P.C. Joshi and P.K. Pal (2011). Night time fog Tiwari, S., S. Payra M. Mohan, S. Verma and D.S.
detection using MODIS data over Northern India. Bhisht (2011). Visibility degradation during foggy
Meteorological Applications, 8(4), 483–494.
 47

Journal of Geomatics Vol 10 No. 1April 2016

period due to anthropogenic urban aerosol at Delhi, Wetzel, M. A., R.D. Bory and L.E. Xu (2004). Satellite
India. Atmospheric Pollution Research, 2, 116-120. microphysical retrievals for land based fog with
validation by balloon profiling. Journal of Applied
Turner, J., R.J. Allam and D.R. Maine (1986). A case Meteorology, 35, 810-829.
study of the detection of fog at night using channel 3 and
4 on the Advanced Very High Resolution Radiometer Yang, L., Y. Xu and M. Wei (2008). Fog detection using
(AVHRR). Meteorological Magazine, 115, 285–290. MODIS data in the Yangtze river delta. Fifth
International Conference on Fuzzy Systems and
Underwood, S.J., G.P. Ellord and A.L. Kuhnert (2004). Knowledge Discovery, IEEE Computer Society, 457-
A multiple-case analysis of noctural radiation-fog 572, DOI:10.1109/FSKD.2008.302.
development off California utilizing GOES nighttime
fog product. Journal of Applied Meteorology, 43, 297-
311.
 48

Journal of Geomatics Vol 10 No. 1 April 2016

Morphometric changes of the Varuna river basin, Varanasi district, Uttar Pradesh
K. Prakash, S. Singh and U.K. Shukla
Centre of Advanced Study, Department of Geology
Banaras Hindu University, Varanasi-221005
Email: [email protected], [email protected], [email protected]

(Received: Jan 10, 2015; in final form: Dec 02, 2015)

Abstract: The Varuna is one of the important interfluves rivers joining the Ganga river near Varanasi city. The
morphometric analysis of the Varuna river basin has been carried out. It covers an area of about 3622 km2 of the Ganga
plain. The dendritic drainage pattern diagnosed in the area exhibits homogeneous permeable substratum and gentle
gradient. Spatio temporal changes (Land Use and Land Cover i.e. LULC) of the Varuna river basin, in Varanasi district,
using Landsat multispectral imageries spanning 38 years (1972, 1988, 2002 and 2010) are also studied. The LULC
patterns illustrate that in early 80’s the basin is largely covered by the salt affected waste land. Later on the salt affected
waste land area is reduced and area of agricultural land and built-up area has increased. The study underlines the necessity
of a scientific approach for the sustainable river basin management, with the help of the hydrological conditions, recent
climatic anomalies and geological control of the basin.

Keywords: Gangetic plain, Interfluves, Varuna river, Morphometric analysis, Confluence-shift

1. Introduction load, sediment-water ratio and slope of the basin as a

result of prevailing climate and neotectonics (Schumm
The interfluves (doab) surfaces of the Gangetic plain, et al., 2000; Raj, 2007). There are many small and large
forming a part of Indo-Gangetic fore land basin system rivers such as Ganga, Yamuna, Brahmputra, Kosi,
(Singh et al., 1996), are the most important, tens to Gomati rivers and several others have shown changes in
hundreds of kilometer wide geomorphic surfaces and their channel courses through Quaternary-Holocene
witness the oldest living civilization of the world. These times in response to changing set of climate and tectonic
plains are drained by numerous snow fed and alluvium conditions (Wells and Don, 1987; Srivastava and Singh,
fed rivers. Interfluve river basins undergo morphometric 1999; Kotoky et al.,2005; Roy and Sinha, 2005; Shukla
changes and transformation of channel patterns and et al., 2012, Shukla, 2013).
their degradation and aggradations in response to
varying set of climate and tectonics influencing the base The focus of the study, the Varuna river emanates at
level of the rivers through time (Denizman, 2003). 25°27′ N & 82°18′E near Mau Aima (Pratapgarh
According to Mesa (2006) geomorphic parameters are district, Uttar Pradesh) flows east-to-southeast for about
important and necessary to explore the basinal dynamics 183km, and makes confluence with the western bank of
and basin architecture. The present study is targeting on the Ganga river at the 83°2'40.088"E 25°19'46.387"N
quantitative approach of watershed and landscape near Raj Ghat bridge, just downstream of the Ganga in
development of the Varuna river, a tributary of Ganga Varanasi city (Fig.1). The Varanasi city is one of the
river, with the help of remote sensing and GIS data. The oldest living civilizations in the world and important
river is flowing deeply incised in to the interfluve pilgrim city of India with a population of about 1.5
surface having a rather narrow valley (Shukla, 2013). million. The name Varanasi itself is believed to have
Interfluve surface making the base level of the rivers is been derived after the name of the rivers Varuna and
made up of silt, sand and clay horizons. Varuna river Assi. Varuna river is one of the foremost controlling
basin is a part of the central alluvial plain of the Ganga drainage system of Varanasi city (Khan et al., 1988).
basin (Singh, 1996; Shukla and Raju, 2008) bounded by Assi river has deteriorated to become a drain, carrying
the Vindhyan rocks in the south forming the peripheral domestic waste and sewerage of Varanasi city, and calls
bulge. For watershed management practices and for immediate attention for its reclamation (Shukla,
geotectonic control over the drainage basin through 2013). The study also analyses spatio-temporal changes
morphometric analysis has been attempted by several in Land Use Land Cover (LULC) of the Varuna river
workers in the recent past (Sreedevi et al., 2004; Pati et basin based on available Landsat imageries. Moreover,
al., 2006; Pati et al. 2008; Thomas et al., 2010; Sarmah the current environmental status of a region and ongoing
et al., 2012). Channel characteristics of the interfluves modifications in terms of urban growth could be better
alluvial river replicate the stability of bank material and appreciated by analysis of spatial and temporal change
erosive power of the stream and any small changes in in land use planning (Turner et al., 1993).
geomorphology (or spatio temporal shift) in the river
basin signify a consequence of variation in sediment

© Indian Society of Geomatics

 49

Journal of Geomatics Vol 10 No. 1 April 2016

Figure 1: Location map of the Varuna basin (a) Map showing subdivision of Ganga plain in Uttar Pradesh
(modified after Singh, 1996); (b) Landsat imagery (MSS Path 153 and Row 42 of 1972) representing Varuna river
basin

Figure 2: Varuna river basin (a) DEM (digital elevation model) from SRTM data; (b) Drainage pattern and 4 sub
basins

2. Methodology parameters such as linear, areal and relief aspects were

extracted using ARC GIS-10.0. Measurements
Shuttel Radar Topogrpaphy Mission (SRTM) data pertaining to the confluence shifting of the Varuna river
(Fig.2a) (of Sep 30, 2000) and SOI (Survey of India) in time and space have been carried out using the
topographic maps (1:50000) are used for extraction of available published data, SOI toposheets, Landsat MSS
the drainage network and sub water shed in the study data and GPS readings.
area. Landsat Multispectral Scanning (MSS) data (of
Sep 30, 1978) is used for delineation of old valley of the 3. Results
Varuna river and further identified and traced out by
handheld 12-channel GPS (make Garmin E-Trax Vista). 3.1. Morphometry
The paleo channels are identified here on the basis of Morphometric analysis done in the present study
higher moister content in soils, textural characters on the incorporates quantitative study of the Varuna river
image and the vegetation pattern in the former river valley, altitude, volume, slope, profiles of the land and
valley. The satellite data were obtained from the Global drainage basin characteristics of the area concerned
Land Cover Facility (http://www.landcover.org). (Kondolf and Hervè, 2003).The stream network and the
Horton’s (1945) and Strahler (1965) methods were catchment area of the Varuna river with four 4th order
adopted in the present study for characterization of sub-watersheds with as are shown in figure 2b.
watershed and drainage network. The morphometric
 50

Journal of Geomatics Vol 10 No. 1 April 2016

Figure 3: (A) Two discreet southward shifting of the confluence of the Varuna river near Varanasi, traced out in
the field with hand GPS; (B) Traces (in doted lines) of the old Varuna River near Sarnath (after Jayaswal et al.);
(C) Present confluence of the Varuna river; (D) Old confluence of the Varuna River; (E) Abandoned channel of
the Varuna River; and (F)Low lying area showing abandoned channel of the Varuna river

Dendritic drainage pattern is the most common and The variations among the stream length ratio (Rl) values
widespread pattern found in the study area (Fig.2b). The are directly related to the geomorphology including
dendritic patterns evolved in the area closely resemble topography and lithology and hence it governs the
to the area having homogeneous bed materials (mainly erosional stage of the watershed and discharge
Gangetic alluvial) with a very gentle regional slope. The (Sreedevi et al., 2004). The Rl values in case of Varuna
perimeter of the Varuna watershed is 482.07 km (Table river basin varies between 0.66 and 0.75 (Table 1) and
1). The values of the bifurcation ratio (Rb) generally set that implies accomplishment of geomorphic maturity.
in between 2.0 and 5.0 for the drainage network. Such The Rho coefficient is used for determining the storage
network generally develops in consistent lithologies and capacity of drainage network (Horton, 1945). The
also signifies the minimum structural control over it. values of Rho coefficient of the Varuna river and its sub-
When the values are higher than 10.0, it indicates that watersheds vary from 0.14 to 0.26 (Table 1). The higher
structural control has played a dominant role on values of Rho coefficient of SW2 and SW3 (Fig.2b) are
drainage network development (Mekel, 1970). Low Rb indicators of higher hydrologic storage during floods
values indicate elongated shape of the basin (Morisawa, and decreased effects of erosion during elevated
1985). The Varuna river belongs to this category. discharge.
 51

Journal of Geomatics Vol 10 No. 1 April 2016

Drainage density (Dd) is one of the significant accounts an Lg value of 1.19, whereas the SW1, SW2,
indicators of the landform development and presents a SW3 and SW4 (Fig.2b) are having the Lg values
numerical measurement of landscape dissection and between 1.61 and 2.63 (Table 1). The higher value of Lg
hence the runoff potential (Smith, 1950). The study area of SW2 indicates geomorphic maturity while other sub-
has very low drainage density varying from 0.19- 0.31, watersheds are charecterized by early mature or late
and a very coarse drainage texture (Table 1), which youth stage of geomorphic advancement.
implies permeable subsurface conditions (since the
basin has chiefly clay and sandy clay subsurface The Rc (Circulatory ratio) value of Varuna Watershed
material) and dense vegetation. (VW) is 0.19, whereas it ranges between 0.15 and 0.23
in other sub-watersheds (Table 1). The low Rc values
Form factor (Ff) is significant factors to envisage the (<0.6) of the sub- watersheds are probably related to
shape of the drainage basin and the flow intensity of attenuated flood- discharge periods. The stage of
watershed with direct relationship to peak discharge evolution of the watersheds can also be explained by the
(Horton 1945, Gregory and Walling 1973). The value of numerical values of Rc of sub- watersheds. The low Rc
Ff (Table1) varies from 0.19 to 0.33. Low Ff values values of the sub- watersheds, imply youth stages of
(<0.4) are characterized by shorter flow peaks of longer watershed development.
duration which in case of Varuna river seems to be its
alluvium fed character and irregular rainfall in the area. The numerical value of Re (elongation ratio) for VW is
0.5, signifying an elongated nature of the basin. Re of 4
Length of overland flow (Lg) is an important sub-watersheds varies from 0.56 and 0.65(Table 1). The
morphometric parameter complementing both elongated shapes of sub- watersheds, with the larger
hydrologic and physiographic advancement of the basin area, are insisting the role of head-ward erosion in
watershed (Horton, 1945). The Varuna watershed development of lengthy channels.

Table 1: Mathematical formula adopted for the quantitative measurement and estimated values of the
morphometric parameter

Parameters/ Sub-Watershed SW1 SW2 SW3 SW4 VW

Linear parameters Unit
Area km2 694.25 1124.3 222.23 523.81 3622.5
Perimeter Km 237.54 302.46 109.41 149.93 482.07
Basin length (Lb) Km 52.27 66.51 25.67 42.53 135.57
Linear Aspect
Bifurcation ratio (Rb) = Nu/N(u+1) Dimensionless 4.03 3.2 2.81 5.33 3.84
Where Nu is number of any given order and N(u+1)
is number in the next higher order
Stream length ratio (Rl) = Lu/L(u-1) Where Lu is Dimensionless 0.66 0.67 0.74 0.75 0.68
stream length order u and L(u-1) is stream segment
of the next lower order
Rho coefficient (ρ) = Rl/Rb Dimensionless 0.16 0.2 0.26 0.14 0.17
Areal Aspect
Drainage density (Dd) = Lt/A km-1 0.29 0.19 0.28 0.31 0.26
Where Lt is the total length of all ordered streams
Stream frequency (Fs) = Nt/A km-2 0.05 0.08 0.1 0.05 0.06
Where Nt is the total number of all ordered streams
Drainage Texture (T) = Dd×FS km-3 0.014 0.015 0.028 0.015 0.015
Length of overland flow (Lg) = 1/2Dd km 1.72 2.63 1.78 1.61 1.92
2
Form factor (Ff) = A/Lb Dimensionless 0.25 0.25 0.33 0.28 0.19
2
Circulatory ratio (Rc) = 4πA/P Dimensionless 0.15 0.15 0.23 0.29 0.19
Elongation ratio (Re) = Dimensionless 0.56 0.56 0.65 0.61 0.5

(SW- sub-watershed, VW- Varuna watershed)

 52

Journal of Geomatics Vol 10 No. 1 April 2016

3.2. Confluence shift basin were investigated during four time spans of 1972,
The confluence shift of the Varuna river is ascertained 1988, 2002 and 2010 (Fig. 4).
by analysing the satellite imageries, toposheets (Survey
of India) and supported by extensive field effort (Fig. 3).
The shifting of the confluence points of the Varuna river
is traced out by hand held 12 channels Garmin E-Trax
Vista GPS with ± 3 meter accuracy in the field (fig. 3a).
The confluence point of the Varuna river has shifted
progressively southwards in detached steps by leaving
behind older valley signatures like depression with
meandering morphology, ponds and meander cut-offs
away from the main river channel (Fig. 3b). Two
discreet channel shifts have been noticed in the field.
First shift is 1.64 km and the second shift is 0.96 km
located north from the present day confluence of Varuna
with the Ganga river (Fig.3a). The remnant valley of the Figure 4: Land use and Land cover (LULC) map of
Varuna river is identified on the basis of width of the the Varuna basin during four time spans of (A) 1972;
abandoned channel/ valley in the field (Fig.3e-f). (B) 1988; (C) 2002; and (D) 2010

It is to be noted that all the tributaries near the Varanasi

city along with the Ganga River are deeply incised and
at present do not have liberty to leave their valleys and
shift (Shukla et al., 2012). Therefore, there are mainly
three possibilities for shift of the tributary confluences.
The causes are (i) when they were not incised and were
freely shifting their channels (ii) in the recent past due
to channel piracy by the Ganga river and (iii) periodic
and linear change in rainfall.

However, the shift in confluence points of the Varuna

river is well comprehended by the dynamic of the main
Ganga River channel. Ganga river acts as the base level
control for the Varuna river and any change in the
dynamics of the former directly affects the later. Before
the Last Glacial Maxima most likely in the late
Quaternary, rivers were not incised and freely migrating
within their broad valleys, shifting laterally for many
kilometres (Shukla et al., 2012). The meanders of the
present-day Ganga river in the study area around
Varanasi appear deformed and elongated, with straight
and highly-convoluted reaches, indicating tectonic
control over the present-day river channel. Around
Varanasi, where presently due to incision (Swamee et
al., 2003; Shukla, 2013) the Ganga River is confined
within a 1–2 km wide narrow valley, it previously had a
much wider valley which was 10–15 km wide. With
time it migrated eastward, leaving behind a large
abandoned meander belt. The location of Varanasi
represents the old flood plain of the Ganga River,
characterized by numerous ponds formed due to the Figure 5:(A) Bar digram showing area and Landuse
abandonment of the channels (Srivastava et al., 2003; land cover changes in Varuna basin from 1972-2010;
Shukla et al., 2012). The incision of the Ganga River
(B) Scatter diagram with smooth line showing
seems related to tectonics along a fault line passing in a
NE–SW direction, controlling the course of the river change in wetland area with time; (C) Rainfall
and dynamics of tributary confluence (Shukla and Raju, demonstrating anamolous rain fall in early 80’s
2008; Shukla et al., 2012; Singh, 2015).
In 1972, LULC of the Varuna river basin reveals that,
3.3. Land use pattern 32% area is identified as salt affected waste land and
The Landsat data has been classified into six major about 60% area is used in agricultural practices (Fig.5a).
classes (agriculture, river, water bodies, salt affected The salt affected waste land area is reduced by 22% in
wasteland, wetland and urban- rural build-up area) and 1988 from 1972 and the large proportion of the
in view of that the land use changes in the Varuna river reclaimed land (from salt affected to useful land) is
occupied by agricultural practices and some proportion
 53

Journal of Geomatics Vol 10 No. 1 April 2016

is consumed in built-up areas (Fig.5a). The LULC management and evolution of groundwater potential for
scenario has drastically changed from 1988 to 2010 watershed planning and their management.
(Fig.4). In these years percentage of built up areas have
increased exponentially, leaving behind the area of Near Varanasi, the Ganga river is flowing along a NE–
agricultural land and surface water bodies. The area of SW tectonic lineament. The confluence of the Varuna
wetland is somehow constant with modest variation river is shifted towards south from 1 to more than 1.5
while streams and water bodies show large variation in km from north from its present confluence with the
studied time span, due to anomalous rain fall in early Ganga river. The study suggests that the
80’s (Fig. 5c). morphodynamic change and confluence shift in the
interfluves river (Varuna river), was controlled largely
Table 2: Land use land cover status in Varuna basin by monsoonal variability as well as directed by tectonic
during 1972-2010 activity in trunk river. LULC has revealed that because
of population growth build up area has drastically
Area (km2) 1972 1988 2002 2010 increased at the cost of agricultural land.

Agricultural 2192.15 2688.05 2869.40 2961.74 Acknowledgement

land
Salt 1155.46 413.36 325.89 195.64 Authors are thankful to the Head of the CAS,
affected Department of Geology, Banaras Hindu University for
wasteland making available the functioning conveniences. S.
Wetland 237.15 327.44 276.62 317.90 Singh is grateful to the UGC, New Delhi for a
fellowship in the form of BSR-SRF.
Built-up 28.04 32.59 57.11 78.84
References
area
Streams and 9.04 160.16 92.57 67.46 Denizman, C. (2003). Morphometric and spatial
water distribution parameter of karstic depressions, lower
bodies Suwannee river basin, Florida. Journal of Cave and
Total area 3621.83 3621.60 3621.58 3621.58 Karst Studies 65(1), pp. 29-35.

Gregory, K.J. and D.E. Walling (1973). Drainage basin

Rainfall data (Fig. 5c) from the India Meteorological form and process - a geomorphological approach.
Department (IMD) support the change in the surface London: Edward Arnold.
water pattern in early 1980. The present study is
confirming the important physiographic modification Horton, R.E. (1945). Erosional development of streams
occurring in the Varuna River basin during the last four and their drainage basins hydrophysical approach to
decades. The morphometric analysis, confluence quantitative morphology. Geological Society of
shifting and change in land use with/ land cover pattern America Bulletine 56 (3), pp. 275-370.
of the river basin highlight the necessity for sustainable
development and management of ecological setup for Jayaswal, V., U.K. Shukla and M. Sharma (2015).
the Varuna river basin. Varanasi and Ganga (Report of Geo-morphological &
Archaeological Investigations, at Ramanagar). Aryan
4. Conclusion Books International, New Delhi.

The GIS techniques have been efficiently utilized for Khan, A.A., P.C. Nawani and M.C. Srivastava (1988).
the assessment of the drainage characteristics of the Geomorphological evalution of the area around
Varuna river watershed and to comprehend the Varanasi, Uttar Pradesh, with aid of aerial photography
significance of morphometric studies and shifting of the and Landsat imageries. Geological Survey of India,
confluence. The low bifurcation ratio and elongation Rec. 113(8), pp. 31-39.
ratio of the Varuna river signify the elongated shape of
the basin. The elongated shapes of sub- watersheds, Kondolf, G.M. and P. Hervè (2003). Tools in fluvial
with the larger basin area, are due to head-ward erosion geomorphology. John Wiley & Sons Ltd., England.
leading to expansion of channels. The study area is
demonstrating very low drainage density because of Kotoky, P., D. Bezbaruah, J. Baruah and J.N. Sarma
permeable subsurface and dense vegetation. The low (2005). Nature of bank erosion along the Brahmaputra
drainage density in the study area leads to very coarse river channel, Assam, India. Current Science 88 (4), pp.
drainage texture. The low values of circulatory ratio are 634-640.
characterized by attenuated flood- discharge periods
and youth stages of watershed development. Thus the Mekel, J.F.M. (1970). The use of aerial photographs in
incorporation of morphometric analysis together with geological mapping. ITC text book of photo-
predictable watershed assessment methods would have interpretations 8, pp. 1-169.
a better understanding the status of land form, drainage
 54

Journal of Geomatics Vol 10 No. 1 April 2016

Mesa, L.M. (2006). Morphometric analysis of a Myths and Scientific Studies. Aryan Book International,
subtropical Andean basin (Tucuman, Argentina). New Delhi, pp. 100-113.
Environmental Geology 50, pp. 1235-1242.
Singh, I.B. (1996). Geological evolution of Ganga plain
Morisawa, M. (1985). Geomorphology text books: - An overview. Journal of Palaeontological Society of
Rivers, form and processes, chapter 5 “Structural and India 41, pp. 99–137.
lithological control”.
Singh, C.K. (2015). Middle Ganga Plain; May be on the
Pati, J.K., V.P. Malviya and K. Prakash (2006). Verge of Seismic Shock. Journal Geological Society of
Basement re-acvtivation and its relation to Neotectonic India 85, pp. 511-513.
activity in and around Allahabad, Ganga Plain. Journal
of Indian Society of Remote Sensing 34, pp. 47-56. Smith, K.G. (1950). Standards for grading texture of
erosional topography. American journal of Science 248,
Pati, J.K., J. Lal, K. Prakash and R. Bhusan (2008). pp.655-668.
Spatio- temporal shift of Western bank of the Ganga
river, Allahabad city and its implications. Journal of Sreedevi, P.D., K. Subrahmanyam and S. Ahmed
Indian Society of Remote Sensing 36, pp. 289-297. (2004). The significance of morphometric analysis for
obtaining groundwater potential zones in a structurally
Raj, R. (2007). Strike slip faulting inferred from controlled terrain. Environmental Geology 47, pp. 412-
offsetting of drainages: lower Narmada basin, western 420.
India. Journal of Earth System Sciences 116, pp. 413–
421. Srivastava, P., I.B. Singh, S. Sharma, U.K. Shukla and
A.K. Singhvi (2003). Late Pleistocene–Holocene
Roy, N.G. and R. Sinha (2005). Alluvial hydraulic changes in the interfluve areas of the central
geomorphology and confluences dynamics in the Ganga Plain, India. Geomorphology 54, pp. 279–292.
Gangetic plains, Farrukhabad–Kannauj area, Uttar
Pradesh, India. Current Science v. 88, pp. 2000–2006. Srivastava, A. and R.P. Singh (1999). Surface
manifestation over a subsurface ridge. International
Sarmah, K., L.K. Jha and B.K. Tiwari (2012). Journal of Remote Sensing 20, pp. 3461– 3466.
Morphomertic analysis of a highland microwatershed in
East Khasi Hills District of Meghalaya, India: Using Strahler, A.N. (1964). Quantitative geomorphology of
remote sensing and geographic information system basin and channel networks. Hand book of Applied
(GIS) techniques. Journal of Geology and Regional Hydrology (Ed. Ven Te Chow). McgrowHill book
Planning 5(5), pp. 142-150. company, New York, section 4-11.

Schumm, S.A., J.F. Dumont and J.M. Holbrook (2000). Swamee, P.K., B. Parkash, J.V. Thomas and S. Singh
Active tectonics and alluvial rivers. Cambridge (2003). Changes in channel pattern of river Ganga
University Press, Cambridge. between Mustafabad and Rajmahal, Gangetic plains
since 18th century. International Journal of Sediment
Shukla, U.K., I.B. Singh, M. Sharma and S. Sharma Research 18(3), pp. 219-231.
(2001). A model of alluvial megafan sedimentation:
Ganga Megafan. Sedimentary Geology 144, pp. 243– Thomas, J., S. Joseph and K.P. Thrivikramaji (2010).
262. Morphometric aspects of a small tropical mountain river
system, the southern Western Ghats, India. International
Shukla, U.K. and N.J. Raju (2008). Migration of the Journal of Digital Earth 3(2), pp. 135-156.
Ganga river and its implications on hydrogeological
potential of Varanasi area, U.P. India. Journal of Earth Turner, M.G., W.H. Romme, R.H. Gardner, R.V.O’
System Science 117, pp. 489–498. Neill and T.K. Kratz (1993). A revised concept of
landscape equilibrium: Disturbance and stability on
Shukla, U.K., P. Srivastava and I.B. Singh (2012). scaled landscapes. Land Ecology 8(3), pp. 213- 227.
Migration of the Ganga river and development of cliffs
in the Varanasi region, India during the late Quaternary: Wells, N.A. and J.A. Don (1987). A reconnaissance of
Role of active tectonics. Geomorphology 171–172, pp. sedimentation on Kosi alluvial fan of India. Special
101–113. publication of SEPM, Tulsa 39, pp. 51-61.

Shukla, U.K. (2013). Varanasi and the Ganga river: A

Geological Perspective. In: V. Jayaswal (Eds) Varanasi:
 55

Journal of Geomatics Vol 10 No. 1 April 2016

Malaria-risk assessment using geographical information system and remote sensing in
Mecha district, West Gojjam, Ethiopia
Emebet Dessalegne1, K.V. Suryabhagavan1 and M. Balakrishnan2
1
School of Earth Sciences, Addis Ababa University, P.O. Box 1176, Addis Ababa, Ethiopia
2
Department of Zoological Sciences, Addis Ababa University, P.O. Box 1176, Addis Ababa, Ethiopia
Email: [email protected]

(Received: Dec 25, 2015; in final form: Feb 24, 2016)

Abstract: Malaria is a major global health problem. About 3.5 billion people are at risk of infection of malaria worldwide,
with environmental factors contributing for about 70–90% of the disease-risk. Over one million cases of malaria are
reported each year, out of which more than 80% is from the sub-Saharan Africa. Ethiopia is a predominantly malaria-
prone country as about 75% of the landscape of the country is favorable for breeding of the malaria vector. Geographical
Information System has emerged as a spatial technology, which integrates a wide range of datasets available from
different sources including remote sensing and Global Positioning System. During the present study, a malaria-risk map
of Mecha district of Ethiopia was prepared by establishing the relationship of various climatic and non-climatic factors
related to the disease using regression analysis. Weighted overlay technique of multi-criteria evaluation was used to
develop the malaria-risk map. Temperature, rainfall, altitude, distance from streams, distance from swamps and ponds,
population density, health facilities and land-use/land-cover patterns were used to prepare malaria-risk areas. Malaria
hazard, elements of risk and vulnerability layer were overlaid, and further verified by ground truth and village-wise
reports of malaria cases to produce the final malaria-risk map. Four categories of malaria-risk ranging from very high to
low were derived. Most of the study area (99.01%) was found to belong to high and moderate malaria-risk. It is suggested
that effective identification and mapping of malaria-risk levels can be made using geospatial tools, to contribute for the
prevention and control of this disease.

Keywords: GIS, Landsat, Malaria, Remote sensing, Regression analysis, Weighted overlay

1. Introduction risk of infection of malaria worldwide with

environmental factors contributing for the disease
Malaria is a major health problem as it affects all age transmission, and its seasonal patterns.
groups of the people in most parts of the world even for
about 70–90% of the disease risk (Bautista et al., 2006; There are several factors associated with this disease
Erin et al., 2014). Around 300−500 million cases and and its control, such as water bodies, rainfall,
more than two million deaths of malaria are reported temperature, population, land-use/land-cover and health
each year, with more than 80% of these from the sub- facilities (Palaniyandi, 2012). Understanding the causal
Saharan Africa (Abdulhakim, 2013). Malaria is factors is a prerequisite to design and implement
essentially an environmental disease, as the vectors appropriate malaria-risk management. So as to mitigate
require specific habitats with surface water for the effect of this risk, effective malaria-risk
reproduction and humidity for adult mosquitoes to management methods are required. Spatial information
survive. The development rate of both the vector and the on malaria distribution helps to prioritize control
malaria parasite are influenced by temperature measures.
(Ashenafi, 2003). Approximately 4–5 million cases of
malaria are reported annually in Ethiopia and malaria is 2. The study area and methods
prevalent in 75% of the extent of the country, putting
over 50 million people at risk (Abdulhakim, 2013). Mecha district lies within 11o 8'−11o39' N latitude and
36o59' 51"−37o 20' E longitude covering a total area of
Integrated use of remote sensing (RS) and Geographical 149,119 km2 (Figure 1), located in the West Gojjam
Information System (GIS) has been successfully Zone in the Amhara region, about 35 km from Bahir
demonstrated in many studies related to mapping of Dar, the capital town of Amhara Regional State of
malaria-risk in different parts of Africa (Hay et al., Ethiopia. There are 44 villages including three town
2000; Kleinschmidt et al., 2001; Sithiprasasna et al., administrative villages in the study area. Mecha district
2005; Dongus et al., 2007). The severity of malaria is a is situated at an altitude range of 1720 m−2800 m above
function of the interactions between Plasmodium, the sea level. The area is characterized by flat lying
parasite; the Anopheles mosquito, the vector; the human topography with some hilly terrain. This district has
host and the environment. Vector abundance combined different climatic variables in different seasons. The
with the probability of the vector feeding of susceptible annual rainfall pattern of the study area varies from
human-host determines the risk of malaria infection, 1000 to 2000 mm. The temperature varies from 23oC to
which is more prevalent in the tropics. It is a serious 27oC. June, July and August are high rainfall months
vector-borne disease. About 3.5 billion people are at and December, January and February are low rainfall

© Indian Society of Geomatics

 56

Journal of Geomatics Vol 10 No. 1 April 2016

months. High temperature is recorded in March, April dependent variable was malaria cases and the
and May and low temperature is recorded in November, independent variables were temperature, rainfall,
December and January. LU/LC, population density, slope, elevation, distance
from swamps and ponds and distance from streams.
2.1 Methodology Elevation variable (8.7) and slope variable (10.7) in the
To develop the malaria-risk map of the study area, regression model are associated with large VIF values.
identification and selection of the major factors To show the spatial clustering of the values associated
contributing for malaria breeding such as land-use/land- with the geographic features in the study area, Moran’s
cover, water bodies, population, elevation, temperature, I. was also computed.
rainfall, ponds and swamps, slope and health station
facilities were done. Regression analysis was applied to 2.1.1 Moran's I: This is a tool to measure spatial
identify the statistical correlations between malaria autocorrelation based on both feature locations and
cases and the above parameters. The mathematical feature values simultaneously. It evaluates whether the
formula applied to the explanatory variables in order to pattern expressed is clustered, dispersed or random
best predict the dependent variable was the following: (Oliveira et al., 2013). The Moran’s index statistic for
spatial autocorrelation is given as:
𝑌 = 𝛽0 + 𝛽1𝑥1 + 𝛽2𝑥2 … . + 𝛽𝑛𝑥𝑛 + 𝜀 (1)
𝒏 𝒏
𝒏 ∑і=𝟏 ∑ј=𝟏 𝒘ij𝒁і𝒁ј
𝐼= ∑𝒏
(2)
where, yis dependent variable, x1, x2, …, xn are 𝑺𝒐 і=𝟏 𝒁і²
independent variables, β0, β1, β2… βn are coefficients
and ε is error term (residual). where, Zі = the deviation of an attribute for feature і
from its mean
In this study, global weighted regression (GWR) and Wij= the spatial weight between і and j,
ordinary least squares regression (OLS) were carried out n= total number of features and
to assess the spatial relationship between the parameters So= the aggregates of all the spatial weights.
and malaria cases and to validate the model
performance. To assess spatial model performance, 2.1.2 Image processing: To produce land-use/land-
values of R-squared, adjusted R-squared, Jarque-Bera cover map of the study area, Landsat TM image of
p-value (JB), Akaike Information Criterion (AIC) and path 170 and row 052 of January 2015 was acquired.
Variance Inflation Factor (VIF) were computed (Ehlkes
et al., 2014). Independent variables greater than 7.5 VIF
(strong multi-collinearity) were cut off to overlay. The

Figure 1: Location map of the study area Mecha District, Ethiopia

 57

Journal of Geomatics Vol 10 No. 1 April 2016

Field data Population

Landsat TM 2015

Image processing Training Health Population

samples station density

Supervised Distance from HS Reclassify

classification

Accuracy Weighted overlay

assessment

LU/LC map Reclassify

Vulnerability map
Malaria-hazard map Element at risk map

AHP

Weighted overlay

Malaria-Risk map

Figure 2: Malaria-risk flowchart

Image processing starting from image pre-processing environmental conditions for malaria transmission
(geometric and radiometric correction), layer stacking based on climatic and non-climatic factors. All factor
(band 2–7), image enhancement and image parameters compatible to hazard analysis were
classification to the final accuracy assessment were generated before weighted overlay. Hazard map was
done in ERDAS Imagine software. To correct the produced by computing and reclassifying the five
satellite image radiometrically, Operational Land parameters viz. meteorological (rainfall and
Imager (OLI) band data were converted to top of temperature) data, distance from ponds and swamps,
atmosphere (TOA) to a planetary reflectance using altitude and distance from streams layers. Each of the
reflectance rescaling coefficients provided in the hazard parameters was ranked according to the
product metadata file. The following equation was used importance for mosquito breeding and transmission.
to convert DN values to TOA reflectance: The process of weighting each factor was performed in
IDRISI software. After assigning weight, the hazard
𝜌𝜆′ = 𝑀𝜌𝑄𝑐𝑎𝑙 + 𝐴𝜌 (3) map was computed by overlaying the five selected
factors.
The ρλ' does not contain the sun angle correction and
hence the image was again converted to TOA Vulnerability map was generated from distance from
reflectance with a correction for the sun angle as health facility map and population density map. The two
follows: layers were overlaid with 54% weight to population
density map and 46% to health facility map. The weight
𝜌𝜆′ 𝜌𝜆′ was given by consulting health experts, who have
𝜌𝜆 = = (4)
𝑐𝑜𝑠(𝜃𝑠𝑧) 𝑠𝑖𝑛(𝜃Ѕ𝐸) advanced knowledge about malaria based on the
regression result coefficients and available information.
After preparing the factors and validating the model and The element at the risk-map was computed by
image processing, three malaria-risk layers were reclassifying the land-use/land-cover pattern of the
generated using their factors. The malaria-risk map was study area. The land-use/land-cover types were ranked
developed by combining the suitability of
 58

Journal of Geomatics Vol 10 No. 1 April 2016

based on the importance from the most important to malaria cases, and in August when higher number of
least important, and vulnerability map was developed cases were registered. Although temperature favours
by reclassifying land-use/land-cover types of the study Plasmodium development, lack of water prevents
area. In this study, malaria-risk was expressed as the breeding and development of the vector. Figure 4 shows
product of malaria hazard map, vulnerability map and the relationship between monthly malaria incidence
element at risk-map using Shook model. To produce the recorded during 2002−2012 with the data on monthly
malaria-risk map, the influence factors were assigned temperature variations in the study area.
for the three components of malaria-risk layers (malaria
hazard, element at risk and vulnerability layer) and
overlaid.

R=V×H×E (5)

where; R=Malaria risk map; H= malaria hazard map;

V= vulnerability map; E= malaria element risk map.

The methodological flowchart is presented in Figure 2.

3. Results

3.1 Malaria vs rainfall

Figure 3 shows the average rainfall and average malaria
cases recorded in the study area during 2002–2012.
Rainfall was the main climatic factor for the prevalence Figure 3: Malaria vs rainfall relationship in the
of malaria in the study area with 44% influence (Tables study area during 2002−2012
1 and 2). The average rainfall of the study area varies
from the lowest 2.056 mm to the highest 418.97 mm per
month. Maximum rainfall was recorded during
June−August and the minimum during
December−February. A higher number of malaria cases
was recorded during May−June and
October−November, and lower in the months of August,
March and April. There was a positive relationship
between malaria cases and rainfall in the months
December to February and May to July, but the
relationship was negative in the months of August to
November.

3.2 Malaria vs temperature

The present study has revealed that temperature also has
influence in prevailing malaria. There was a negative Figure 4: Malaria vs temperature relationship in the
relationship during the months of December−April, study area (2002−2012)
when the study area had lower number of recorded

Table 1: Malaria cases and rainfall and temperature in Mecha district (2002–2012)

Months July Aug. Sep. Oct. Nov. Dec. Jan. Feb. Mar. Apr. May June
Malaria case 39164 2770 3638 4760 5705 4487 3313 3162 2909 3070 7188 7683
Rainfall 418.9 388.3 238.2 100.3 35.4 10.4 2.9 2.06 29.7 60.9 175.6 358.5
Temperature 20.07 20.5 20.78 19.72 15.52 17.45 18.94 19.89 22.5 23.78 22.76 20.69

3.3 Regression analysis for model validation and adjusted R2 for this model were 657.72329, 0.8052
As shown in Table 2, rainfall, temperature and and 0.7566, respectively. Multiple R-squared and
population density have strong positive relationship and adjusted R-squared were both statistics derived from the
altitude and slope have strong negative relationship with regression equation to quantify model performance. In
malaria incidence. These are the main factors for this model, R2 was 80.5299%. Hence, 80.53% variation
malaria prevalence in the study area relative to other in the dependent variable (malaria cases) could be
factors analysed (Tables 2 and 3). The AIC, multiple R2 explained by the model.
 59

Journal of Geomatics Vol 10 No. 1 April 2016

Table 2: Result of regression analysis
Variable Coefficient T-stat Probability Robust-t Robust-p R2 (%) VIF
Intercept -75035357 -4.74 0.000041* -7.994938 0.000000*
Pop. density +69.1356 4.037 0.000319* 6.851215 0.000000* 85.54 2.0
Health station Distance -0.031 -4.051 0.007826* -1.130710 0.000000* 19.6 1.7
Slope -11.04497 -0.12 0.897635 -0.191535 0.849318 16.68 10.7
Temperature +115.4472 0.52 0.526501 0.615132 0.542817 51.04 2.4
Rainfall +752.86 4.76 0.000039* 7.638637 0.000000* 64.02 1.9
Elevation -114.44 -2.06 0.488905 -0.848799 0.402299 17.04 8.7
Distance from swamps -0.04183 -4.85 0.009351* -0.746534 0.000014* 59.00 2.2
Distance from streams -0.08 -0.58 0.56 -0.77 0.44 12.71 1.4
* indicates a statistically significant p-value (p < 0.01).
index value was 0.357501, and hence spatial features
Coefficients are values that represent the strength and and their associated data values tended to be clustered
type of relationship the explanatory variable has to the (positive spatial autocorrelation) as it was greater than
dependent variable. When the relationship is positive, 0. The tool returns a Z score of 3.429816, which
the sign for the associated coefficient is also positive (+) indicated that standard deviations were away from the
and negative relationships have negative (−) signs. mean.
When the relationship is strong, the coefficient is large
such as rainfall, temperature, altitude, elevation and
population density. Weak relationships are associated
with coefficients near (zero) such as distance from
swamps and ponds and distance from streams.
Regression analysis computed a p-value for the
coefficients associated with each independent variable.
P-value was used to reject the null hypothesis for
statistical test that states for all purposes, the coefficient
is zero and the associated explanatory variable is not
useful for the model.

The regression analysis revealed small p-values. Hence,

the explanatory variables were important to the model
with a value different from zero (the coefficient was not
zero). Table 3 shows that some of the variables are both
negative and positive. The explanatory variable that
showed negative and positive significance were slope,
distance from streams, distance from swamps and
temperature. This indicates that these variables were not Figure 5: Malaria hazard map of the study area
statistically significant (P>0.01). However, population
3.4 Areas of malaria hazard
density, distance from health stations, rainfall and
Rainfall, altitude, streams, temperature, swamps and
distance from swamps and ponds were significant
(P<0.01). ponds were the predictors of presence of malaria with
percentage influence of 44%, 38%, 4%, 7% and 7%,
respectively. Rainfall and altitude were the dominant
Table 3: Results of explanatory regression analysis
factors for the existence of malaria as a hazard as
Variables Variable significance
compared with other selected factors, and streams
(%)
showed the least percentage influence (Principal
Negative Positive
Eigenvector) for malaria prevalence. The consistency
Population density 0.00 100 ratio for the Eigenvector of weights was within an
Slope 75.44 24.56 acceptable range with the value 0.02. Table 4 shows the
Altitude 85.44 14.56 weight, rank and degree of vulnerability of the selected
Distance health stations 100 0.00 parameters of malaria hazard in the study area. Figure 5
Distance from streams 60.18 39.82 shows the malaria hazard-risk map of the study area,
Distance from swamps & 100 0.00 which shows the level of malaria vulnerability in an
ponds extent of 46.77 km (0.31%) as very high, 64504.51 km2
Rainfall 0.00 100 (43.25%) high, 76446.74 km2 (51.26%) moderate and
Temperature 21.58 78.42 8122.58 km2 (5.44%) low. Thus, most of the study area
is subjected to high and moderate malaria hazard-risk.
In the present model, p-value was small (0.000015), and
hence the null hypothesis was rejected. There was 3.5 Malaria vulnerability
spatial autocorrelation between the values associated Figure 6 shows that in an extent of 1878.04 km2 (1.26%)
with the geographic features in the study area. Moran’s was moderately vulnerable for malaria 39794.21 km2
 60

Journal of Geomatics Vol 10 No. 1 April 2016

(26.68%) was vulnerable for malaria at low level and Table 4: Characteristic of factors in relation to
107446.73 km2 (72%) was vulnerable for malaria at malaria hazard area identification
very low level. Thus, the majority of the study area is Factors Weight Class Rank Degree of
under very low malaria-risk. vulnerability
Rainfall 44 98–100 mm 5 Very low
100–103 mm 4 Low
103–105 mm 3 Moderate
105–108 mm 2 High
>108 mm 1 Very high
Altitude 38 <2000 m 1 Very high
2000–2200 m 2 High
2200–2400 m 3 Moderate
2400–2600 m 4 Low
>2600 m 5 Very low
Distance 7 0–500 m 1 Very high
from 500–2000 m 2 High
Swamps 2000–3500 m 3 Moderate
and 3500–5000 m 4 Low
ponds >5000 m 5 Very low
Distance 4 0–500 m 1 Very high
from 500–2000 m 2 High
Streams 2000–3500 m 3 Moderate
3500–5000 m 4 Low
>5000 m 5 Very low
Figure 6: Malaria vulnerability
Temperat 7 <15oC 5 Very low
3.6 Element at risk map of malaria ure 15–17oC 4 Low
The results of NDVI values ranged between -0.266708– 17–19oC 3 Moderate
0.569553. The lowest (negative) NDVI values indicate 19–21oC 2 High
the water bodies. The highest NDVI values indicate >21oC 1 Very high
plantation and bush lands. Table 5 shows the results of
NDVI value for each of the LU/LC types of the area. Table 5: NDVI values for each land-use/land-cover
Figure 7 illustrates that 48523.66 km2 (32.54%) had types
very high, 21365.29 km2 (14.32%) had high, 75556.48 Land-use/land-cover NDVI
km2 (50.66%) had moderate and 3673.56 km2 (2.46%) types values
had low level of malaria vulnerability. Irrigation 0.143
Water bodies -0.075
3.7 Identifying malaria-risk areas Farmland 0.144
As shown in Table 6, element at risk and malaria Plantation 0.245
vulnerability had 63%, 31% and 6% of weight influence
Grassland 0.178
of hazard element at risk and vulnerability for the
existence of malaria in the study area. Malaria hazard Bareland 0.148
layer was the dominant factor for the final malaria-risk Wetlands 0.103
map. There was no area, which was free from malaria- Settlement 0.114
risk. In the study district, 33.59 km2 (0.23%) area was Bush and shrublands 0.188
of very high, 69305.82 km2 (46.47%) was high,
76830.96 km2 (51.52%) was moderate and 2948.61 km2 3.8 Malaria risk levels of villages in Mecha district
(1.97%) was of low malaria-risk. The majority of the All villages in the study area fall within the risk of
study area was subject to high and moderate risk of malaria prevalence. One of the villages (Merawi town)
malaria. The final malaria-risk model map has revealed fall in the very high malaria-risk area and the other
that all parameters, analysed during the study had villages are in the high to low levels of malaria-risk.
different weight influence for the prevalence of malaria Figure 8 shows the villages in the study area showing
in the Mecha district of Ethiopia. However, rainfall was the levels of malaria-risk.
the most dominant factor for the prevalence of malaria
in the present study area, where as altitude had limiting
effect for the prevalence of the disease as altitude has
negative correlation with temperature. The total area
and degree of vulnerability for malaria prevalence in the
final malaria-risk map and its layers (malaria hazard
map, element at risk map and vulnerability map) are
presented in Table 7.
 61

Journal of Geomatics Vol 10 No. 1 April 2016

Figure 7: Element at risk map of malaria

Table 6: Characteristic of factors in relation to

malaria risk area identification Figure 8: Malaria-risk map showing the status of
Factors Weight Rank Degree of villages in the study area
Vulnerability
Hazard map 63 1 Very high 4. Discussion
2 High
3 Moderate The importance of GIS techniques is recognized in areas
4 Low of disease prevalence and treatment (Carlos et al.,
Element at 31 1 Very high 2010). The GIS-based malaria incidence mapping has
risk map 2 High been used for risk assessment at national, regional, and
2 Moderate local levels in the context of resource allocation,
4 Low management and to combat the disease (Saxena et al.,
2009). Probability of the transmission of malaria to the
Vulnerability 6 2 Moderate
present study area was determined by climatic, non-
map 3 Low
climatic and biological factors.
4 Very low
Areas near water bodies showed a low prevalence of
Table 7: Summary of the results for malaria-risk
malaria. There is no stagnation of water in the river.
and its layers
When there is heavy rain fall, mosquitoes cannot
develop when the river flows fast in the absence of
Type of Area Area Degree of stagnant water bodies. However, in some areas, ponds
area (km2) (%) Vulnerability are created close to the streams and rivers in order to
Malaria 46.77 0.31 Very high store water for the dry season. Such ponds act as
hazard map 64504.51 43.25 High resourceful areas for mosquitoes to develop and
76446.74 51.26 Moderate contribute to increase malaria prevalence (Bautista et
8122.58 5.44 Low al., 2006). A negative association between distance
Vulnerability 1878.04 1.26 Moderate from swampy areas and malaria-risk exists in the study
map 39794.21 26.68 Low area. It is already revealed that a strong positive
107446.73 72 very Low association exists between malaria incidence and water
Element at 48523.66 32.54 Very high bodies (Yihenew, 2007; Aster, 2010; Abdulhakim,
risk map 21365.29 14.32 High 2013).
75556.48 50.66 Moderate
3673.56 2.46 Low Higher elevation in general has long been recognized to
Final 33.59 0.225 Very high be negatively associated with malaria due to its
Malaria 69305.82 46.47 High association with cooler temperatures that slows the
risk map 76830.96 51.52 Moderate development of anopheline vectors and the Plasmodium
2948.61 1.97 Low parasites they transmit (Patz et al., 2003). Malaria
prevalence decreases with increase in altitude,
particularly above 2000 m (Patz et al., 2003; Bautista et
al., 2006). The present study shows higher negative
 62

Journal of Geomatics Vol 10 No. 1 April 2016

correlation between monthly incidence of malaria and localities helps the health authorities to minimize
altitude (Yihenew, 2007; Yazoume et al., 2008). expenditure. The present study shows negative
correlation between monthly incidence of malaria and
Rainfall results in an increase in the prevalence of distance from health facilities. The relationship of
malaria as rains provide good breeding sites for the distance from health stations and malaria incidence was
mosquito vectors (Stephen, 2006; Omukunda et al., statistically significant.
2013). The correlation coefficient for the association
between monthly rainfall and monthly incidence of The high human population density has caused over-
malaria was found greater than that for the association cultivation, and severe environmental manipulations
between other variables assessed. A rise in temperature leading to extensive drought and recurrent famine in
enhances the survival chances of Anopheles mosquitoes many areas. This leads to more movement of people
and the Plasmodium and thus accelerates the from one area to another resulting in transmission of
transmission dynamics of malaria. There was a negative malaria (Wakgari et al., 2006; Wiseman et al., 2006).
relationship between malaria and temperature during Malaria-risk may increase in certain regions due to
the months January to April and in August, when the population movement by labor related to agriculture,
study area had the minimum and maximum mining, conflict and refugees (Martens and Hall, 2000).
temperatures, respectively. Hence, an increase in Work opportunities and resettlement programs in
temperatures does not mean an increase in the malaria malaria endemic areas can easily attract a large number
transmission risk if this is accompanied by a decrease in of people, making them vulnerable to the disease
rainfall. Although temperature favors Plasmodium to (Meron, 2010). Major environmental transformations
develop, lack of water prevents development of the like deforestation and new construction take place
vector. during resettlement, enhancing the proliferation of
mosquito breeding sites, and result in malaria outbreaks
The habitats of mosquitoes differ according to the (Martens et al., 1995; Kathleen, 2002; Aster, 2010). The
vegetation and the nature of local environment. Land- present study shows strong positive correlation between
use/land-cover types with plantation and bush lands monthly incidence of malaria and human population
have strong association with malaria indicating that this density.
LU/LC may be a proxy for predictors of elevated
malaria-risk (Richard and Poccard, 1998; Yasuoka and Findings of the present study show that, a model-based
Levins, 2007; Ehlkes et al., 2014). The relationship of malaria-risk map can be developed by establishing the
NDVI to Entomological Inoculation Rate (EIR) is relationship of various parameters using remote sensing
highly correlated. The lower the NDVI value indicates and geographic information system. It also reveals that
the lower the vegetation level and the area is thought to remote sensing and GIS techniques can be effectively
be dry (Ceccato et al., 2005). The present results show used in mosquito larval habitat identification and risk
high correlation between the incidence of malaria and area mapping. The risk area identification map indicates
vegetation due to the prevalence of high NDVI value as affected areas. The final malaria-risk map of the study
reported earlier (Oliveira et al., 2013; Ehlkes et al., area shows that the entire study area has malaria-risk
2014; Solomon et al., 2015). factors. The study area falls under very high, high,
moderate and low risk areas. The malaria-risk map
In areas with low slopes, water tends to be logged, and developed can support decision makers to take
such conditions accelerate chances for water stagnation. precautions in space and time so as to control and
Absence of proper water drainage may lead to the manage malaria incidence.
creation of stagnant water pools, which in turn,
encourages breeding and survival of mosquitoes Acknowledgements
(Thomson et al., 1999). The present study shows that
slope gradient has negative influence on malaria The authors are grateful to the School of Earth Sciences,
incidence. This relationship could be attributed to the College of Natural and Computational Sciences, Addis
different slope types found on different geographic Ababa University, Addis Ababa, for providing financial
locations across the studied landscapes. The bottom support to carry out this study.
areas characterized by flat or gentle slopes are mostly
under swamps and water bodies. As slope increases References
from lower parts to middle and upper slopes, mosquito
populations cannot sustain. Abdulhakim, A. (2013). GIS and remote sensing
integrated for malaria risk mapping, Kersa, Ethiopia,
Presence of health institutions in a particular area is very MSc. Thesis. Teri University, New Delhi.
important for treatment of patients, awareness creation
and to implement preventive measures (Meron, 2010). Ashenafi, M. (2003). Design and water management of
These in turn influence the prevalence of disease in the irrigation systems to control breeding
locality. Absence and distant health institutions result in of anopheles mosquitoes. Case study: Hara irrigation
difficulties in accessibility and enhanced cost of project, Arba Minch, Ethiopia, MSc. Thesis,
treatment. Therefore, people who are near to health Wageningen University, Wageningen.
institutions are safer relative to those who are away from
such centers. Identification of potential malaria-risk
 63

Journal of Geomatics Vol 10 No. 1 April 2016

Aster, T. (2010). GIS and RS based assessment of Martens, P. and L. Hall (2000). Malaria on the move:
malaria risk mapping for Boricha Woreda, Ethiopia, Human population movement and malaria transmission.
MSc. Thesis, Addis Ababa University, Addis Ababa. Emerging Infectious Diseases 6:103−109.

Bautista, C.T., A.S. Chan, J.R. Ryan, C. Calampa, M.H. Martens, W.J., L.W. Niessen, J. Rotmans, T.H.Jetten
Ropers, A.W. Hightower and A.J. Magill (2006). and A.J. McMichael (1995). Potential impact of global
Epidemiology and spatial analysis of malaria in the climate change on malaria risk. Environmental Health
Northern Peruvian Amazon. American Journal of Perspective 103:458−64.
Tropical Medicine and Hygiene 75:1216−22.
Meron, M. (2010). Web GIS in decision support to
Carlos, A.V., A.F. Julián, M.C. Zulma, R. Patricia, C.L. control malaria: Case study in TiroAfetaWoreda,
Myriam and D. Sofia (2010). Correlation between Oromia Region, Ethiopia, MSc. Thesis, Addis Ababa
malaria incidence and prevalence of soil-transmitted University, Addis Ababa.
helminthes in Colombia: An ecologic evaluation.
Biomédica 30:501–508. Oliveira, E.C., E.S. Santos, P. Zeilhofer, R. Souza-
Santos and M. Atanaka-Santos (2013). Geographic
Ceccato, P., S. Connor and M.C. Thomson (2005). information systems and logistic regression for high-
Application of geographical information systems and resolution malaria risk mapping in a rural settlement of
remote sensing technologies for assessing and the southern Brazilian Amazon. Malaria Journal,
monitoring malaria risk. Parassitologia 47: 81−96. doi:10.1186/1475-2875-12-420.

Dongus, S., D. Nyika, K. Kannady, D. Mtasiwa, H. Omukunda, E., A. Githeko, M.F. Ndong, E.
Mshinda and U. Fillinger (2007). Participatory mapping Mushinzimana and G. Yanr (2013). Effect of swamp
of target areas to enable operational larval source cultivation on distribution of anopheline larval habitats
management to suppress malaria vector mosquitoes in in Western Kenya. Journal of Vector Borne Disease 49:
Dar es Salaam, Tanzania. International Journal of 61–71.
Health Geographics 6:37. doi:10.1186/1476-072X-6-
37. Palaniyandi, M. (2012). The role of remote sensing and
GIS for spatial prediction of vector-borne diseases
Ehlkes, L., C. Anne, B. Kreuels, R. Krumkamp and O. transmission. Journal of Vector Borne Disease 49: 197–
Adjei (2014). Geographically weighted regression of 204.
land cover determinants of Plasmodium falciparum
transmission in the Ashanti Region of Ghana, Patz, J.A., A.K. Githeko, J.P. McCarty, S. Hussein, U.
International Journal of Health Geographics 13:13−35. Confalonieri and N. Wet (2003). Climate change and
infectious diseases. Climate Change and Human Health:
Erin, M., S. Jennifer, G. Katya, Y. Amrish, B. Teun, O. Risks and Responses103–137.
Wycliffe, K. Simon, D. Chris, A. Thomas, J.C. Smith
and C. Nakul (2014). The cost effectiveness of malaria Richard, Y. and I. Poccard (1998). A statistical study of
control interventions in the highlands of Western NDVI sensitivity to seasonal and inter- annual rainfall
Kenya. PLoS One.; 9: e107700. variations in Southern Africa. International Journal of
Remote Sensing 19:2907–2920.
Hay, S.I., J.A. Omumbo, M.H. Craig and R.W. Snow
(2000). Earth observation, geographic information Saxena, R., B.N. Nagpal, A. Srivastava, S.K. Gupta and
system and Plasmodium falciparum malaria in sub- A.P. Dash (2009). Application of spatial technology in
Saharan Africa. Advances in Parasitology 47:173–215. malaria research & control: Some new insights. Indian
Journal of Medical Research 130:125−132.
Kathleen, W. (2002). A review of control methods for
African malaria vectors. Environmental Health Project, Sithiprasasna, R., W.J. Lee, D.M. Ugsang and K.J.
Activity Report 108, U.S. Agency for International Linthicum (2005). Identification and characterization of
larval and adult anopheline mosquito habitats in the
Development, Washington.
Republic of Korea: Potential use of remotely sensed
data to estimate mosquito distributions. International
Kleinschmidt, I., B.L. Sharp, G.P.Y. Clarke, B. Curtis
Journal of Health Geographics 4:1−11.
and C. Fraser (2001). Use of generalised linear mixed
models in the spatial analysis of small area malaria
Solomon, N., K.V. Suryabhagavan and E. Gadisa
incidence rates in KwaZulu Natal, South Africa.
(2015). Visceral Leishmaniasis (Kala-Azar) risk
American Journal of Epidemiology 153:1213−21.
mapping using geo-spatial tools: A case study in Kafta
Humera district, North Western Ethiopia, International
Kleinschmidt, I., J. Omumbo, O. Briët, N. Van de
Journal of Geoinformatics 11:21−26.
Giesen, N. Sogoba and N.K. Mensah (2001). An
empirical malaria distribution map for West Africa.
Stephen, M. (2006). Landscape determinants and
Trop Med Int Health 6:779–86.
remote sensing of mosquito larval habitats in the high
 64

Journal of Geomatics Vol 10 No. 1 April 2016

land of Kenya. American Journal of Tropical Medicine expenditure and determinants of demand at the
and Hygiene 74:69−75. household level. Tropical Medicine and International
Health 11:419–431.
Thomson, MC., S.J. Connor, U. D’Alessandro, B.
Rowlingson, P. Diggle, M. Cresswell, C. Dark and G. Yasuoka, J. and R. Levins (2007). Impact of
Brian (1999). Predicting malaria infection in Gambian deforestation and agricultural development on
children from satellite data and bed net use surveys: The Anopheline ecology and malaria epidemiology.
importance of spatial correlation in the interpretation of America Journal of Tropical Medicine and Hygiene
results. American Journal of Tropical Medicine and 76:450–460.
Hygiene 61:2–8.
Yazoume, Y., S. Osman, K. Bocar and S. Rainer (2008).
Wakgari, D., A. Ahmed and B. Yemane (2006). The Environmental factors and malaria transmission risk:
interplay between population dynamics and malaria Modelling the risk in a holoendemic area of Burkina
transmission in Ethiopia. Ethiopian Journal of Health Faso. Ashgate Publishing Ltd., Surrey.
Development 20:137−144.
Yihenew, A. (2007). Irrigation and socio-economic
Wiseman, V.B., L. McElroy and W. Conteh (2006). factors related to malaria transmission in Ziway. M.Sc
Malaria prevention in the Gambia: Patterns of Thesis, Addis Ababa University, Addis Ababa.
 65

Journal of Geomatics Vol 10 No. 1 April 2016

Object based classification techniques for citrus orchards
Kadapala Bharath Kumar Reddy1, J. Mohammed Ahamed2, R. Hebbara, Uday Raj2 and Rama Rao1
1
Indian Institute of Space Science and Technology, Thiruvananthapuram, Kerala, India
2
RRSC – South, National Remote Sensing Centre, Bengaluru, Karnataka, India
Email: [email protected], (mohammedahamed_j, hebbar_kr, uday_raj)@nrsc.gov.in

(Received: Mar 02, 2016; in final form: Apr 13, 2016)

Abstract: Cultivation of citrus orchards is characterized by small field size, sparse distribution with different age groups
and coexisting spectrally similar crops. Thus, classification of citrus crop using satellite data is quite challenging. The
present study was carried out to evaluate object based classification techniques for mapping citrus orchards located at
Indi hobli of Bijapur district, Karnataka state. Two date LISS-IV and single date Cartosat-1 data were used for
classification. Spectral signature of young orchards less than 5 to 6 years and poorly managed orchards were mixed with
pomegranate, sugarcane and grape orchards. Two approaches of segmentation techniques namely, threshold & clump
and lambda schedule, were tested. The results revealed that the single date satellite data showed classification accuracy
(around 75%) using both threshold and clump and lambda schedule segmentation approach. Inclusion of second date data
along with vegetation indices significantly improved the mapping accuracy (around 85%) by eliminating short duration
crops from evergreen citrus orchards. The study explored the potential use of high resolution data for inventory of citrus
orchards and the methodology could be refined for operational application using textural and contextual information.

Keywords: Object based image analysis, Full lambda schedule Segmentation, Threshold and clumping, Citrus, Brovey
transform

1. Introduction spectral, texture and contextual information in an

integrative way(Blaschke, 2010). Dissatisfaction of
Image classification is perhaps the most important part using pixels solely in the classification has been
of digital image analysis. It is very nice to have a "pretty mentioned long back (Cracknell, 1998). Spectrally
picture" or an image, showing a magnitude of colors similar but compositionally different land cover may be
illustrating various features of the underlying terrain, misclassified. Similarly, the spectral heterogeneity of
but it is quite useless unless it is known what the colors the land cover can lead to rogue pixels appearing within
mean (Korgaonkar, 2012). The data obtained through classes creating a ‘salt and pepper’ effect. In addition to
remote sensing satellites has huge applications in the this, the increased application of higher resolution
fields of agriculture, urban modeling, disaster imagery is problematic as it is difficult to classify
management etc. For all these applications a well accurately using traditional pixel-based methods. The
classified data is required i.e., image classification has increased amount of spatial information often leads to
to be carried out in order to group all the pixels into an inconsistent classification of pixels(Whiteside,
several land cover classes and this further can be used 2005). In GEOBIA the spatial information in the
according to user needs. neighborhood is also considered which allows
increasing the dimensionality of the feature space of a
1.1 Classification methods pixel when compared to the traditional pixel based
Classification methods can be mainly categorized into methods where, only spectral values of the pixel are
two types namely: used in the feature space and hence giving us the more
1. Pixel based classification reliable results than those methods. Hence, in this study
2. Object based classification the object based classification methods were evaluated
to identify the best object based method in mapping
1.1.1 Pixel based classification: The pixel based citrus orchards.
methods use the digital number associated with the pixel
in order to assign it to a specific class i.e., the spectral 2. Data used
information of the pixel is used for the classification.
There are mainly two methods in pixel based Multi-spectral images of LISS IV camera and
classification methods, Supervised and Unsupervised CARTOSAT data were used. Indi taluk of Bijapur
classification. District, Karnataka was identified for the study. This
taluk is located with the following boundaries of
1.1.2 Object based classification: The object based coordinates: Upper bound ULX: 75o 43’36.13”, ULY:
image analysis (OBIA) or specifically Geographic 17o 0’ 20.49” and Lower bound LRX: 76o 03’3.16”
Object Based Image Analysis (GEOBIA) when it comes LRY: 16o 56’17.04”. The dominant crop is Citrus with
to satellite images delineate readily usable objects from small amount of grapes, sugarcane and other crops
imagery while at the same time combining image grown in the region.
processing and GIS functionalities in order to utilize

© Indian Society of Geomatics

 66

Journal of Geomatics Vol 10 No. 1 April 2016

3.1 Segmentation using full lambda schedule
This method iteratively merges adjacent segments based
on a combination of spectral and spatial information.
Merging occurs when the algorithm finds a pair of
adjacent regions, i and j, such that the merging cost ti,j is
less than a defined threshold lambda value:
|𝑂𝑖 |. |𝑂𝑗 | 2
. ‖𝑢𝑖 − 𝑢𝑗 ‖
|𝑂𝑖 | + |𝑂𝑗 |
𝑡𝑖,𝑗 =
𝑙𝑒𝑛𝑔𝑡ℎ(𝜕(𝑂𝑖 , 𝑂𝑗 )) 1

where,
Oi is the area of region i.
Oj is the area of region j.
Figure: 1a) LISS IV May Figure: 1b) LISS IV ui is the average value in region i.
November uj is the average value in region j.
║ui-uj║is the Euclidean distance between the
Classification was done classification for the data spectral values of regions i and j.
obtained in the month of May (figure 1a) and also using length(∂(Oi,Oj)) is the length of the common boundary
the stacked data of May and November. Summer crops of Oi and Oj (Robinson, 2002).
get harvested in the month of May and monsoon crops
get harvested in the month of November (figure 1b). While not necessarily derived from the Pixel Probability
Therefore, these two months data was used to exclude Layer, the raster object segments will have the zonal
seasonal crops as much as possible in the images. mean pixel probabilities as attributes. Output from the
Probability Pixels to Raster Objects Operator contains
LISS IV sensors have three bands with a spatial pixels that are grouped as raster objects which have
resolution of 5.8 m. CARTOSAT-1 panchromatic has associated probability attributes (Erdas, 2013).
one broad band with very good spatial resolution—2.5
m. Combining these two images to yield a three band 3.2 Threshold and clumping
data set with 2.5 m resolution provides the best This operator performs a threshold operation on a pixel
characteristics of both sensors (NRSC, 2004). probability layer which keeps only pixels with a
probability greater than or equal to the threshold value
Brovey Transform was used to merge the two data sets. specified. It converts the pixels to binary values (0’s and
The Brovey Transform was developed to visually 1’s), then performs a contiguity operation (clump) on
increase contrast in the low and high ends of an image’s the binary values of 1(Imagine Objective, 2010).
histogram (i.e., to provide contrast in shadows, water
and high reflectance areas such as urban features). Now the objects formed were edited from the above two
Consequently, the Brovey Transform should not be used methods by applying filters over the objects to refine the
if preserving the original scene radiometry is important. results. Focal filter was used in the beginning to make
However, it is good for producing RGB images with a the object edges sharp and remove very narrow strips
higher degree of contrast at the tails of the image from the objects formed. This filter replaces the most
histogram. repeating value in the window of specific size around
the pixel. The optimum value for this was found to be 3.
Our objective was to perform a binary class The next filter used was probability filter which
classification for the identification of Citrus orchards in removes the objects of lower probability and gives us
the given area. Object based analysis of the image was more reliable objects after filtering. Later size filter was
carried out to identify the citrus orchards and mark all used to filter out very small objects. Finally, re-clump
the other regions as the background. filter was used on objects to form bigger objects. Raster
to vector conversion was made using polygon tracing
3. Methodology and work flow and later two more filter are applied on the vector
objects. The smoothening filter which eliminates the
In ERDAS imagine, Objective Workstation was used. sharp edges of objects and the island filter takes out
The first step was pixel level binary classification of the island like gaps in the formed objects were used in post
image and assigns probability for each and every pixel processing operation. The work flow is pictorially
depending on how close it is to Citrus class. For this step represented in a flowchart in the Appendix-1.
sixtyfour training samples were given from both Citrus
and Non Citrus equally covering all the variations. The 4. Results and discussions
next step i.e., objects formation step which is an
important step in the analysis. Two different methods The main aim of the entire process was to classify the
were followed to create objects in this step. One was mature citrus orchards. It can be seen how different
segmentation using full lambda schedule and the second parameter sets (Table 1) had performed in citrus
was threshold and Clump. dominant and non-citrus dominant regions from images
shown (Figure 2a, 2b).
 67

Journal of Geomatics Vol 10 No. 1 April 2016

It was also observed that few parameter sets had this method are mentioned in table 1. To overcome the
classified the citrus very well in its dominant region but above mentioned problem the second method, i.e., Point
there was a lot of misclassification in the non-citrus based assessment was carried out. Here, random points
dominant region. Two different assessment methods were generated with 30samples per class. Later analysis
were used to calculate accuracy. The first method is was carried out to identify correctly classified samples
from area estimation based on the actual data obtained using ground based GPS locations. The results of this
from the government statistics. This method might not assessment method are described in the form of
give precise idea about which method had the best confusion matrix (Table 2). It can be seen that highest
performance because if some non- citrus dominant accuracy of 85.07 per cent was observed for the set 11
regions are classified as citrus and vice-versa, it may in which NDVI was a stacked along with multi-data data
match the reported area even though spatially the using lambda schedule segmentation.
classification was incorrect. The area accuracy results of

Table 1: Object creation methods, parameters used and respective area statistics and accuracy obtained in
identifying citrus orchards

Single Date/ Object Creation No. Spectral Texture Size Shape Probability Estimated Mapping
Multi Date Method Threshold area (ha) Accuracy
(%)
Set1 0.8 1 0.1 0.05 NA 2490 61.81
Segmentation- Set2 0.6 1 0.15 0.05 NA 1146 63.59
Lambda Schedule
Single Date Set3 0.6 1 0.15 0.05 NA 1140 63.27
Threshold/ Clump Set4 NA NA NA NA 0.95 1888 95.24
Set5 NA NA NA NA 0.9 1152 63.91
Set6 NA NA NA NA 0.95 1557 86.40
Segmentation- Set7 0.5 1 0.02 0.02 NA 843 46.79
Lambda Schedule Set8 0.6 1 0.15 0.05 NA 1402 77.79
Multi Date
Set9 0.8 1 0.15 0.05 NA 1411 78.30
Threshold/ Clump Set10 NA NA NA NA 0.9 1460 81.03
Multi Date Segmentation- Set11 0.7 1 0.15 0.05 NA 1440 79.90
+ NDVI Lambda Schedule

4.1 Lambda schedule segmentation (single date) increasing the probability filter value to 0.6 and
In raster object operation (ROO) process the probability reducing the spectral value to 0.6, which gave very less
filter values were tested iteratively from 0.45 to 0.8. The area estimate 1146 hectares as compared with 1800
optimized value for probability filter was found to be hectares of reported area. However, misclassification of
0.6 for lambda schedule and 0.7 for the Threshold and non-citrus regions was reduced.
Clump process. For Lambda Schedule it was observed
that by giving 0.7 as probability filter value area under 4.2 Threshold and clump segmentation (single date)
citrus drastically changed resulting in under In threshold and clump the threshold probability in ROC
classification and decreasing it to 0.5, resulted in was initially set to be 0.95 (table 2) and probability
misclassifying non-citrus regions. Hence the optimum value of 0.6 in ROO, which resulted in smaller object
value for this parameter was found to be 0.6. In size and misclassification of non-citrus segments. Even
threshold and clump method the value was increased to though the area estimated looked very precise (95.24 per
0.7 because in this step, only high probable pixels were cent) spatially there was significant misclassification.
taken into account in the raster object creation (ROC) Hence, the probability filter value in ROO step was
step itself which precedes ROO step in segmentation. increased to 0.8 which reduced the misclassification. It
Hence, higher values were needed to be kept as cut-off also had smaller segments of objects existing adjacently
for probability filter. which could join to form a bigger object if threshold in
the object formation step is reduced. To achieve this,
The set 1 estimated the area as 2490 hectares and overall there was a need to decrease probability value in ROC.
accuracy was 73.13 per cent. But it was observed that a Hence, two changes were made i.e., the threshold
lot of non-citrus regions were classified as citrus probability was set to 0.9 and probability filter in ROO
orchards. Hence, changes were made in probability was set to 0.7.
filter in ROO step and spectral value in ROC step by
 68

Journal of Geomatics Vol 10 No. 1 April 2016

` Figure 2a Figure 2b
Figure 2: From top left to bottom right in each of the figures show the classification results for 11 parameter
sets from (a) citrus dominant region; and (b) non-citrus dominant region

Table 2: Table of confusion matrices for selected sets using point based assessment method
Lambda Schedule (Set 1) Threshold and Clump(Set 5)
Citrus Non citrus Total Citrus Non citrus Total
Classified classified
Citrus 31 15 46 Citrus 27 11 38
Non citrus 3 18 21 Non Citrus 7 22 29
Total samples 34 33 Total samples 34 33
taken taken
Accuracy 91.17 54.55% 73.13% Accuracy 79.41 66.67% 73.13%
% %
Lambda Schedule (Set 2) Lambda Schedule (Set 9)
Citrus Non citrus Total Citrus Non citrus Total
classified classified
Citrus 28 10 38 Citrus 29 9 38
Non citrus 6 23 29 Non Citrus 5 24 29
Total samples 34 33 Total samples 34 33
taken taken
Accuracy 82.35 69.70% 76.12% Accuracy 85.29 72.72% 79.10%
% %
Threshold and Clump (Set 10) Lambda Schedule (Set 11)
Citrus Non citrus Total Citrus Non citrus Total
classified classified
Citrus 30 12 42 Citrus 31 6 37
Non citrus 4 21 25 Non Citrus 4 26 30
Total samples 34 33 Total samples 34 33
taken taken
Accuracy 88.23 63.63% 76.12% Accuracy 91.14 78.90% 85.07%
% %
 69

Journal of Geomatics Vol 10 No. 1 April 2016

The best methods in the two types of algorithms using made on young citrus plants too but unconvincing
single date data had accuracies around 76.12 per cent for results were observed because of the resolution of this
lambda schedule segmentation (set 2 of table 2) and data as young citrus plants get mixed-up with other
73.13 per cent (set 5 of table 2) for threshold and clump orchards. The young plants were not identifiable in the
method of segmentation. Hence there is a need to data; they looked similar to other fallow lands which
increase the feature space to capture the seasonal made the method specific for mature orchards at this
variability of crops and improve the classification. The level of resolution.
data for the month of November was added which gave
us better results (set 9 and 10 of table 2) compared to Further this work can be validated by using this
above mentioned single date classification methods. technique in dominant citrus growing regions in India.
Use of higher resolution data products may enhance the
4.3 Multi-date and multi-date with thematic layer classification accuracy. Reported statistics from state
included departments include both mature and young citrus
Classification carried out using multi-date data, set 9 orchards. Since, the classification technique targeted on
and set 10 (table 2) reduced the misclassification. mature citrus orchards, area estimated by this technique
Overall accuracy was 79.1 and 76.12 per cent might be on par with the reported data.
respectively. The results of both the methods were
observed and found that threshold clump method of References
segmentation had more misclassification occurring in
non-citrus regions. This was observed clearly in the Blaschke, T. (2010). Object based image analysis for
second accuracy assessment (Table 2). Also it was remote sensing. ISPRS Journal of Photogrammetry and
observed that threshold and clump method had unusual Remote Sensing, 1-6.
shaped objects. However, lambda schedule
segmentation method reduced the misclassification
significantly and the boundary shape of the objects Blaschke T. and J. Stroble (2001). What’s wrong with
created matched the field boundaries. Therefore, by pixels? Some recent developments interfacing remote
considering the above factors into account lambda sensing and GIS, GIS-Zeitschrift für
schedule segmentation method was used to include one Geoinformationsysteme 6:12-17.
more additional layer (Set 11 of table 1) of Normalized
difference vegetation index (NDVI) to further attempt Cracknell, A. (1998). Synergy in remote sensing —
in improving the classification accuracy.
what’s in a pixel? International Journal of Remote
Since NDVI layer describes additional spectral Sensing, 2025–2047.
information, iteratively a value of 0.7 was selected as
spectral weightage instead of 0.8 which performed Crisp, D.J. P. Perry and N.J. Redding (2002). Fast
better by reducing misclassification with multi-date segmentation of large images. Edenberg, South
data. A lot of misclassification were removed which is Australia: Defence Research and Surveillance &
desired. It was observed that temporal data with Reconnaisanse Division,DSTO Electronics and
additional NDVI as a layer included in classification
Surveillance Research Laboratory.
using lambda schedule segmentation significantly
reduced the misclassification and also increased the
mapping accuracy 79.9 per cent (table 1). Erdas (2010). IMAGINE Objective. Norcross, USA:
ERDAS Inc.
5. Conclusion and limitations
Erdas (2013). Erdas Field Guide.
From the above results it was observed that lambda
schedule segmentation algorithm worked well as a
raster object creation method with multi-date data and Korgaonkar, G.S., R.R.Sedamkar and K. Bhandari
when the data was combined with NDVI which adds the (2012). Hyperspectral image classification on decision
additional spectral information the results were level fusion. International Journal of Computer
enhanced. Although through the first assessment Applications (IJCA), 1-2.
method where only total area is taken into consideration
threshold and clumping method shows an accuracy of NRSC (2004). IRS-P6 data user’s Manual. Hyderabad.
over 95.24% for set 4, however when checked with
ground based GPS points, misclassification was clearly
observed (table 2). Similar reason can also be concluded Robinson, D. J. (2002). Implementation of a fast
for other methods using different parameter sets. Even algorithm for segmenting SAR imagery. Edinburgh,
in the set 11 where NDVI was used along with multi- South Australia, Australia: DSTO Electronics and
date data, it was observed that trees along the road side Surveillance Research Laboratory.
and clusters of trees were classified as citrus.
Whiteside, T. and W. Ahamad (2005). A comparison of
This method is applicable only to the mature Citrus
object-oriented and pixel-based methods in northern
orchards at this resolution limits. A similar attempt was
 70

Journal of Geomatics Vol 10 No. 1 April 2016

Australia. Proceedings of SSC2005 Spatial intelligence, of the Spatial Sciences Institute, September 2005.
innovation and praxis: The national biennial Conference Melbourne: Spatial Sciences Institute.

APPENDIX – 1

(Flowchart describing the methodology adopted in identification of citrus orchards)

 71

Journal of Geomatics Vol 10 No. 1 April 2016

Evaluation of predictive ability of support vector machines and naive Bayes trees
methods for spatial prediction of landslides in Uttarakhand state (India) using GIS
Binh Thai Pham1,2, DieuTien Bui3, Indra Prakash4 and M.B. Dholakia1

1
Department of Civil Engineering, Gujarat Technological University, Ahmedabad - 382424, India
2
Department of Geotechnical Engineering, University of Transport Technology, Hanoi, Vietnam
3
Geographic Information System Group, Department of Business Administration and Computer Science, Telemark
University College, HallvardEikasPlass 1, N-3800 BøiTelemark, Norway
4
Bhaskarcharya Institute for Space Applications and Geo-Informatics (BISAG), Gandhinagar, India
Email: [email protected] or [email protected]

(Received: Feb 11, 2016; in final form: Mar 29, 2016)

Abstract: The main objective of this study is to apply and evaluate the predictive capability of the Support Vector
Machines (SVM) and Naïve Bayes Trees (NBT) methods for spatial prediction of landslides in a part of Uttarakhand
state (India). SVM is one of the most efficient machine learning methods that has been applied widely in landslide
prediction whereas NBT has not been applied for landslide problems. In these models, a total of 430 historical landslide
locations have been first identified to construct landslide inventory map. Landslide locations have been split randomly
into two parts to generate training dataset (70% landslide locations) and testing dataset (30% landslide locations).
Secondly, landslide affecting factors such as slope angle, slope aspect, elevation, plan curvature, lithology, soil, land
cover, distance to roads, distance to rivers, distance to lineaments, and rainfall have been selected to assess the spatial
relationship with landslide occurrences. The predictive capability of these factors has been evaluated using the Gain Ratio
method. Using training dataset, the SVM and NBT models have been constructed to assess the susceptibility of landslide
occurrences. Finally, the performance of the SVM and NBT models have been validated and compared using receiver
operating characteristic curve technique and statistical index-based evaluations. The results show that both the SVM and
NBT models perform well for spatial prediction of landslides. Out of these, the SVM model (AUC = 0.881) outperforms
the NBT model (AUC = 0.832). Overall, SVM and NBT indicate promising methods which could be used for spatial
prediction of landslides in landslide prone areas. Moreover, the results obtained from this study could be helpful for
planning and decision making in landslide hazard management.

Keywords: Landslides, GIS, Support vector machines, Naïve Bayes trees

1. Introduction (GIS) is known as standard tool for integration of

different types of data collected from various sources.
Landslide is one of the most devastating natural
disasters causing loss of human life and properties all Many methods have been applied for spatial prediction
over the world. India is known as one of the most of landslides using GIS in recent decades. These
affected countries by landslides in Asia (Guha-Sapir et methods are based on main approaches (i) expert
al., 2014). Approximately 300 people die and 46 USD opinion-based approach and (ii) data mining based
millions in properties loss every year in India (GSI, approach (Song et al., 2012). Expert opinion-based
2009). Most of landslides (about 80%) have occurred in approach is subjective because it is based on the
Himalayan area (Onagh et al., 2012). Landslide studies perspective of experts in selecting variables and
have been turning into urgent tasks not only in India but assigning weights to variables. On the other hand, data
also all over the world in order to reduce their harmful mining based approach is objective as it uses machine
impaction to human life. learning algorithms to determine factors leading to
landslide occurrences and calculate weights of the
Spatial prediction of landslides is the probability of factors during learning of models. Out of these
potential instability of slopes related to a set of casual approaches, data mining based approach is more
factors (Guzzetti et al., 2005). It can be carried out by commonly utilized for spatial prediction of landslides.
analyzing the spatial relationship between past landslide Common machine learning methods are logistic
events and a set of geo-environmental factors. It is based regression (Devkota et al., 2013; Lucà et al., 2011),
on an assumption that future landslides will occur under decision tree (Pradhan, 2013; Yeon et al., 2010),
same conditions with previous landslides (Ermini et al., artificial neural network (Choi et al., 2010; Zare et al.,
2005). Landslide susceptibility map is a final outcome 2013) and support vector machines (Kavzoglu et al.,
of spatial prediction of landslides. It helps in land use 2014; Pradhan, 2013). In addition, Naïve Bayes Trees
planning and decision making for landslide hazard (NBT) is also an efficient machine learning technique
management (Pham et al., 2015a; Wang et al., 2009). To that has been applied successfully in other fields but
produce this map, Geographic Information System landslide problems.

© Indian Society of Geomatics

 72

Journal of Geomatics Vol 10 No. 1 April 2016

In the present study, Support Vector Machines (SVM) are two types of soils in this study area viz., silty and
and NBT methods have been applied and compared for loamy. Loamy soil occupies 73.73% of area.
spatial prediction of landslides. A small portion of
Uttarakhand state, India had been selected as the study Land cover in this study includes four categories such
area. Receiver Operating Characteristic (ROC) curve as dense forest, open-forest, non-forest, and scrub land.
method and statistic index-based evaluations have been Non-forest is the dominant land cover (39.02%).
used to validate and compare these landslide models.
The analysis process has been done by using GIS 3. Methodology
application and Weka 3.7.12 software.
Methodology of this study includes four main steps (i)
2. Description of study area constructing database for spatial prediction of
landslides, (ii) evaluating predictive capability of
The study area lies between TehriGarhwal district and landslide affecting factors, (ii) constructing landslide
PauriGarhwal district of Uttarakhand state in India models (SVM, NBT) to assess landslide susceptibility,
(longitudes 78o29’01’E to 78o37’06’’E and latitudes (iii) evaluating landslide models, (iv) constructing
29o56’38’’N to 30o09’37’’N) covering an area of about landslide susceptibility maps using Geo-informatics
323.815 km² (Fig. 1). The study area is situated in technology.
subtropical monsoon region. The highest temperature is
about 45°C in summer season whereas the lowest 3.1. Data collection and interpretation
temperature is around 1.3°C in winter season. The 3.1.1. Landslide inventory map: Landslide inventory
humidity varies from 25% to 85%. Heavy rainfall often map is a compilation of landslide locations that have
occurs in monsoon season (June to September) with been occurred in the past and present. It is one of the
annual average rainfall ranging from 770mm to most important data for spatial prediction of landslides.
1684mm. In the study area, many landslides occur every year
(Onagh et al., 2012). By interpretation of Google Earth
images using Google Earth pro 7.0, a total of 430
landslide location has been identified based on
morphology and texture of past landslides to construct
landslide inventory map (Fig. 2). These landslide
locations have been evaluated in comparison with
historical landslide reports, newspaper records and
extensive field investigation. Subsequently, 70%
landslide inventory (301 locations) has been used to
construct training dataset for building landslide models
whereas 30% remaining landslide inventory (129
locations) has been utilized to build testing dataset for
validating landslide models.

Figure 1: Location of the study area

Topographically, the study area is occupied by high

mountains and intervening deep valleys (Pham et al.,
2015b). Elevation ranges from 380m to 2180m (above
mean sea level) with average elevation of 1081m. Slope
angles are relatively steep up to 70 degrees. Slope angles
of 15 to 45 degrees occupy the largest area
(85.45%).Geologically, six lithological groups have
presented in this study area namely Amri group
(quartzite, phyllite),Blaini and Krol group (boulder bed
and limestone), Bijni group (quartzite, phyllite), Jaunsar
group (phyllite and quartzite), Manikot shell limestone
Figure 2: Landslide locations and elevation map of
(limestone), Tal group (sandstone, shale, quartzite,
the study area
phyllite, and limestone) (Pham et al., 2015b). Baliana
and Krol group and Bijni group are dominant with
3.1.2. Landslide affecting factors: Landslides in the
30.1% and 28.1% of the study area, respectively. There
study area usually occur under various conditions such
as geological activities, rainfall, geomorphological
 73

Journal of Geomatics Vol 10 No. 1 April 2016

characteristics, vegetation, human activities (Sarkar et has been constructed by buffering lineament network
al., 1995; Sengupta et al., 2010). Based on the generated from Landsat-8 sattelite image in the study
mechanism of landslide occurrences and geo- area. Eleven landslide affecting factors and their classes
environmental characteristics of the study area, eleven is shown in Table 1.
landslide affecting factors (slope angle, slope aspect,
elevation, plan curvature, lithology, soil, land cover, 3.2. Methods for spatial prediction of landslides
distance to roads, distance to rivers, distance to 3.2.1. Support vector machines (SVM): SVM was
lineaments, and rainfall) have been selected to assess the first proposed by Vapnik (1995) that is one of the most
spatial relationship between them and landslide effective machine learning techniques for classification
occurrences for spatial prediction of landslides. (Kavzoglu et al., 2014). It is based on the statistical
learning theory in order to find an optimal hyper-plane
Maps of these landslide affecting factors have been in separating two classes for classification (Tien Bui et
generated using GIS application. Specifically, slope al., 2015). SVM can be trained in two main steps (i) the
angle map (Fig. 3a), slope aspect map, curvature map, original input space is first mapped into a high-
and elevation map have been extracted from DEM with dimensional feature space, (ii) the optimal hyper-plane
20 m resolution generated from ASTER Global DEM in the feature space is determined by maximizing the
collected from United States Geological Survey margins between classes (Abe, 2005).
(http://earthexplorer.usgs.gov). Lithological map (Fig.
3b) has been extracted from state geological map on a The performance of the SVM method depends
scale of 1:1000000. Land cover map (Fig. 4a) has been significantly on the choice of the kernel function (Dixon
extracted from state land cover map at a scale of and Candade, 2008). According to Tien Bui et al.
1:1000000 (http://www.ahec.org.in/wfw/maps.htm). (2012), Radial Basis Function (RBF) is one of the most
Soil map (Fig. 4b) has been generated from state soil widely used in landslide models among kernel
map at a scale of 1:1000000. Rainfall map has been functions. Therefore, RBF has been selected in this
constructed using meteorological data for 30 years from study in training SVM. In addition, two calculating
1984 to 2014 (NCEP, 2014). Distance to roads map has parameter of radial basis function kernel have been
been constructed by buffering road networks on the high optimized to obtain the best performance of the SVM
slopes (larger than 15 degrees) in the study area. model such as regularization parameter (C = 0.1) and
Distance to rivers map has been generated by buffering kernel width (  = 1).
river networks on the high slopes (larger than 15
degrees) in the study area. Distance to lineaments map

Table 1: Landslide affecting factors and their class utilized in this study

No. Landslide causal factors Classes

1 Slope angle (degree) (1) 0-8; (2); (3) 8-15; (4) 15-25; (5) 25-35; (6) 35-45; (7)> 45
(1) flat[-1]; (2) north [0-22.5 and 337.5-360]; (3) northeast [22.5-67.5]; (4) east [67.5-
2 Slope aspect 112.5]; (5) southeast [112.5-157.5]; (6) south [157.5-202.5]; (7) southwest [202.5-
247.5]; (8) west [247.5-292.5]; (9) northwest [292.5-337.5]
(1) 0-600; (2) 600-750; (3) 750-900; (4) 900-1050; (5) 1050-1200; (6) 1200-1350;
3 Elevation (m)
(7) 1350-1500; (8) 1500-1650; (9) 1650-1800; (10) > 1,800
4 Curvature (1) concave (<-0.05); (2) flat (-0.05-0.05); (3) and convex (> 0.05)
(1) Amri group [quartzite, phyllite], (2) Blaini and Krol group [boulder bed and
limestone], (3) Bijni group [quartzite, phyllite]; (4) Jaunsar group [phyllite and
5 Lithology
quartzite]; (5) Manikot shell limestone [limestone]; (6) Tal group [sandstone, shale,
quartzite, phyllite, and limestone].
6 Soil (1) coarse-loamy; (2) skeletal-loamy; (3) fine-loamy; (4) mixed-loamy; (5) fine-silt
7 Land cover (1) non-forest; (2) dense-forest; (3) open-forest; (4) scrub-land;
(1) 0-900; (2) 900-1000; (3) 1000-1100; (4) 1100-1200; (5) 1200-1300; (6) 1300-
8 Rainfall (mm)
1400; (7) 1400-1500; (8) > 1,500
(1) 0-50; (2) 50-100; (3) 100-150; (4) 150-200; (5) 200-250; (6) 250-300; (7) 300-
9 Distance to lineaments (m)
350; (8) 350-400; (9) 400-450; (10) 450-500; (11) > 500
10 Distance to roads (m) (1) 0-40; (2) 40-80;(3) 80-120; (4) 120-160; (5) 160-200; (6) > 200
11 Distance to rivers (m) (1) 0-40; (2) 40-80;(3) 80-120; (4) 120-160; (5) 160-200; (6) > 200
 74

Journal of Geomatics Vol 10 No. 1 April 2016

(a) (b)

Figure 3: (a) Slope map and (b) Lithological map

(a) (b)

Figure 4: (a) Land cover map and (b) Soil map

 75

Journal of Geomatics Vol 10 No. 1 April 2016

3.2.2. Naïve Bayes Tree (NBT): NBT was first introduced corresponding to five susceptible index intervals such as
by Kohavi (1996) which is a hybrid approach of Naïve Bayes very low, low, moderate, high, and very high (Table 2 and
and decision tree classifiers. Decision tree is one of the most Fig. 6).
commonly used in machine learning techniques that is based
on a tree-like hierarchy for classification (Zhao and Zhang,
2008). Naïve Bayes is based on Bayes' theorem that
considers all attributes are independent to maximize the
posterior probability in determination the classified classes
(Pham et al., 2015b; Soni et al., 2011). NBT is also a
classification tree based method; however, it contains both
nodes and leaves. The nodes consist of univariate splits as
normal decision tree. The leaves include Naïve Bayes
classifier (Kohavi, 1996). NBT is trained in two main steps
(i) decision tree classifier is utilized to segment the data (ii)
naïve Bayes classifier uses each segment of the data to create Fig 5: The Gain Ratio (GR) values of landslide affecting
leaves for classifying variables. The performance of NBT is factors using
considered better than naïve Bayes and decision tree
(Kohavi, 1996). The performance of landslide susceptibility maps has been
validated by overlaying them with landslide inventory map
In this study, NBT has been applied at the first time for to calculate landslide density for each susceptible class.
spatial prediction of landslides. Its performance has been Landslide density is a ratio of the percentage of landslide
compared with the performance of the SVM method. pixels and the percentage of all pixels on each susceptible
class. It can be observed from Table 2 that both landslide
4. Results susceptibility maps perform well due to the fact that most of
landslides have occurred on high and very high classes.
4. 1. Predictive capability of landslide affecting factors:
Evaluation of predictive capability of landslide affecting 4. 3. Evaluation and comparison of landslide models
factors to landslide models in the study area has been carried The performance of landslide models (SVM, NBT) has been
out using Gain Ratio (GR) technique. It is known as an validated using statistical index-based evaluations and ROC
efficient feature selection method in evaluation the technique. Statistical indices such as sensitivity and
importance of input data to models (Quinlan, 1986). Factors specificity have been taken into account to evaluate the
with higher GR values have better predictive capability than performance of landslide models. Sensitivity is the
those with lower GR values. Factors with zero GR value probability of the landslide pixels that have been classified
have no contribution to landslide models, thus it must be correctly as “landslide” class (Pham et al. 2016a). It indicates
removed in analyzing process. how good landslide models for classification of landslide
pixels. On the other hand, specificity is the probability of the
Predictive capability of eleven landslide affecting factors to non-landslide pixels that have been classified correctly as
landslide models in this study is shown in Fig. 5. It can be “non-landslide” class (Pham et al. 2016a). It shows how
observed that distance to roads has the highest predictive good landslide models for classification of non-landslide
capability to landslide models (GR = 0.149), followed by pixels.
rainfall (GR = 0.058), curvature (GR = 0.042), elevation and
slope angle (GR = 0.037), soil types (GR = 0.025), lithology ROC curve is a standard method to evaluate general
(GR = 0.021), land cover (GR = 0.01), slope aspect (GR = performance of landslide models. It is constructed by
0.008), distance to rivers (GR = 0.006), and distance to plotting pairs of values (“sensitivity” and “100-specificity”).
lineaments (GR = 0.003), respectively. In general, all eleven Area under ROC curve (AUC) is used to evaluate
landslide affecting factors are having contribution to quantitatively the performance of landslide models. If the
landslide models in the study area (GR > 0). Therefore, they AUC value is larger than 0.8, the performance of landslide
all have been selected for landslide analysis in the present models is good and acceptable (Kantardzic, 2011).
study.
The performance of landslide models is shown in Fig. 7 and
4. 2. Landslide susceptibility map Fig. 8. It can be observed that both landslide models perform
Landslide models such as SVM and NBT have been well for spatial prediction of landslides in this study (AUC >
constructed using training dataset. Thereafter, landslide 0.8). Out of these, the SVM model (AUC = 0.881)
susceptibility mapsusing these landslide models have been outperforms the NBT model (AUC = 0.832). More
constructed. Firstly, landslide susceptibility indices(LSIs) specifically, the SVM model (sensitivity = 84.4%, specificity
have been generated for all pixels in whole study area. These = 81.6%) performs better than the NBT model (sensitivity =
susceptible indices have been then reclassified into five 82.5%, specificity = 76.6%) for classification of both
intervals using geometrical interval method (Frye, 2007). landslide and non-landslide pixels. For both the SVM model
Five landslide susceptibility classes have been named
 76

Journal of Geomatics Vol 10 No. 1 April 2016

and the NBT model, the classification of landslide pixels is
significantly better than those of non-landslide pixels.

Table 2: Landslide density on landslide susceptibility maps

LSIs % Pixels % Landslides Landside Density

Class
SVM NBT SVM NBT SVM NBT SVM NBT
Very low 0-0.01 0-0.001 0.176 0.164 0.005 0.011 0.026 0.064
Low 0.01-0.039 0.001-0.009 0.248 0.205 0.012 0.004 0.047 0.021
Moderate 0.039 - 0.121 0.0089-0.043 0.260 0.257 0.035 0.016 0.134 0.063
High 0.121-0.352 0.043-0.207 0.130 0.203 0.061 0.078 0.472 0.384
Very high 0.352-1 0.207-1 0.186 0.171 0.887 0.891 4.772 5.205

(a) (b)

Figure 6: Landslide Susceptibility Maps (LSM) using (a) the SVM model and (b) the NBT model
 77

Journal of Geomatics Vol 10 No. 1 April 2016

100 widely used feature selection techniques (Karegowda et
al., 2010). The results show that in this area, distance to
roads is important factor that has the highest predictive
80 capability for landslide models. It is reasonable because
most of identified landslides are on or adjacent to roads
or highways. In addition, rainfall is also having high
contribution to landslide models; this observation
60 confirms the landslides occurrences during long-term
Sensitivity

heavy rain (Sengupta et al., 2010). Other factors are also

having contribution to landslide models. Therefore, all
40 eleven landslide affecting factors are appropriate for
SVM (AUC = 0.881) spatial prediction of landslides in the present study.
NBT (AUC = 0.832)
20
Conclusions

Overall, SVM and NBT indicate promising methods for

spatial prediction of landslides which could be also used
0 in other landslide prone areas. Out of these, the SVM
0 10 20 30 40 50 60 70 80 90 100 model is having better predictive capability than NBT
100-Specificity model. The present study would be helpful for land use
Figure 7: The performance of landslide models using planning, decision making and management of landslide
ROC curve hazard prone areas.

Acknowledgement

Authors are thankful to the Director, Bhaskarcharya

Institute for Space Applications and Geo-Informatics
(BISAG), Department of Science & Technology,
Government of Gujarat, Gandhinagar, Gujarat, India for
providing facilities to carry out this research work.

References

Abe, S. (2005). Support vector machines for pattern

classification, 2. Springer.

Choi, J., H.J. Oh, J.S. Won and S. Lee (2010).

Figure 8: The performance of landslide models using Validation of an artificial neural network model for
statistical index-based evaluations landslide susceptibility mapping. Environmental Earth
Sciences, 60(3), pp 473-483.
4.4 Discussions
The results show that both landslide models perform Devkota, K.C., A.D. Regmi, H.R. Pourghasemi, K.
well for spatial prediction of landslides. Out of these, Yoshida, B. Pradhan, I.C. Ryu, M.R. Dhital and O.F.
the SVM model outperforms the NBT model. This Althuwaynee (2013). Landslide susceptibility mapping
corroborates the fact of early study that the SVM model using certainty factor, index of entropy and logistic
is one of the most efficient methods compared to regression models in GIS and their comparison at
conventional methods (Hassanien et al., 2015; Salama Mugling–Narayanghat road section in Nepal Himalaya.
et al., 2013; Tien Bui et al., 2012). It might be due to the Natural Hazards, 65(1), pp 135-165.
reason that NBT uses an assumption that all parameters
are independent which is not really true in landslide Dixon, B. and N. Candade (2008). Multispectral landuse
problems (Tien Bui et al., 2015). classification using neural networks and support vector
machines: One or the other, or both? International
In addition, it is also necessary to evaluate the predictive Journal of Remote Sensing, 29(4), pp 1185-1206.
capability of landslide affecting factors to landslide
models in order to select suitable parameters to Ermini, L., F. Catani and N. Casagli (2005). Artificial
construct dataset for model learning process (Pham et neural networks applied to landslide susceptibility
al. 2016b). It is because the performance of landslide assessment. Geomorphology, 66(1), pp 327-343.
models depends significantly on the quality of input
data that is constructed from landslide affecting factors Frye, C. (2007). About the geometrical interval
(Pradhan, 2013). In the present study, the predictive classification method.
capability of eleven landslide affecting factors has been https://blogs.esri.com/esri/arcgis/2007/10/18/about-
evaluated using the GR method which is one of the most the-geometrical-interval-classification-method/.
 78

Journal of Geomatics Vol 10 No. 1 April 2016

GSI (2009). Geological Survey of India. bayes, multilayer perceptron neural networks and
www.portal.gsi.gov.in. functional trees methods. Theor Appl Climatol, pp 1-19.

Guha-Sapir, D., P. Hoyois and R. Below (2014). Annual Pham, B.T., D. Tien Bui, MB. Dholakia, I. Prakash and
disaster statistical review 2013: The numbers and H.V. Pham (2016a). A comparative study of least square
trends. Brussels: CRED. support vector machines and multiclass alternating
decision trees for spatial prediction of rainfall-induced
Guzzetti, F., P. Reichenbach, M. Cardinali, M. Galli and landslides in a tropical cyclones area. Geotechnical and
F. Ardizzone (2005). Probabilistic landslide hazard Geological Engineering 34:1-18.
assessment at the basin scale. Geomorphology, 72(1–4),
pp 272-299. Pham, B.T., D. Tien Bui, I. Prakash and M.B. Dholakia
(2016b). Rotation forest fuzzy rule-based classifier
Hassanien, A.E., M.A. Salama, J. Platos and V. Snasel ensemble for spatial prediction of landslides using GIS.
(2015). Rough local transfer function for cardiac Natural Hazards:1-31.
disorders detection using heart sounds. Logic Journal of
IGP. Pradhan, B. (2013). A comparative study on the
predictive ability of the decision tree, support vector
Kantardzic, M. (2011). Data mining: Concepts, models, machine and neuro-fuzzy models in landslide
methods and algorithms. John Wiley & Sons. susceptibility mapping using GIS. Computers &
Geosciences, 51, pp 350-365.
Karegowda, A.G., A. Manjunath and M. Jayaram
(2010). Comparative study of attribute selection using Quinlan, J.R. (1986). Induction of decision trees.
gain ratio and correlation based feature selection. Machine Learning, 1(1), 81-106.
International Journal of Information Technology and
Knowledge Management, 2(2), pp 271-277. Salama, M.A., O.S. Soliman, I. Maglogiannis, A.E.
Hassanien and A.A. Fahmy (2013). Rough set-based
Kavzoglu, T., E.K. Sahin and I. Colkesen (2014). identification of heart valve diseases using heart sounds.
Landslide susceptibility mapping using GIS-based Rough Sets and Intelligent Systems-Professor Zdzisław
multi-criteria decision analysis, support vector Pawlak in Memoriam. Springer, pp 475-491.
machines and logistic regression. Landslides, 11(3), pp
425-439. Sarkar, S., D.P. Kanungo and G.S. Mehrotra (1995).
Landslide hazard zonation: A case study in Garhwal
Kohavi, R., 1996. Scaling Up the Accuracy of Naive- Himalaya, India. Mountain Research and Development,
Bayes Classifiers: A Decision-Tree Hybrid, KDD. 15(4), pp 301-309.
Citeseer, pp. 202-207.
Sengupta, A., S. Gupta and K. Anbarasu (2010).
Lucà, F., M. Conforti and G. Robustelli (2011). Rainfall thresholds for the initiation of landslide at
Comparison of GIS-based gullying susceptibility Lanta Khola in north Sikkim, India. Natural hazards,
mapping using bivariate and multivariate statistics. 52(1), pp 31-42.
Northern Calabria, South Italy, Geomorphology,
134(3–4), pp 297-308. Song, Y., J. Gong, S. Gao, D. Wang, T. Cui, Y. Li and
B. Wei (2012). Susceptibility assessment of earthquake-
NCEP (2014). Global weather data for SWAT. induced landslides using Bayesian network: A case
http://globalweather.tamu.edu/home. study in Beichuan, China. Computers & Geosciences,
42(0), pp 189-199.
Onagh, M., V. Kumra and P.K. Rai (2012). Landslide
susceptibility mapping in a part of Uttarkashi district Soni, J., U. Ansari, D. Sharma and S. Soni (2011).
(India) by multiple linear regression method. Predictive data mining for medical diagnosis: An
International Journal of Geology, Earth and overview of heart disease prediction. International
Environmental Sciences. Journal of Computer Applications, 17(8), pp 43-48.

Pham, B.T., D. Tien Bui, P. Indra and M.B. Dholakia Tien Bui, D., B. Pradhan, O. Lofman and I. Revhaug
(2015a). Landslide susceptibility assessment at a part of (2012). Landslide susceptibility assessment in Vietnam
Uttarakhand Himalaya, India using GIS – based using support vector machines, decision tree and naive
statistical approach of frequency ratio method. Bayes models. Mathematical Problems in Engineering.
International Journal of Engineering Research &
Technology 4(11), pp 338-344. Tien Bui, D., T.A. Tuan, H. Klempe, B. Pradhan and I.
Revhaug (2015). Spatial prediction models for shallow
Pham, B.T., D. Tien Bui, H.R. Pourghasemi, P. Indra landslide hazards: A comparative assessment of the
and M.B. Dholakia (2015b). Landslide susceptibility efficacy of support vector machines, artificial neural
assesssment in the Uttarakhand area (India) using GIS: networks, kernel logistic regression and logistic model
A comparison study of prediction capability of naïve tree. Landslides, pp 1-18.
 79

Journal of Geomatics Vol 10 No. 1 April 2016

Vapnik, V.N. (1995). The nature of statistical learning
theory. Springer Verlag, New York.

Wang, W.D., C.M. Xie and X.G. Du (2009). Landslides

susceptibility mapping based on geographical
information system, GuiZhou, south-west China.
Environmental geology, 58(1), pp 33-43.

Yeon, Y.-K., J.-G. Han and K.H. Ryu (2010). Landslide

susceptibility mapping in Injae, Korea, using a decision
tree. Engineering Geology, 116(3), pp 274-283.

Zare, M., H.R. Pourghasemi, M. Vafakhah and B.

Pradhan (2013). Landslide susceptibility mapping at
Vaz Watershed (Iran) using an artificial neural network
model: A comparison between multilayer perceptron
(MLP) and radial basic function (RBF) algorithms.
Arabian Journal of Geosciences, 6(8), pp 2873-2888.

Zhao, Y. and Y. Zhang (2008). Comparison of decision

tree methods for finding active objects. Advances in
Space Research, 41(12), pp 1955-1959.
 80

Journal of Geomatics Vol 10 No. 1 April 2016

Early estimation of crop sown area by integrating multi-source data
D. Ram Rajak1,2, Rajesh K. Jain2 and Shibendu S. Ray3
1
Space Applications Centre, ISRO, Ahmedabad – 380015, India
2
Institute of Management, NIRMA University, Ahmedabad-382481, India
3
Mahalanobis National Crop Forecasting Centre, MOA, New Delhi-110049, India
Email: [email protected]

(Received: Feb 01, 2016; in final form: Mar 30, 2016)

Abstract: Satellite based remote sensing (RS) data at different spatial and temporal scales can provide crop sown area
estimates needed by decision makers. Due to conflict of spatial resolution versus temporal frequency of data collection,
getting early and accurate crop sown area at large scale is very difficult. This paper presents a methodology of early
estimation of crop sown area at large scale by making use of high temporal coarse spatial resolution data and low temporal
fine spatial resolution data. It also uses previous years’ data for extracting a-priori knowledge of crop sowing area. Early
crop area estimate was made for Gujarat state (India) for 2011-12 rabi season. Multi-date MODIS (MODerate resolution
Imaging Spectroradiometer) data and two-date Resourcesat-2 AWiFS (Advance Wide Field Sensor) data upto mid-
December were used for crop sown area early estimates. Multi-date MODIS data for previous five years provided a-priori
information on crop presence / absence over the previous five crop seasons. While ISODATA was used for classifying
multi-date MODIS and AWiFS data; hierarchical decision tree approach was used for integrating multi-source
information. Incorporating two date AWiFS data and a-priori information with multi-date MODIS data increased crop
sown area early estimates accuracy significantly.

Keywords: Crop sown area, MODIS, AWiFS, ISODATA, Decision tree

1. Introduction based observations) in India. Parihar and Oza (2006)

have described the concept and program details. While
Traditionally, the crop sown area is estimated from the Xiangming et al. (2006) have used multi-date MODIS
sample data collected by different government (MODerate resolution Imaging Spectroradiometer) data
institutes/departments or survey agencies. These for mapping paddy rice in South and South-East Asia.
methods are time consuming and labour intensive. Wardlow et al. (2007) have used MODIS NDVI time
Human subjectivity and biases increase in-accuracy series data for crop classification in U. S. Central Great
levels of these estimates. Satellite based remote sensing Plains.
(RS) data is one of the few sources for deriving frequent
and reliable estimates of the crop sown area at different Nigam et al. (2009, 2011) used atmospherically
spatial scales. A number of research studies and corrected surface reflectance values in Red and NIR
projects have demonstrated the usefulness of RS data in bands of INSAT 3A CCD to compute NDVI at
making crop inventory over different parts of the world continental scale and these NDVI values were further
(Mc-Donald and Hall, 1980; Sharman, 1993; De Roover validated with global product to judge its spatio-
et al., 1993; Navalgund et al., 1991; Oza et al., 1996; temporal profiles and its range over different natural
Dadhwal et al., 2002; Parihar and Oza, 2006; targets. Vyas et al. (2013) demonstrated that the
Xiangming et al., 2006; Wardlow et al., 2007; Nigam et spatially distributed crop sowing dates derived using
al., 2009; Nigam et al., 2012; Vyas et al., 2013; Ray et INSAT 3A CCD data at 1 km×1 km resolution could be
al., 2014; Sharma et al., 2014; Parihar, 2016; Karam et captured at regional scale in India. Ray et al. (2014)
al., 2016; Rajak and Jain, 2016). Large Area Crop have summarized an operational methodology of
Inventory Experiment (LACIE, 1974-1977), covering multiple forecasting of multiple crops in India. Sharma
USA, USSR, Brazil, Argentina, India etc. was one such et al. (2014) concluded that multi-year multi-date
study (Mc-Donald and Hall, 1980). Europe-wide crop MODIS data could be used for monitoring gross annual
production estimation was carried out by European changes of major rabi crops at regional scale.
Union under Monitoring Agriculture with Remote
Sensing (MARS) project (Sharman, 1993; De Roover et Sud et al. (2015) presented a critical review of the
al., 1993). In India, Crop Acreage and Production literature related to concepts in crop area estimation and
Estimation (CAPE) project demonstrated the potential crop yield assessment. In addition, country-experiences
of remotely sensed data for crop inventory (Navalgund were also reported while bringing out several issues and
et al., 1991). Dadhwal et al. (2002) have reviewed problems with regard to crop area and crop yield
Indian experience of crop inventory using RS data. A estimation. Karam et al. (2016) proposed a management
multiple crops acreage estimation and production tool for annual inventory and monitoring of cultivated
forecasting program based on multiple inputs was lands using RapidEye and Landsat ETM+ imagery over
developed under FASAL (Forecasting Agricultural a test area in Bekaa Valley, Lebanon. The study
output using Space, Agro-meteorological and Land concluded that satellite imagery was essential for the

© Indian Society of Geomatics

 81

Journal of Geomatics Vol 10 No. 1 April 2016

definition of the existing cropping patterns in the pilot and harvested by September-October) include paddy
area and it helped in better estimation of seasonal (rice), cotton, groundnut, castor, jowar, bajra, tobacco,
irrigation needs at the scheme level. Parihar (2016) arhar, maize, sugarcane etc. Cotton, castor, tobacco, and
detailed the sequential developments in the use of single sugarcane are the crops which continue beyond kharif
and multi-date optical and microwave remote sensing season and usually available in the field during
data for crop production forecasting in India. A November – December period. Wheat, rapeseed &
methodology was developed by Nigam et al. (2015) mustard, cumin, potato etc. are the major crops grown
using high temporal vegetation index data at 1km spatial during rabi season (crops sown during winter and
resolution available from Indian geostationary satellite harvested before summer) in the state.
(INSAT 3A) to monitor progress of rabi crop area at
country scale. The rabi crop area estimates obtained at The satellite data used in this study include multi-date
the end of crop season at country level showed mean MODIS reflectance data over Gujarat for 2006-2007 to
deviation of −18.1% with respect to reported DAC 2011-2012 rabi seasons and multi-date AWiFS data of
statistics during rabi season 2011–2012. At national full 2011-12 season (full season data was used for
scale, the INSAT- estimated rabi crop area showed obtaining the classified image that was used as reference
16.36% deviation from AWiFS derived rabi area at 2 classified image). MODIS is a sensor aboard the
km × 2 km grid, but no attempt was made to estimate NASA’s Terra and Aqua satellites. The sensor views the
crop sown area at state level. Sharma et al. (2014) entire Earth's surface every 1 to 2 days, acquiring data
estimated rabi sown area for Gujarat state using full in 36 spectral bands at different spatial resolutions.
season multi-date MODIS data but no attempt was made Terra MODIS surface reflectance 8-day L3 global 250m
to arrive at an early estimation of the crop sown area. (MOD09Q1) data were downloaded from the website:
Rajak and Jain (2016) have emphasized the importance https://lpdaac.usgs.gov. The data used in the present
of two-source data i.e. AWiFS (Advance Wide Field study were from Julian day 281 (9th October) to Julian
Sensor) and LISS-III (Linear Imaging Self-scanning day 89 (30th March) in subsequent year for six rabi
Sensor - 3) to improve crop acreage estimation at district seasons. Thus, for 2010-11 rabi season, the time series
scale. contained 23 NDVI composite images, prepared from
surface reflectance data. Overall, 115 MODIS images
There is no doubt that multi-source data analysis will (23 images x 5 seasons) were processed and analysed
become increasingly widespread in the future, due to over 5 years (2006-07 to 2010-11) for getting multi-year
increasing ease and lower cost of data collection, crop a-priori information. MODIS derived 9-dates
storage and manipulation. Availability of multi-source NDVI time series upto mid-December was used for
data of the same object at different times provides 2011-12 crop sown area early estimation. Two sets of
information on time varying characteristics of the Resourcesat-2 AWiFS data (Path/Row 92/56) for 2011-
object, if proper tools are available to analyse the multi- 12 season were prepared. Set 1 having AWiFS data
source data. In case of satellite based multi-source data, limited to mid-December and Set 2 having full rabi
there are a number of commercial image processing season data from October to March were analysed
software that may be used for storage, archival, and separately. While Set 1 comprised of November 17 and
visualization of multi-source data. Extraction of the December 11, 2011 data, there were 7 dates of cloud-
required information from such a multi-source dataset free data in Set 2. The sensor characteristics of AWiFS
has remained a challenging task for long. Integration of are given in Table 1.
information available from the previous years’ data with
current year’s multi-source RS data has not been Table 1: Characteristics of AWiFS onboard
explored adequately for early estimation of crop sown Resourcesat-2
area.
Sr Characteristic Value
In this study, a methodology developed for early No
estimation of crop sown area over Gujarat state (India) 1 Spectral band: Green (B2) 0.52 – 0.59 µm
using current year’s multi-sensor data along with the 2 Spectral band: Red (B3) 0.62 – 0.68 µm
information extracted from previous 5 years’ RS data is 3 Spectral band: NIR (B4) 0.77 – 0.86 µm
presented. The methodology uses multi-source 4 Spectral band: SWIR (B5) 1.55 – 1.70 µm
information i.e. multi-date satellite data from MODIS, 5 Spatial Resolution 56 m
two date satellite data from AWiFS, and a-priori crop 6 Swath 740 km
history derived from multi-year multi-date MODIS 7 Revisit period 5 days
data. 8 Quantisation 12 bit
(Source: Bhuvan, 2012)
2. Study area and data used
Crop and other land use / land cover details collected in
This study was conducted over Gujarat, a western state
field during crop season 2011-12 were used for training
of India with geographic area of 19.6 Million ha.
signature generation and validation in image
Gujarat is an agriculturally rich state with a large classification. A sample of major crop parameters
number of crops grown in mainly rabi and kharif crop collected during the field visit is shown in Table 2.
seasons. The major crops sown in kharif season (crops
associated with the monsoon, mostly sown in June-July
 82

Journal of Geomatics Vol 10 No. 1 April 2016

Table 2: A sample of major crop parameters (NIR-R) to the sum of spectral values (NIR+R). Its
collected during the field visit in 2011-12 theoretical value ranges from -1.0 to +1.0. To avoid
working with real/float numbers and to store in integer
Field ID 20111213-02 - 20111212-07 - 20111213-05 formats, it is sometimes scaled by multiplying and
13 Dec - 12 Dec - 13 Dec adding some constant values. In this case, the real values
Date
2011 2011 2011 were multiplied by 100 and then 100 was added to get
Lat. (N) 23.872 - 23.685 - 23.774 the scaled NDVI values.
Long. (E) 71.990 - 72.238 - 71.989
Scaled NDVI = 100 + 100 * (NIR-R) / (NIR+R).
Crop Type Wheat - Mustard - Castor
CRI - Flowering - Capsule So the theoretical range of scaled NDVI became 0 to
Crop Stage
Formation 200, hence the values were stored in single byte format.
Field Size (m) 100x100 - 100 x 100 - 200x200
Synthetic 200x200 - 200 x 200 - - Screening & smoothening of MODIS time series
Field Size (m) data: An algorithm, named Harmonic ANalysis of
Crop 50 - 50 - 70 Time Series (HANTS) was employed for detecting the
Fraction (%) cloud contaminated pixels in MODIS data and for
Date of November - Oct III - N/A temporally interpolating the remaining non-
Sowing III week (Kharif) contaminated data to construct gapless images at a
- - - - - - prescribed interval of 8 days. The HANTS algorithm
was devised and developed at NLR (Nationaal Lucht-
en Ruimtevaartlaboratorium), in collaboration with
3. Methodology
KNMI (Koninklijk Nederlands Meteorologisch
Instituut) and Alterra (Menenti et al., 1993; Verhoef et
This study has three major data analysis components,
al., 1996; Roerink et al., 2000) the Netherlands. Its
namely analysis of multi-date and multi-year MODIS
applications have been successfully demonstrated for
data, analysis of multi-date AWiFS data and integration
vegetation monitoring, land surface temperature studies
of multi-source parameters for early crop sown area
and generating cloud-free weather images (Wen et al.,
estimation.
2004; Xu and Shen, 2013; Sharma et al., 2014). The
HANTS software can be downloaded free from the
3.1 Processing of MODIS and AWiFS data
Naval Research Laboratory (NRL) website
MODIS MOD09Q1 data were converted from HDF-
http://www.nlr.org/space/earth-observation/ or
EOS format to ERDAS raster image format for further
http://gdsc.nlr.nl/gdsc/en/tools/hants.. It can be
image processing through ERDAS IMAGINE software.
implemented through MATLAB also (Abouali, 2012).
MODIS surface reflectance values at pixel level were
obtained by multiplying 16-bit unsigned data values
The HANTS algorithm can provide a much better
with a factor of 0.0001 (Vermote et al., 2011).
smoothing of time series than most other methods can
offer. Further details on HANTS can be found
In case of AWiFS data the Digital Numbers (DN) stored
elsewhere (see Roerink et al., 2000; Wen et al., 2004;
in the original data were converted to spectral radiance
Verhoef et al., 2005). The basic mechanism is to
by using saturation radiance value available in data
calculate a Fourier series to the data, identify and
header file.
remove outliers and replace them with the value
produced by the Fourier series. This process is
Accurate co-registration: Spatial characteristics of any
controlled by five parameters, which have to be set at
land use / land cover including crops derived from
the beginning of each HANTS run. The HANTS control
different sources need to have common geo-referencing
parameters used are shown in Table 3.
so that they can be integrated. It requires very accurate
geo-referencing of the multi-source data. In this study,
Examples of thin cloudy data normalisation and thick
the multi-date MODIS data at 250m spatial resolution,
cloudy data interpolation using HANTS technique are
which are well co-registered among each other, were co-
shown in Fig. 1.
registered with Resourcesat-2 AWiFS data at 56m
spatial resolution. Both the datasets were brought to a
common projection system of Albers Conical Equal 3.2 Extracting data specific crop patterns
Data specific pattern analysis is an important element of
Area (ACEA) projection and WGS84 datum.
multisource data mining. The idea is to mine each data
type separately to get data patterns using appropriate
NDVI and scaled NDVI calculation: An index derived
mining techniques instead of combining all the data into
from spectral values in Near Infra-Red (NIR) and Red
one huge dataset before mining. Then assemble the data
(R) bands of electromagnetic spectrum, called
specific attributes (information) derived from all the
Normalized Difference Vegetation Index (NDVI) is
data types separately and perform integrated analysis on
extensively used for studying vegetation using RS data.
them. This approach promotes data specific knowledge
It responds to changes in the amount of green biomass,
discovery at each data source independently to
chlorophyll content and canopy water stress. NDVI is
maximize utility of data specific information. In this
defined as ratio of the difference of spectral values
study, based on the field knowledge, rabi crop locations
 83

Journal of Geomatics Vol 10 No. 1 April 2016

were identified on the images and crop specific then matched with the reference patterns and classified
temporal NDVI patterns were derived from multi-date into different classes based on their temporal profiles.
MODIS (2006-07 to 2010-11) as well as multi-date
AWiFS data (2011-12, reference dataset). RS Data

(a) (c)
Sig. Ext. ISODATA

Mask
GT

Spectral
Pattern
(b) (d) Ref. Matching Spectral
Sign. Clusters

Rationale –
Matching of Temporal Spectral
Profiles of Clusters With
Figure 1: Examples of Harmonic ANalysis of Time Reference Temporal Spectral
Series (HANTS) applications: (a) scaled NDVI Profiles Prepared from GT
profile of a rabi crop with thin cloudy atmosphere, Classified Image Crop Layer
(b) scaled NDVI profile of another rabi crop with Figure 2: Conceptual flow of ISODATA clustering
missing data due to thick cloudy atmosphere; (c) and spectral pattern matching showing major data
after applying HANTS over profiles in (a); and (d) analysis components
after applying HANTS over profiles shown in (b)
Multi-date full-season MODIS NDVI time series data
3.3 ISODATA Classification from 2006-07 to 2010-11 were subjected to data
The Iterative Self-Organizing Data Analysis Technique screening and extrapolation before classification. The
(ISODATA) was used as an unsupervised classifier by HANTS corrected data were then classified using
recognizing multi-temporal patterns in the dataset (Ball ISODATA clustering. The temporal spectral profiles of
and Hall, 1965). It is an iterative and heuristic procedure ISODATA clusters were matched with the reference
which assigns first an arbitrary initial cluster vector temporal spectral profiles and the clusters were labelled
based on user’s input. In the second step, it classifies to rabi-crop and other than rabi classes. Thus classified
each pixel of the data to the closest cluster in spectral images were obtained for 5 rabi seasons i.e. 2006-07 to
domain. Merging and splitting of clusters is done, if 2010-11.
conditions are met. Clusters are merged if either the
number of pixels in a cluster is less than a certain Two sets of multi-date AWiFS data, as described
threshold or if the centres of two clusters are closer than earlier, were processed to prepare classified images. Set
a certain threshold. Similarly, clusters are split into two 1 was classified by subjecting 7 dates of NDVI stack to
different clusters if the cluster standard deviation ISODATA clustering and then labelling of clusters to
exceeds a predefined/user defined value and the number rabi-crop and other classes. In case of Set 2, firstly
of pixels is twice the threshold for the minimum number December 11 AWiFS data was classified by ISODATA
of pixels in a cluster. In the third step the new cluster in vegetation (crops, forest, shrubs, plantations etc.) and
mean vectors are calculated based on all the pixels in other classes and ΔNDVI (i.e. NDVIDec11 – NDVINov11)
that cluster. The second and third steps are repeated until image was used for classifying vegetation class in 4 sub-
the "change" between two consecutive iterations is classes based on NDVI gradient.
small. ISODATA has been found very effective at
identifying spectral clusters in data. It is especially very 3.4 Integration by decision tree approach
useful while analysing a new data as we don’t need to Decision tree approach is a commonly used method of
know much about the data beforehand. Care has to be information extraction in data mining. The objective is
taken that the data is structured well otherwise to create a model that estimates the value of a target
ISODATA may take long time if data is largely variable based on several input variables available from
unstructured. Conceptual flow of the procedure single or multiple data sources. A decision tree is a flow-
followed in ISODATA clustering and spectral matching chart-like structure, where each internal (non-leaf) node
is shown in Fig. 2. denotes a test on an attribute, each branch represents the
outcome of a test, and each leaf (or terminal) node holds
Before classifying 9-dates MODIS NDVI time-series a class label. The topmost node in a tree is the root node
data of 2011-12, temporal patterns of reference crops (Quinlan, 1993).
were obtained from 9-dates NDVI values at the crop
locations collected during the in-season field surveys. In this study three input images namely (i) a-priori crop
Stacked 9-dates NDVI data was subjected to ISODATA history image from 5-years multi-date MODIS data, (ii)
and clusters were obtained for 2011-12 season. The 9-dates MODIS derived classified image, (iii) 2-dates
temporal NDVI patterns of the ISODATA clusters were
 84

Journal of Geomatics Vol 10 No. 1 April 2016

AWiFS derived classified images were integrated to
yield an output image.

The a-priori rabi crop history image was prepared by

integrating 5 classified images of 2006-07 to 2010-11
seasons and the crop pixels were grouped in following
5 classes:
 Apr1: pixels with crop all the 5 years.
 Apr2: pixels with crop in 2010-11 and any 3
seasons.
 Apr3: pixels with crop in 2010-11 and any 2
seasons.
 Apr4: pixels with crop in 2010-11 and any 1 season.
 Apr5: pixels with crop in 2010-11 only.

Decreasing weightages were given to these classes,

highest to Apr1 and lowest to Apr5, while this image
was integrated for final classification of crop sown area
for 2011-12 rabi season. Figure 3: Scaled NDVI images obtained from
MODIS reflectance data over Gujarat from October
4. Results and Discussion 8, 2011 to December 11, 2011 (2011-12 rabi season)

The MODIS / Terra MOD09Q1 and Resourcesat-2

AWiFS image sub-sets for Gujarat state were extracted
from the original data from October to mid-December
2011. While MODIS two bands i.e. red (620 – 670 nm)
and near infrared (841 – 871 nm) data were used, all the
four bands data from AWiFS (Green: 0.52-0.59µm,
Red: 0.62-0.68µm, NIR: 0.77-0.86µm, and SWIR:1.55-
1.70µm) data were used in this study.

The MODIS reflectance images were used to derive

scaled NDVI for all the nine 8-day interval dates (8, 16,
24 October, 1, 9, 17, 25 November, 3 and 11 December,
2011) of 2011-12 rabi season. Scaled NDVI images
obtained from MODIS reflectance data are shown in
Fig.3. The temporal variations of forest cover NDVI are
clearly visible from October to December images. The
spectral contrast between forest regions (for example
Gir Forest and Dangs Forest) and their surroundings has
continuously decreased from October to December. It is
because of decrease in Forest NDVI and increase in crop Figure 4: Colour Composite prepared from 2 dates
NDVI during this time period. AWiFS data (Red: ΔNDVI, Green: red band of Dec
11, Blue: NIR band of Dec 11). The regions with high
In case of Resourcesat-2 AWiFS Set 1 data, cloud free ΔNDVI are red and pink in the image. The rabi crop
images were available for 2 dates (November 17 and area are visible in pink colour
December 11) over the selected study period of October
to mid-December. Colour Composite prepared from 2 A subset of the MODIS / Terra MOD09Q1 8-day
dates data (Red: ΔNDVI, Green: red band of Dec 11, reflectance data for Red and NIR bands was created for
Blue: NIR band of Dec 11) is shown in Fig. 4. In case all the 5 seasons (2006-07 to 2010-11). Multi-date
of AWiFS Set 2 data, similar methodology was NDVI images were prepared from the reflectance
followed and the NDVI time series was classified to rabi images. The HANTS corrected NDVI stacked images
crop classified image using ISODATA clustering. As were subjected to ISODATA clustering for each crop
this dataset contained full rabi season data, the crop season. Based on the temporal spectral profiles shape
temporal spectral patterns of major crops were well matching with the reference profiles, unknown clusters
discriminated from each other and this image was were classified to different classes. Classes were
considered as reference classified image. Typical merged to form rabi crop and other-than-rabi-crop
temporal NDVI patterns of three major rabi crops of theme images. The visual profile matching was carried
Gujarat are shown in Fig. 5. out based on the overall NDVI pattern, peak value, time
of peak, duration of peak value, growth gradient, decay
gradient etc.
 85

Journal of Geomatics Vol 10 No. 1 April 2016

with higher gradient values of both the ΔNDVIs (see
Level 2, Table 4).

Table 4: The hierarchical decision rules applied for

obtaining the classified image by integrating
multisource data

Level MODIS AWiFS A-priori Output,

9-Dates 2-Dates (5- if YES
years)
Non-rabi- Non - Non
crop class vegetation Rabi
Figure 5: Typical temporal NDVI patterns of three 1
class or crop
major rabi crops and a kharif crop of Gujarat ΔNDVI< 0
derived from multi-date AWiFS data Rabi crop Veg. class Crop at Rabi
with with least 1 crop 1
2
These five classified images for 2006-07 to 2010-11 ΔNDVI>0.10 ΔNDVI> year
seasons were used to create a-priori rabi crop history 0.05
image for 2011-12 season. The a-priori rabi crop image Rabi crop Veg. class Crop at Rabi
is shown in Fig. 6. 3 with with least 2 crop 2
ΔNDVI>0.10 ΔNDVI> 0 years
Rabi crop Veg. class Crop at Rabi
with with least 3 crop 3
4
ΔNDVI>0.05 ΔNDVI> years
0.05
Rabi crop Veg. class Crop at Rabi
5 with with least 4 crop 4
ΔNDVI>0.05 ΔNDVI> 0 years
Rabi crop Vegetation -
Non
6 class Rabi
crop
Note: In case of MODIS 9-Dates, ΔNDVI = (mean
NDVI of 7,8,9 – mean NDVI of 4,5,6) and in case of
AWiFS 2-Dates, ΔNDVI = (NDVI Dec 11 – NDVI Nov
17).

While the crop sown area estimated from 9-dates

MODIS data was found to be 2.370 million hectares
Figure 6: The a-priori rabi crop image prepared (0.913 Mha with NDVI>0.1 and 1.457 Mha with
from 5-year multi-date MODIS data. While Apr1 NDVI>0.05); the integrated dataset yielded rabi crop
represent the pixels with rabi crop for all the 5 years, sown area of 1.125 million hectares. This shows that the
Apr5 represents rabi crop during the previous classification based on only MODIS data overestimated
season i.e. 2010-11 the rabi sown area; as rabi crop area at the end of crop
season was reported to be 2.045 million hectares (DES,
The crop sown area image for 2011-12 was obtained by 2016). The rabi crop sown area as on mid-December
integrating information extracted from 9-dates of 2011 was reported to be 1.238 million hectares (DES,
MODIS images, 2-dates of AWiFS images and the a- 2016), indicating that the area estimated by the proposed
priori rabi crop history image prepared from 5-years of methodology is almost same as reported. However, it is
multi-date MODIS images. The hierarchical decision not expected that the RS data upto mid-December
rules were applied for obtaining this classified image should match the area estimated reported by mid-
and are given in Table 4. The pixels belonging to the December. It is expected that the data upto mid-
fields where crop was grown during all the 5 rabi December may pick-up the crop sown at most upto
seasons were assigned a-priori category Apr1. November end. This indicates that the proposed
Similarly, Apr2 to Apr5 categories were assigned as methodology is slightly overestimating the crop sown
defined in Section 3.4. While forming decision rules for area. The degree of over estimation could not be
integration, Apr1 and Apr2 (crop at least 4 years) were quantified as November-end crop sown area estimates
given the highest weightage. The pixels belonging to are not available from any source. The classified image
Apr1 & Apr2 were classified to rabi crop class although obtained through the proposed methodology was
ΔNDVI (AWiFS) was just greater than 0.0 with ΔNDVI checked for its accuracy in classifying the rabi crops for
(MODIS) was just greater than 0.05 (see Level 5, Table 2011-12. It was compared with the reference crop sown
4). However, Apr5 was given the least weightage as the area image obtained using full rabi season AWiFS data.
pixels belonging to it were classified to rabi crop only
 86

Journal of Geomatics Vol 10 No. 1 April 2016

It was found that crop sown area images upto mid- References
December could pick up almost 57% of full season rabi
crop area. Sharma et al. (2014) estimated the rabi sown Abouali, M. (2012). MATLAB Implementation of
crop area of 1.798 million hectare using full season Harmonic ANalysis of Time Series (HANTS).
MODIS data, but early estimation was not attempted. http://www.mathworks.com/matlabcentral/fileexchang
e/38841-matlab-implementation-of-harmonic-analysis-
Government of India (Department of Agriculture and of-time-series--hants-/.
Farmers’ Welfare) estimated that almost 60% of full
season rabi crop was sown as on December 16, 2011. Ball, G.H. and D.J. Hall (1965). ISODATA: A method
The mapping accuracy of rabi sown area was found to of data analysis and pattern classification. Stanford
be 88.6% with respect to the reference full season Research Institute, Menlo Park, United States. Office of
AWiFS image based rabi sown area. The full season Naval Research.Information Sciences Branch.
crop sown area determined from the reference classified
image was 1.972 million hectares. While the Bhuvan (2012). Resourcesat-2 Handbook.
Government of India (DES, 2016) estimates were 1.238 http://bhuvan.nrsc.gov.in/bhuvan/PDF/ Resourcesat-
and 2.045 million hectares as on December 16, 2011 and 2_Handbook.pdf. Accessed on Jan 28, 2016.
February 24, 2012, respectively.
Dadhwal, V.K., R.P. Singh, S. Dutta and J.S. Parihar
5. Conclusion (2002). Remote sensing based crop inventory: A review
of Indian experience. Tropical Ecology, 43(1): 107-122.
A technique for early estimating crop sown area over a
large region has been developed using multi-source De Roover, B., S.C. De Mudker and R. Goossens
remote sensing data. Crop sown area of Gujarat state (1993). The regional inventory (MARS) in Belgium.
(India) for 2011-12 rabi season was estimated by mid- Proc. Int. Symp. “Operationalization of Remote
December. The multi-source data used in this study Sensing”, 19-23 April, ITC, Enschede, The
included in-season MODIS 8-day composite, and two Netherlands.
dates AWiFS data along with multi-year, multi-date
MODIS derived a-priori crop image. Initial crop growth DES (2016). Personal communication. Crops and
trends in terms of NDVI values were obtained from NFSM division, Department of agriculture, cooperation
multi-date MODIS data from October to mid-December and farmers’ welfare, Ministry of agriculture and
2011. ISODATA clustering algorithm was used to farmers’ welfare, Government of India.
cluster similar trends of temporal NDVI patterns.
Spatial distribution of crop and non-crop fields/clusters Karam, Fadi, Nabil Amacha, Wassim Katerji,
was obtained from two dates AWiFS data (November Weicheng Wu, Alfonso Domínguez and Safa Baydoun
17 and December 11). Integration of the temporal (2016). Using remote sensing to improve crop water
patterns derived from MODIS data, spatial clusters allocation in a scarce water resources environment.
derived from AWiFS data, and a-priori image derived International Journal of Science and Research, 5(1):
from multi-year MODIS data was carried out using 1481-1495. ISSN (Online): 2319-7064.
hierarchical decision tree approach. Reference crop
sown area was obtained by using full crop season multi- Mc-Donald, R.B. and F.G. Hall (1980). Global crop
date AWiFS data for 2011-12. It was found that the forecasting. Science, 280: 670-679.
integration of in-season multi-source data provided the
crop sown area estimates closer to the reference Menenti, M., S. Azzali, A. de Vries, D. Fuller and S.
estimates. Prince (1993). Vegetation monitoring in Southern
Africa using temporal Fourier analysis of
Acknowledgements AVHRR/NDVI observations. Proc. Int. Symp. Rem.
Sens. in Arid and Semi-Arid Regions, Lanzhou, China,
This study was carried out as a part of the Ph. D. 1993, 287-294.
research activity undertaken by the first author. He
thanks Director, Space Applications Centre, Navalgund, R.R., J.S. Parihar, Ajai and P.P. Nageshwar
Ahmedabad for the motivation he provided. He Rao (1991). Crop inventory using remotely sensed data.
expresses thanks to Project Director – FASAL for Current Science, 61: 162–171.
allowing him to access FASAL project RS data for this
research. He also thanks Dr. J. S. Parihar for co-guiding Nigam, R., B.K. Bhattacharya, K.R. Gunjal, N.
his Ph.D. study. The constant encouragement and Padmanabhan and N.K. Patel (2009). A continental
technical guidance provided by Dr. Markand Oza is scale vegetation index from Indian geostationary
highly appreciated and acknowledged. The MODIS data satellite. ISPRS Archives XXXVIII-8/W3 Workshop
MOD09Q1 support from NASA EOSDIS LP DAAC Proceedings: Impact of Climate Change on Agriculture.
(https://lpdaac.usgs.gov) is acknowledged. http://www.isprs.org/proceedings/XXXVIII/8-
W3/B5/3-22_F.pdf.

Nigam, R., B.K. Bhattacharya, K.R. Gunjal, N.

Padmanabhan and N.K. Patel (2011). Continental scale
 87

Journal of Geomatics Vol 10 No. 1 April 2016

vegetation index from Indian geostationary satellite: Quinlan, J.R. (1986). Induction of decision trees.
algorithm definition and validation. Current Science, Machine Learning, 1:81–106.
100(8):1184-1192.
Rajak, D.R., H.A. Patel, K.N. Chaudhari, N.K. Patel, S.
Nigam, R., B.K. Bhattacharya, K.R. Gunjal, N. Panigrahy and J.S. Parihar (2010). Spatial temporal
Padmanabhan, and N.K. Patel (2012). Formulation of sowing pattern of rapeseed-mustard crop in India using
time series vegetation index from Indian geostationary multi-date IRS AWiFS data. ISPRS Archives
satellite and comparison with global product. Journal of XXXVIII- 8/W20; Workshop Proceedings: Earth
the Indian Society of Remote Sensing, 40:1–9. Observation for Terrestrial Ecosystems.http://www.int-
arch-photogramm-remote-sens-spatial-inf-
Nigam, R., S. Vyas, B.K. Bhattacharya, M.P. Oza, S.S. sci.net/XXXVIII-8-W20/56/2011/isprsarchives-
Srivastava, N. Bhagia, D. Dhar and K.R. Manjunath XXXVIII-8-W20-56-2011.pdf.
(2015): Modeling temporal growth profile of vegetation
index from Indian geostationary satellite for assessing Rajak, D.R. and Jain R.K. (2016). Toward improved
in-season progress of crop area, GIScience & Remote crop acreage estimation using multi-source information.
Sensing, DOI: 10.1080/15481603.2015.1073036. 19th Nirma International Conference on Management
http://dx.doi.org/10.1080/15481603.2015.1073036 (NICOM-2016), Macro and Micro Dynamics for
Empowering Trade, Industry and Society. Excel India
Oza, M.P., N. Bhagia, S. Dutta, J.H. Patel and V.K. Publishers, New Delhi. ISBN: 978-93-85777-07-3,
Dadhwal (1996). National wheat acreage estimation for Page: 82,92.
1995-96 using multi-date IRS-1C WiFS data. J. Indian
Soc. Remote Sensing, 24(4): 243-254. Ray, S.S., S. Neetu, Mamatha and S. Gupta (2014). Use
of remote sensing in crop forecasting and assessment of
Oza, M.P., N. Bhagia, D.R. Rajak and V.K. Dadhwal impact of natural disasters: operational approaches in
(2002). All India wheat inventory using multidate IRS India.http://www.fao.org/fileadmin/templates/rap/files/
WiFS and weather data. Proceedings of the ISPRS Project/Expert_Meeting_17Feb2014_/P2-
Comm. VII Symposium “Resource and Environmental 3_Remote_Sensing_for_Crop_Monitoring.pdf.
Monitoring”, Dec. 3-6 2002, Hyderabad, India, IAPRS
and GIS, 34(7): 338-343. Roerink, G.J., M. Menenti and W. Verhoef (2000).
Reconstructing cloudfree NDVI composites using
Parihar, J.S., S. Panigrahy, N.K. Patel, D.R. Rajak, H.A. Fourier analysis of time series, Int. J. Remote Sensing,
Patel and G.N. Patel (2010). Methodology for multiple 21(9): 1911–1917.
pre-harvest rapeseed-mustard acreage estimation at
state and national levels using multi-date AWiFS data. Sharman, M.J. (1993). The agriculture project of the
Scientific Report No. EOAM/SAC/FASAL- Joint Research Centre: Operational use of Remote
TD/SN/41/2010, Space Applications Centre (ISRO), Sensing for agricultural statistics. Proc. Int. Symp.
Ahmedabad, India. “Operationalization of Remote Sensing”, 19-23 April,
ITC, Enschede, The Netherlands.
Parihar, J.S. and M.P. Oza (2006). FASAL: An
integrated approach for crop assessment and production Sharma N., D.R. Rajak, R.K. Jain, S. Palria and J.S.
forecasting. Proc. SPIE 6411, Agriculture and Parihar (2014). Monitoring rabi crop area using multi-
Hydrology Applications of Remote Sensing, 641101; year MODIS data: A case study of Gujarat, India.
doi:10.1117/12.713157. Journal of Geomatics, 8 (2): 211-215.
http://www.isgindia.org/JOG/abstracts/October-
Parihar, J.S. (2016). FASAL concept in meeting the 2014/JOG-VOL08nO2-Oct2014_pg86%20-90.pdf
requirements of assessment and forecasting crop
production affected by extreme weather events. Sud U.C., T. Ahmad, V.K. Gupta, H. Chandra, P.M.
Mausam, 67(1): 93-104. Sahoo, K. Aditya, M. Singh and A. Biswas (2015).
Research on Improving Methods for Estimating Crop
Patnaik, C. and V.K. Dadhwal (1995). Study of crop Area, Yield and Production under Mixed, Repeated and
discrimination by simulated LISS-III data. J. Indian Continuous Cropping. http://gsars.org/wp-
Soc. Remote Sensing, 23(4): 195-199. content/uploads/2016/01/WP_Synthesis-of-Literature-
and-Framework_Improving-Methods-for-Estimation-
Potgieter, A.B., A. Apan, G. Hammer and P. Dunn of-Crop-Area-190116.pdf.
(2005). “Spying” on the winter wheat crop – Generating
objective planted area and crop production estimates Verhoef, W., M. Menenti and S. Azzali (1996). A color
using MODIS imagery. Proceedings of SSC 2005 composite of NOAA-AVHRR based on time series
Spatial Intelligence, Innovation and Praxis: The national (1981–1992). International Journal of Remote Sensing,
biennial Conference of the Spatial Sciences Institute, 17:231–235.
September, 2005. Melbourne: Spatial Sciences Institute.
ISBN 0-9581366-2-9. Verhoef, Wout, Albert van der Kamp and Koelemeijer
Robert (2005). Climate indicators from time series of
 88

Journal of Geomatics Vol 10 No. 1 April 2016

NDVI images (CITISEN). Final report, National data for crop classification in the U.S. Central Great
Aerospace Laboratory NLR, NLR-CR-2005-474. plains. Remote Sensing of Environment, 108: 290–310.

Vermote, E.F., S.Y. Kotchenova and J.P. Ray (2011). Wen Jun, Su Zhongbo, and Ma Yaoming (2004).
MODIS surface reflectance user’s guide (ver 1.3). Reconstruction of a Cloud-free Vegetation Index Time
MODIS land surface reflectance science computing Series for the Tibetan Plateau. Mountain Research and
facility. http://modis- Development 24(4):348-353, doi:
sr.ltdri.org/guide/MOD09_UserGuide_v1_3.pdf (last http://dx.doi.org/10.1659/0276-4741(2004).
accessed April 1, 2016).
Xiangming Xiao, S. Boles, S. Frolking, C. Li, J.Y.
Vyas, S., R. Nigam, N.K. Patel and S. Panigrahy (2013). Babu, W. Salas and B. Moore III (2006). Mapping
Extracting regional pattern of wheat sowing dates using paddy rice agriculture in South and Southeast Asia
multispectral and high temporal observations from using multi-temporal MODIS images. Remote Sensing
Indian geostationary satellite. J Indian Soc Remote of Environment, 100: 95 – 113.
Sens, 41(4):855–864, DOI 10.1007/s12524-013-0266-
3. Xu, Yongming and Yan, Shen (2013). Reconstruction
of the land surface temperature time series using
Wardlow, B.D., S.L. Egbert and J.H. Kastens (2007). harmonic analysis. Computers & Geosciences 61:126–
Analysis of time-series MODIS 250 m vegetation index 132.
 89

Journal of Geomatics Vol 10 No. 1 April 2016

Formosat-2 with Landsat-8 temporal - multispectral data for wheat crop identification
using Hypertangent Kernel based Possibilistic classifier
Rohit Nandan1, Ankit Kamboj2, Anil Kumar3, A. Senthil Kumar3 and K. Venkata Reddy1
1 National Institute of Technology, Warangal
2Gurukul Kangri Vishwavidyalaya, Haridwar
3Indian Institute of Remote Sensing, Dehradun, India

Email: [email protected], [email protected], [email protected], [email protected] and [email protected]

(Received: Feb 27, 2016; in final form: Apr 18, 2016)

Abstract: Agriculture plays major role in India’s economy, and provides undoubtedly the largest livelihood with its
allied sectors. Crop type identification serves in number of applications such as crop yield forecasting, collecting crop
production statistics, facilitating crop rotation records, mapping soil productivity, identification of factors influencing
crop stress, assessment of crop damage and monitoring farming activity. To identify particular crop type in a single date
imagery is a challenging task. However, Classification facilitates the multi-temporal images by taking into account
changes in reflectance as a function of plant phenology. This research work deals with Possibilistic c-means classifier
with Hypertangent kernel for wheat (Triticumaestivum) identification in Haridwar, Uttarakhand, India. The vegetation
index outputs of Formosat-2 and Landsat-8 (Operational Land Imager) sensors were arranged in chronological order of
their date and prepared three temporal datasets which cover whole phenological cycle of wheat. It was evaluated that for
2.7, 2.5, and 2.5 values of weighted constant (m), images of 4 (Formosat-2: 04 Dec 2014, 30 Jan 2015, 21 Feb 2015;
Landsat-8: 16 Mar 2015), 5 (Formosat-2: 04 Dec 2014, 30 Jan 2015, 21 Feb 2015; Landsat-8: 16 Mar 2015, 01 Apr
2015), and 6 (Formosat-2: 04 Dec 2014, 30 Jan 2015, 21 Feb 2015, 09 Apr 2015; Landsat-8: 16 Mar 2015, 01 Apr 2015)
date combination respectively represent the nicely separated wheat crop from other vegetation and were easily
differentiated between early harvested and late harvested wheat crops. This study demonstrates that 5 date combination
was sufficient to discriminate late harvested wheat crops and 6 date combination was sufficient to discriminate early
harvested wheat crops.

Keywords: Wheat identification, Phenology, Soft classification, Possibilistic c-means (PCM), Kernel, Weighted constant

1. Introduction 2012). Hence, multi-temporal approach is more

beneficial to identify specific crop.
Maps of crop type are created by national and
multinational agricultural agencies, insurance agencies, The atmospheric effect on satellite imagery is common
and regional agricultural boards to prepare an inventory which affects the usability of data. For temporal
of what was grown in certain areas and when? This analysis, the data should be free of rain, haze and clouds
serves in number of applications such as crop yield while acquiring data covering the crop growth season.
forecasting, collecting crop production statistics, But this requirement is rarely met. This hampers the
facilitating crop rotation records, mapping soil temporal analysis results and provides gaps in temporal
productivity, identification of factors influencing crop data sampling (Steven et al., 2003). The use of multiple
stress, assessment of crop damage and monitoring sensors for filling up these periods of long absence in
farming activity. temporal data may provide a solution to this problem
(Shang et al., 2008).
Traditional methods of obtaining specific crop maps are
through census and ground surveying. In order to For crop identification, classification facilitates the
standardize measurements, remote sensing offers an multi-temporal images by taking into account changes
efficient and well-founded means to map crop type and in reflectance as a function of plant phenology. Pixel
acreage. Spectral reflectance of vegetation varies with based classification method has been traditionally used
respect to change in plant phenology, stage and crop to identify specific crops. But, the technique is efficient
health. In India, different crops are grown in the vicinity only when the spatial resolution of sensors match the
of each other, so spectral response of target class may Land Use/ Land cover class on the ground and there
overlap with other class(s). Therefore, crop mapping should not exist any spectral mixing at the inter-class
using single date imagery is a real challenge (Wardlow boundaries. The problem of occurrence of mixed pixels
et al., 2007; Masialeti et al., 2010). The information can be tackled by applying fuzzy based classification
about the varying pattern of growth cycle and the approach. The fuzzy set theory was proposed by Zadeh
occurrence of varying crop phenology stages in the time (1965) to handle the uncertainty in class assignment.
domain can help in discriminating or identifying crops Bezdek et al. (1981) improved a fuzzy based
using remote sensing technique (Murty et al., 2003; Niel classification technique; Fuzzy c-means (FCM) which
et al., 2004; Doriaswamy et al., 2006; Misra et al., was put forward by Dunn, 1973. In FCM, the
membership value is a measure of the “degree of

© Indian Society of Geomatics

 90

Journal of Geomatics Vol 10 No. 1 April 2016

sharing” of the pixel for the class. While in the case of
𝜌𝑁𝐼𝑅 −𝜌𝑅𝑒𝑑
Possibilistic c-means (PCM); the membership value 𝑁𝐷𝑉𝐼 = (1)
𝜌𝑁𝐼𝑅 +𝜌𝑅𝑒𝑑
represents “the degree of belongings or compatibility or
where, 𝜌𝑁𝐼𝑅 represents reflectance at Near Infrared band
typicality” (Chawla 2010). PCM was proposed by
Krishnapuram and Keller (1993). PCM algorithm has and 𝜌𝑅𝑒𝑑 represents reflectance at Red band.
the capability to extract single class and handle the
Possibilistic c-means algorithm is the modified form of
problem of noises and outliers, which commonly exist
FCM which was proposed by Krishnapuram and Keller
in the remote sensing data.
(1993). The objective function for PCM is given in
FCM and PCM fail to give results with higher accuracy, equation (2):
𝑁 𝐶
when the classes are linearly non-separable (Wu, 2006). 𝑚 2
In such a situation, if kernels are included in the existing 𝐽𝑚 (𝑈, 𝑉) = ∑ ∑(µ𝑖𝑗 ) ||𝑋𝑖 − 𝑉𝑗 || 𝐴
𝑖=1 𝑗=1
algorithms then it has the capability to handle mixed 𝐶 𝑁
pixels with linearly non-separable classes. Kernels are 𝑚
tools which take data to a higher dimension, such that + ∑ 𝜂𝑗 ∑(1 − 𝜇𝑖𝑗 )
the classes are linearly separable by a hyper plane. 𝑗=1 𝑖=1

Zhang and Chen (2002) presented a kernel based Fuzzy (2)

c-means (FCM) algorithm which was tested on where, U is the matrix between the number of pixel and
spherical dataset and real iris data. Kumar et al. (2006) number of classes. The equation (2) is subject to
studied the effect of different kernels while generating constraints,
density estimation using SVM with respect to overall 𝑚𝑎𝑥
sub-pixel classification accuracy of multi-spectral data. for all i 𝑗 µ𝑖𝑗 > 0
for all j ∑𝑁
𝑖=1 𝜇𝑖𝑗 > 0
A crop can be discriminated by exploiting the variations for all i, j 0 ≤ 𝜇𝑖𝑗 ≤ 1
in spectral response of various crops in a multi-
dimensional feature space produced by different where, 𝑋𝑖 is the vector denoting spectral response of a
spectral bands, or time domain or both (Dadhwal et al., pixel i, 𝑉𝑗 is a collection of vector of cluster centres, 𝜇𝑖𝑗
2002). Working with temporal data, the number of is class membership values of a pixel, c and N are
spectral bands also increases. Hence, to reduce the number of clusters and pixels respectively, m is a
spectral dimensionality of the data, temporal indices are weighting component (1<m<∞), which controls the
generated. The vegetation indices maximize the degree of fuzziness.
sensitivity of plant biophysical parameters and perform
radiometric correction in the satellite imagery (Jensen, 𝜂𝑗 is dependent on the shape and average size of cluster
2009). j and is computed as in Eq. (3):
Wheat has proven itself to be a highly adapted crop ∑𝑁 𝑚 2
𝑖=1 𝜇𝑖𝑗 𝑑𝑖𝑗
across the world. Wheat is main cereal crop in India and 𝜂𝑗 = 𝐾 ∑𝑁 𝑚 (3)
𝑖=1 𝜇𝑖𝑗
staple food of millions of Indians, particularly in the
The class memberships, 𝜇𝑖𝑗 are obtained from equation
northern and north-western parts of the country. Hence,
importance of monitoring crop cannot be unnoticed. (4):
1
Specific crop mapping has essential role for crop 𝜇𝑖𝑗 = 1 (4)
𝑑2
𝑖𝑗
𝑚−1
acreage and yield estimation. 1+( )
𝜂𝑗

The major objective of this research is to propose a soft where, 𝑑𝑖𝑗 represents the distance between the pixels
classifier algorithm which has the capability to extract value i and mean of the class j.
wheat crop with better accuracy dealing with non-
linearity within the classes. Specific objectives include: PCM is robust to handle mixed pixels but it fails to
(1) To implement Hypertangent kernel based correctly classify the pixels when the classes are linearly
Possibilistic classifier for specific crop identification in non-separable (Wu, 2006). Kernel methods provide a
bi-sensor multi-spectral data, and (2) To evaluate compatible and reliable framework for developing
number of temporal images and optimized value of nonlinear technique of classification and have useful
weighted constant that may be best suited for wheat crop properties when dealing with low number of training
identification. data, presence of heterogeneous land cover and different
noise sources in the data.
2. Temporal vegetation index and classification
approach In this paper a robust supervised fuzzy classification
technique, Kernel based Possibilistic c- Means
The Normalized difference vegetation index (NDVI) algorithm (KPCM) has been presented. Its basic idea is
layers of different date were prepared and stacked in to transform the low dimensional input data into a
chronological order and prepared three temporal higher dimensional feature space via a kernel method.
datasets of vegetation index. The NDVI was proposed After the implementation of kernels, the classes become
by Kriegler et al. (1969) and the mathematical linearly separable and PCM is performed on the feature
expression is given by equation (1). space.
 91

Journal of Geomatics Vol 10 No. 1 April 2016

prepared three datasets which cover major five growth
The mathematical expression of Hypertangent kernel phases (Tillering, Stem extension, Heading, Flowering
used in this study is given in equation (5) (Kaur et al., and Ripening) of wheat crop. The datasets for Formosat-
2012): 2 and Landsat-8 are shown in Table 2(a) and 2(b)
2
‖𝑥𝑖 −𝑥𝑗 ‖ respectively.
𝐾(𝑥, 𝑥𝑖 ) = 1 − 𝑡𝑎𝑛ℎ (− ),𝜎 > 0 (5)
𝜎2

Hence, the objective function of Kernel based

Possibilistic c-Means (Zhang and Chen, 2003) is given
by equation (6):
𝑁 𝑐

𝐽𝑚 (𝑈, 𝑉) = ∑ ∑ 𝜇𝑖𝑗 𝑚 𝐾(𝑋𝑖 , 𝑉𝑗 )

𝑖=1 𝑗=1
𝐶 𝑁
𝑚
+ ∑ 𝜂𝑗 ∑(1 − 𝜇𝑖𝑗 )
𝑗=1 𝑖=1
(6)
The updated membership value can be computed as
given by equation (7):
1
𝜇𝑖𝑗 = 1 (7)
𝐾(𝑋𝑖 ,𝑉𝑗 ) 𝑚−1
1+( )
𝜂𝑗

where, Figure 1: Study area: East Haridwar, Uttarakhand,

2 India
𝐾(𝑋𝑖 , 𝑉𝑗 ) = ||∅(𝑋𝑖 ) − ∅(𝑉𝑗 )|| (8)
where, ∅ represents the kernel function. And Table 1(a): Formosat-2 Sensors Specifications
2 (http://www.satimagingcorp.com/satellite-
||∅(𝑋𝑖 ) − ∅(𝑉𝑗 )||
sensors/other-satellite-sensors/formosat-2/)
= 𝐾(𝑋𝑖 , 𝑋𝑖 ) + 𝐾(𝑉𝑗 , 𝑉𝑗 ) − 2𝐾(𝑋𝑖 , 𝑉𝑗 )
(9) Band Wavelengt Spatial
The Hypertangent Kernel equation (5) helps to evaluate h (in µm) Resolution (in m)
the above kernel function. Band 1 –Blue 0.45 – 0.52 8

3. Study area and data used Band 2 –Green 0.52 – 0.60 8

Band 3 –Red 0.63 – 0.69 8
The study area under this research is East side of
Haridwar, Uttarakhand, India towards National Band 4 – Near 0.76 – 0.90 8
Highway 74 as shown in fig. 1. Uttarakhand is a state Infrared (NIR)
in the northern part of India. The Haridwar district P - Panchromatic 0.45 – 0.90 2
shares its boundaries by Dehradun in the north, Pauri
Garhwal district in the east while, west and south are Table 1(b): Landsat-8 OLI Sensors Specifications
bounded by districts of Uttar Pradesh state. The central (Landsat-8 Data User Handbook-June 2015)
latitude and longitude of the study area taken are
29°52'20.3124"N and 78°10'25.0998"E respectively. Band Wavelength Spatial
The land is fertile with river Ganga flowing through the (in µm) Resolution
district and tourism and agriculture remains the (in m)
backbone of the district. The major crops grown in the Band 1- Coastal/ 0.433-0.453 30
study area are rice, wheat, lentil, groundnut, mustard Aerosol
and plantations like citrus fruits, mango, litchi etc. In Band 2- Blue 0.450-0.515 30
this study, remotely sensed temporal images of Band 3- Green 0.525-0.600 30
Formosat-2 and Landsat-8 (Operational Land Imager) Band 4- Red 0.630-0.680 30
were used for wheat crop identification. The sensors Band 5- Near 0.845-0.885 30
specifications are shown in Table 1(a) and 1(b). Infrared (NIR)
Band 6-Short 1.560-1.660 30
Wheat (Triticumaestivum) is the Rabi season crop in Wave Infrared
India. The sowing window of wheat is 2nd week of Band 7- Shot Wave 2.100-2.300 30
November to 4th week of December in Haridwar district. Infrared
Then it passes through a series of developmental phases
Band 8- 0.500-0.680 15
from sowing to harvest. The harvesting period for wheat
Panchromatic
crop is mid-March to first week of April.
Band 9- Cirrus 1.360-1.390 30
The combination of Formosat-2 and Landsat-8 (OLI)
images NDVI outputs have been combined and
 92

Journal of Geomatics Vol 10 No. 1 April 2016

Table 2(a): Formosat-2 temporal datasets harvested and late harvested wheat crops were evaluated
by using the maximum membership difference between
Formosat-2 both classes among all three NDVI temporal datasets.
Date Reference code The Flowchart of the methodology adopted is shown in
04 Dec 2014 F1 Fig. 2:
30 Jan 2015 F2
21 Feb 2015 F3 Formosat-2 and Landsat-8 OLI Temporal
09 Apr 2015 F4 Data

Table 2(b): Landsat-8 temporal datasets

Atmospheric Corrections
Landsat-8 (OLI) Temporal NDVI Dataset
Date Reference code
16 Mar 2015 L1 Creation of Training Data
01 Apr 2015 L2

Reference data for wheat was identified on the imagery Supervised Soft Classification using
from the GPS data collected during a field visit on 16 Hypertangent Kernel based Possibilistic c-
Means Classifier
March 2015. During field survey the early sowing and
late sowing wheat crop samples were collected based on
the information provided by concern farmers. Further,
Evaluation for optimized weighted
this reference data was used as a training and testing constant and best temporal date
data for classification and validation. combination
4. Methodology
Figure 2: Methodology adopted
Four temporal data of Formosat-2 from 04 Dec 2014 to
09 April 2015 and two temporal data of Landsat-8 (OLI) 255
of 16 March 2015 and 01 April 2015, sensors have been 205
NDVI (on 8 bit scale)

used. These data were geometrically corrected for both

sensors by using reference image as Formosat-2 21 Feb 155
2015 date. The Temporal Images were atmospherically 105
corrected by ATCOR. The NDVI images were
generated and linearly stretched in the scale of 0 to 255 55
by image enhancement technique, because of good 5
visibility. The SMIC (Sub-Pixel Land Cover Mapping
Image Classifier), a JAVA based image processing -45
package (Kumar et al., 2006) supports 8 bit as well as
16 bit scale imagery has been used.
Temoral Formosat-2 and Landsat-8 data
Early Harvested Wheat
The vegetation index outputs of Formosat-2 and
Late Harvested Wheat
Landsat-8 (Operational Land Imager) sensors were
arranged in chronological order of their dates and
prepared three temporal datasets of 4 (Formosat-2: 04 Figure 3: Spectral growth curve of wheat at training
Dec 2014, 30 Jan 2015, 21 Feb 2015; Landsat-8: 16 Mar sites
2015), 5 (Formosat-2: 04 Dec 2014, 30 Jan 2015, 21 Feb
2015; Landsat-8: 16 Mar 2015, 01 Apr 2015), and 6 5. Result and discussion
(Formosat-2: 04 Dec 2014, 30 Jan 2015, 21 Feb 2015,
09 Apr 2015; Landsat-8: 16 Mar 2015, 01 Apr 2015) The spectral growth curve of early harvested and late
date combination, which cover whole phenological harvested wheat crop has been shown in fig.3. The
cycle of wheat. graph is showing the variation of NDVI between both
classes with respect to time.
Hypertangent kernel based PCM soft classifier has been
applied on the all three date combination of temporal The weighted constant; m was optimized for 4, 5 and 6
vegetation index datasets. By using testing data the date combination for two classes; early harvested wheat
optimized weighted constant and best temporal crop and late harvested wheat crop. The value of m was
vegetation index datasets were evaluated by observing varied from 1.5 to 3.0 for each classification. However,
the difference in membership between the target class the testing sites data were used to find the difference
and other class(s). The maximum membership between membership values of both output fraction
difference containing weighted constant value will be images. As the difference tends to be constant the
the optimized weighted constant. Best temporal corresponding value of m can be considered as its
vegetation index datasets to identify two classes of early optimized value. The graphs shown in figure 4(a), 4(b)
and 4(c) depict the difference between the membership
 93

Journal of Geomatics Vol 10 No. 1 April 2016

values of classes as m was varied for 4, 5 and 6 date and 5(b) represent nicely separated membership values
combination at unbiased sites. While in the case in late harvested and early harvested wheat crop fields
Hypertangent kernel based PCM optimized m was for 5 and 6 date combination temporal images. The
observed at 2.7, 2.5 and 2.5 respectively. classified images membership values have been
stretched up in 8 bit scale for good visibility.
(a) (a)
0.15
Difference betweeen
Membership value

0.1
0.05
0
1.5 1.7 1.9 2.1 2.3 2.5 2.7 2.9 3

Weighted Constant (m)

Difference between Early Harvested and Late
Harvested classes in Early Harvested test sites
Difference between Late Harvested and Early
Harvested classes in Late Havested test sites
(b)
(b)
0.3
Difference Between
membership value

0.2

0.1

0
1.5 1.7 1.9 2.1 2.3 2.5 2.7 2.9 3
Weighted Constant (m)
Difference between Early Harvested and Late
Harvested classes in Early Harvested test sites
Difference between Late Harvested and Early
Harvested classes in Late Havested test sites
Figure 5: Classified images extracted by using Hyper
Tangent Kernel based PCM classifier, (a) Class1-
(c) Early harvested wheat crop (Six date; optimized m=
0.3
Difference in membership

2.5); (b) Class2- Late harvested wheat crop (Five

date; optimized m= 2.5)
0.2

0.1 6. Conclusion
value

0 The study carried out in this research indicates the

1.5 1.7 1.9 2.1 2.3 2.5 2.7 2.9 3 potential of bi-sensor approach in the temporal analysis
Weighted constant (m) studies. It has been observed that to identify wheat crop,
Difference between Early Harvested and Late the 5 growth phases covering images are enough. The
Harvested classes in Early Harvested test sites Formosat-2 temporal data shows less spectral mixing
Difference between Late Harvested and Early than Landsat-8 because of the finer resolution of its
Harvested classes in Late Havested test sites sensor. Therefore, five and six date combination have
more spectral mixing than four date combination
Figure 4: Weighted Constant (m) vs Difference because of coarser resolution Landsat-8 images
between membership values for (a) 4 date contribution in datasets. The results indicate that 5 date
combination datasets; (b) 5 date combination combination was sufficient to discriminate late
datasets; (c) 6 date combination datasets harvested wheat crops and 6 date combination was
sufficient to discriminate early harvested wheat crops.
The four dates classified results were not appropriate to Therefore, the study conclude that the Hypertangent
discriminate both wheat classes. The Five dates and six kernel based Possibilistic c-means classification
dates classified results were represented the high approach is robust to handle the mixed pixels to identify
difference between membership values of both classes wheat crop. This selected best date can help in providing
at unbiased sites (The wheat crop area, which have not a temporal window for monitoring the wheat crop. This
been used as training sites). However, five dates were approach would help in generating accurate maps with
found sufficient to discriminate late harvested wheat the help of optimum number of strategically selected
crops. The six dates classified results were also temporal remote sensing images covering the growing
evaluated for good discrimination of early harvested season of the crop, and help in saving resources spent in
wheat crop from other non-interest classes. Figure 5(a) mapping too.
 94

Journal of Geomatics Vol 10 No. 1 April 2016

Acknowledgements Krishnapuram, R. and J.M. Keller (1993). A
possibilistic approach to clustering. Fuzzy Systems,
The authors would like to thank Sponsors the National IEEE Transactions on, 1(2), pp. 98-110.
Space Organization, National Applied Research
Laboratories and jointly supported by the Chinese Kumar, A., S.K. Ghosh and V.K. Dadhwal (2006). Sub-
Taipei Society of Photogrammetry and Remote Sensing pixel land cover mapping: SMIC system. ISPRS
and the Center for Space and Remote Sensing Research, International Symposium on Geospatial Databases for
National Central University of Taiwan for providing Sustainable Development, Goa, India.
Formosat-2 temporal images of Haridwar district,
Uttrakhand, India. The first author is thankful to IIRS Kumar, A., S.K. Ghosh, and V.K. Dadhwal (2006).
ISRO for allowing ten month M. Tech project on study Study of mixed kernel effect on classification accuracy
of specific crop identification by using kernel based using density estimation. In Proceedings of the ISPRS
fuzzy classifier. Commission VII Symposium (Vol. 36, No. Part 7).

References Masialeti, I., S. Egbert and B.D. Wardlow (2010). A

comparative analysis of phenological curves for major
Aggarwal, R., A. Kumar, P.L.N. Raju and Y.K. Murthy crops in Kansas. GIScience and Remote Sensing, 47(2),
(2014). Gaussian kernel based classification approach pp. 241-259.
for wheat identification. ISPRS-International Archives
of the Photogrammetry, Remote Sensing and Spatial Misra, G., A. Kumar, N.R. Patel, R. Zurita-Milla and A.
Information Sciences, 1, pp. 671-676. Singh (2012). Mapping specific crop - A multi sensor
temporal approach. In Geoscience and Remote Sensing
Bezdek, J.C., C. Coray, R. Gunderson and J. Watson Symposium (IGARSS), 2012 IEEE International
(1981). Detection and characterization of cluster conference, pp. 3034-3037.
substructure I. Linear structure: fuzzy c-lines. SIAM
Journal on Applied Mathematics, 40(2), pp. 339-357. Murthy, C.S., P.V. Raju and K.V.S. Badrinath (2003).
Classification of wheat crop with multi-temporal
Chawla, S. (2010). Possibilistic c-means-spatial images: Performance of maximum likelihood and
contextual information based sub-pixel classification artificial neural networks. International Journal of
approach for multi-spectral data. Unpublished Ph.D Remote Sensing, 24(23), pp. 4871-4890.
thesis, University of Twente (ITC), Enschede. pp. 14-15
Shang, J., H. McNairn, C. Champagne and X. Jiao
Dadhwal, V.K., R.P. Singh, S. Dutta and J.S. Parihar (2008). Contribution of multi-frequency, multi-sensor,
(2002). Remote sensing based crop inventory: A review and multi-temporal radar data to operational annual crop
of Indian experience. Tropical Ecology, 43(1), pp. 107- mapping. In Geoscience and Remote Sensing
122. Symposium, 2008. IGARSS 2008. IEEE
International (Vol. 3, pp. III-378). IEEE.
Department of the Interior U.S. Geological Survey
(2015). Landsat-8 Data User Handbook, pp.9-10. Steven, M.D., T.J. Malthus, F. Baret, H. Xu and M.J.
Chopping (2003). Inter-calibration of vegetation indices
Doraiswamy, P.C., B. Akhmedov and A.J. Stern (2006). from different sensor systems. Remote Sensing of
Improved techniques for crop classification using Environment, 88(4), pp. 412-422.
MODIS imagery. In Geoscience and Remote Sensing
Symposium, 2006. IGARSS 2006. IEEE International Van Niel, T.G. and T.R. McVicar (2004). Determining
Conference, pp. 2084-2087. temporal windows for crop discrimination with remote
http://www.satimagingcorp.com/satellite- sensing: a case study in south-eastern
sensors/other-satellite-sensors/formosat-2/(Retrieved Australia. Computers and Electronics in Agriculture,
30 May 2016) 45(1), pp. 91-108.

Jensen, J.R. (2009). Remote sensing of the environment. Wardlow, B.D., S.L. Egbert and J.H. Kastens (2007).
2nd Edition. Pearson Education, pp. 382-399. Analysis of time-series MODIS 250 m vegetation index
data for crop classification in the US Central Great
Kaur, P., P. Gupta and P. Sharma (2012). Review and Plains. Remote Sensing of Environment, 108(3), pp.
comparison of kernel based fuzzy image segmentation 290-310.
techniques. International Journal of Intelligent Systems
and Applications (IJISA), 4(7), pp. 50. Wu, X.H. (2006). A possibilistic c-means clustering
algorithm based on kernel methods. In
Kriegler, F.J., W.A. Malila, R.F. Nalepka and W. Communications, Circuits and Systems Proceedings,
Richardson (1969). Preprocessing transformations and 2006 International Conference on (Vol. 3, pp. 2062-
their effects on multispectral recognition. In Remote 2066). IEEE.
Sensing of Environment, 4(1), pp. 97.
Zadeh, L.A. (1965). Fuzzy sets. Information and
control, 8(3), pp. 338-353.
 95

Journal of Geomatics Vol 10 No. 1 April 2016

Zhang, D.Q. and S.C. Chen (2003). Kernel-based fuzzy
Zhang, D. and S. Chen (2002). Fuzzy clustering using and possibilistic c-means clustering. In Proceedings of
kernel method. In in The 2002 International Conference the International Conference Artificial Neural Network,
on Control and Automation, 2002. ICCA, 2002. pp. 122-125.

-------------------------------------------------

India’s Journey Towards Excellence in Building Earth Observation Cameras

George Joseph Notion Press
ISBN: 9789352069989
Rs.450/; US$ 10.99

The book traces the evolution of earth observation cameras in Indian

Space Research Organization (ISRO); how from a humble beginning of
a two band framing camera ISRO went on to developing world class
imaging system from space and the innovations carried out in the
course of development of these sensors. The book also discusses the
rational to choose various camera specifications based on the
application needs.
The book starts with the beginning of the space program in India and
systematically chronicles the journey of the development of advanced
space based imaging system. The book also provides some basic
technical insights into the building of space based remote sensing
cameras, which have been presented in a way that can be understood
by non‐specialists too. In addition the book brings out the
management aspects and the role played by leadership which make
ISRO stand out as a successful high‐tech organisation.

In addition to students and professionals in the field who will get a broad account of the functioning of
space based camera systems and the nuances in the design, development and deployment of them, those
in policy making and technical management in space agencies across the globe will also find the book
useful to understand the path taken by India to achieve pre‐eminence in this field.

“I hope the book will interest a broad range of readers both within and outside the country.
Those who were part of the journey will feel a sense of satisfaction and proud of what they
could achieve and the younger readers will be inspired and encouraged to be part of this
excitement. The book should interest all those who want to know how India has achieved
preeminence in space based remote sensing.”

A.S. Kiran Kumar

Chairman ISRO/Secretary DOS
(In the foreword to the book)

Author: Dr. George Joseph was director of the Space Applications Centre (ISRO), Ahmedabad. Under his
overall guidance the development of electro‐optical sensors started in ISRO.
 96

Journal of Geomatics Vol 10 No. 1 April 2016

Crime mapping analysis of Ajmer city -A GIS approach
Ravi Sharma, Sarvesh Palria and Parul Bhalla
Department of Remote Sensing and Geo-informatics, M.D.S. University Ajmer – 305009
Email: [email protected]

(Received: Jan 23, 2015; in final form: Mar 18, 2016)

Abstract: Due to unemployment, the crime is increasing. Criminal acts like murder, rape, kidnapping, home breaking,
theft, robbery are prohibited but exists all around the world. India is one major country where crime is increasing. The
present study deals with the crime mapping of Ajmer city with GIS approach. The study is having objectives to generate
crime maps to identify the crime pattern of Ajmer city. It is based on time series analysis to identify crime direction and
hotspots of crimes. It also analyses type of hotspot, proximity of crimes to police stations, displacement of crime across
time, crime rate of each ward and the socio-economic characteristics of city. This analysis provides insight to police with
a view to decreasing crime rate.

Key Words: Crime rate, GIS, Hotspot, Land use/land cover, Crime map

1. Introduction historical events and predict future events. GIS based

DSS can also be used to get critical information in
Crime is a human tendency, therefore, its distribution tactical planning and response. GIS helps to identify
across the landscape is not geographically random. potential suspects of crime and thus leads to decrease
Place plays a vital role in understanding crime. To the crime rate.
reduce crime, geography of crime needs to be
understood as crime has an inherent geographical The present study focuses on the creation of the
quality. When a crime occurs, it happens at a place with database of spatial and non-spatial attribute, calculation
a geographical location (Chainey and Ratcliffe, 2005). of crime rate and crime density in Ajmer city and
For someone to commit a crime, one must come from a identification of hotspots of different crime occurring in
place (such as their home, work or school). This place Ajmer city. Temporal crime pattern of Ajmer city is also
could be the same location where the crime is analysed and illustrated using GIS tools.
committed or is often close to the place where crime
occurred. 2. Study area

Jaishankar et al. (2004) showed that use of GIS provides Ajmer is the district of Rajasthan state. Ajmer city is the
a convenient tool for crime pattern analysis due to its head quarter of the district. Ajmer city has a population
geographic referencing capabilities. It provides valuable of 542321. Ajmer city is a religious and tourist place. It
information concerning property of crimes including is surrounded by NH-8, NH-14, NH-79 and NH-89.
data on the social and physical characteristics of areas Ajmer city is surrounded by the Aravali hills. Ajmer city
that contribute to localized criminal activity. stretches from 26˚23’ North to 26˚23’ North and 74˚36’
Thangavelu et al. (2013) discussed about the importance East to 74˚40’ east. It has nine Police Stations (PS) and
of GIS, as it can be used as a tool to identify factors one Mahila PS. All nine PS boundaries of Ajmer city
contributing to crime and thus allow police department have been selected for study. Figure 1 shows the
to proactively respond to the situations before they location map of study area.
become problematic. Crime analysis mapping is a
valuable problem solving tool because it can lead to the
identification of new problems facing law enforcement,
lend a visual perspective to an analysis, assist in the
development of an effective response, aid in the
formation of partnerships by providing a common point
of reference and assist evaluation procedures (Velasco
and Boba, 2005).

A study of GIS based Decision Support System (DSS)

for crime mapping in Ahmedabad city was conducted
by Patel et al. (2014). They found that GIS based DSS
is important as it uses geography and analysis as an
interface for integrating and accessing massive amounts
of location-based information. GIS based DSS allows
police personnel to plan effectively for emergency Figure 1: Location map of study area
response, determine mitigation priorities, analyze

© Indian Society of Geomatics

 97

Journal of Geomatics Vol 10 No. 1 April 2016

3. Data used and methodology murder, attempt to murder, rape, kidnapping, robbery,
home breaking day and night, automobile theft and other
The IPC crime incidents data is collected from the SP theft are given in table 1. IPC crimes in different police
Office, Crime Branch & each PS of Ajmer city from stations of Ajmer city are given in table 2.
2009 to 2014 (Up to March 2014). Crime type data about

Table 1: Statistics of crime types in Ajmer city for various year

Year
2014 Total
Crime Type 2009 2010 2011 2012 2013
(Upto March)
Murder 11 60 16 14 12 0 60
Home breaking 82 436 55 58 134 32 436
Robbery 16 49 9 5 8 5 49
Kidnapping 27 155 17 22 64 7 155
Rape 10 103 20 13 35 14 103
Automobile theft 143 928 129 162 286 53 928
Other thefts 112 513 98 72 134 30 513
Attempt to murder 13 64 16 8 14 2 64
Total 414 2308 360 354 687 143 2308

Table 2: IPC crimes in different police stations of Ajmer city

Year
Police Station 2014(Upto
S.No. 2009 2010 2011 2012 2013 Total
Name March)
1 Civil Lines 75 52 43 42 101 18 331
2 Clock Tower 56 21 29 25 49 11 191
3 Ganj 9 14 7 26 52 15 123
4 Dargah 31 19 31 26 25 5 137
5 Alwargate 29 36 46 39 78 13 241
6 Kotwali 71 67 77 64 81 13 373
7 Adarshnagar 26 27 22 33 82 16 206
8 Christianganj 73 83 80 62 157 33 488
9 Ramganj 44 31 25 37 62 19 218
Total 414 350 360 354 687 143 2308

For spatial data analysis, Cartosat-I data and toposheets

were used. Non-spatial attribute data were collected
from all PS and crime branch office. GPS was used to
collect geographic locations of crime incidence. Census
data and ward map were collected from Statistical
department and Nagar Nigam of Ajmer.

Thematic maps were generated on the yearly basis and

year-wise maps were prepared for the period 2009- 2014
(upto March). Year-wise maps help to show the change
detection of crime pattern and the direction of crime.
Crime wise and crime type maps were prepared. These
thematic maps were analyzed to find out the hotspot
areas and then spatio-temporal crimes were identified.
The methodology is given in figure 2.

4. Result and discussion

Crime map of Ajmer city for the period 2009 – 2014

with PS boundary overlaid is given in figure 3. Location
map of PS of Ajmer city according to municipal
boundary is given in figure 4. Cartosat-I data was used Figure 2: Flow chart of methodology
for road network delineation and base map preparation
from toposheet.
 98

Journal of Geomatics Vol 10 No. 1 April 2016

4.2 Civil Lines PS
After studying about 2009–2014 map of Civil Lines PS,
it can be seen that –murder and attempt to murder both
were not found in the category of hotspot area.
Kidnapping mainly occurred near Lohakhan bus stand.
Home breaking at Civil Lines area (Near Shastrinagar
road). Two wheeler thefts happened at parking areas of
Ajmer hospital, roadways bus stand, near Mission
school, old RPSC, Civil Lines hospital, Collectorate
parisar, and at Ajmer club park. Other theft area
occurred at bus stand, Session court, Rajasthan Board,
Figure 3: Crime map of Ajmer city (2009 – 2014) Lohakhan, Civil Lines and Ajmer club. Hotspots were
with police station (PS) boundary found in 500meter buffer zone of Civil Lines PS and Out
Post (OP) Civil Lines.

4.3 Clock tower PS

After studying about 2009–2014 map of Clock tower
PS, it was found that murder and attempt to murder both
are not found in the category of hotspot area.
Kidnapping mainly occurred near Trambe station and
Home breaking night at Janta market.

Two wheeler thefts occurred at Sant Francis hospital,

Babu colony, Vimla market; Outside Jain namkin, GCA
parking, Apna market. Four wheelers were mainly
stolen from Sant Francis hospital and Babu colony.
Other thefts occurred at Padav. Robbery occurred at
Figure 4: Location map of Police Stations (PS) of Martin Bridge.
Ajmer city according to municipal boundary
Hotspots were found in 100 meters buffer zone of Clock
By observing and analyzing the crime data of Ajmer tower PS and OP Kesarganj were at high rate. No other
city’s nine PS, crime maps for various years as well as hotspot was found in buffer zone of OP Usarigate.
for various PS were prepared. Four illustrative such
maps given in figure 5 (a – d). Composite crime map of 4.4 Ganj PS
Ajmer city (2009 - 2014) with municipal boundary of After studying about 2009–2014 map of Ganj P.S., it
Ajmer city is given in figure 6. The PS-wise result and was found that murder and attempt to murder both were
patterns are discussed below. not found in the category of hotspot area. Kidnapping
occurred at mainly 300 meter buffer zone of Ganj PS.
4.1 Aadarsh Nagar PS Home breaking night, two wheeler theft and other theft
After studying about 2009–2014 map of Aadarsh Nagar crimes mainly occurred in 300 meter buffer zone of OP
PS, it was observed that –murder and attempt to murder Anasagar and PS Ganj. Home breaking occurred mainly
both were not found in the category of hotspot areas. near Dehli gate and two wheeler theft mainly at
Kidnapping had mainly occurred near Sethi colony. Ramprashad ghat.
Home breaking and night crimes took place in Sethi
colony and the 500 meter buffer zone of Aadarsh Nagar Proximity of crime with PS and outpost police station is
outpost police station. Robbery was mainly seen in very high and all hotspots were found 300 meter buffer
Ricco industrial area, where mainly the official people zone to PS and OP.
visited.
4.5 Dargah PS
Two wheelers and four wheelers were mainly stolen After studying about 2009–2014 map of Dargah PS, it
from the industrial area near Aadarsh Nagar and was found that murder and attempt to murder both were
Pravatpura circle and Hatundi circle. not found in the category of hotspot area. Terrorist
prefer a place where crowd is present because in such
In the year 2009, permanent hotspot can be seen at area one bomb blast may damage life of many people.
Pratapura circle. In the year 2010, hotspot is shifted to Due to this reason an explosion occurred near courtyard
Aadarsh Nagar colony, near Balupura road. outside the Dargah of Khawaja Moinuddin Chishti in
Ajmer (11 Oct 2007). Dargah is sensitive area for
The Aadarsh Nagar colony development was at peak terrorist attack so CCTV camera and metal detectors are
this year. Criminals are attracted at high density essentially required in such area.
population area. In the year 2011, Pratappura circle
(North and South) can be seen as major hotspot. No robbery and four wheeler theft case was found in last
Makupura circle can also be seen as the major Hotspot. 5 years in Dargah PS. area because of high crowd
 99

Journal of Geomatics Vol 10 No. 1 April 2016

(a) (b)

(c)
(d)

Figure 5: Crime map (2009 – 2014) of four Police Stations (PS) viz., (a) Aadarsh Nagar, (b) Alwar gate, (c ) Ganj
and (d) Christianganj
 100

Journal of Geomatics Vol 10 No. 1 April 2016

and place shorting for parking area. Other theft hotspot Two wheeler thefts mainly occurred at Sabzi mandi,
was found at Khwaja Mohinidin Chishti Dargah parisar Railway hospital, Shiv temple (Ajay nagar), HMT
because of high crowd. Home breaking occurred at parisar and Chungi naka. Other theft occurred at
night at Holidara. Two wheeler theft occurred at Nala Rambag circle, near OP Ramganjand in Dorai at
market and Dhanmandi. Violence crime mainly Hastivihar colony.
occurred at Andarkot.
Hotspots were found in 500 meters buffer zone of OP
4.6 Kotwali PS Ramganj and OP Dorai and proximity of crime with PS
After studying about 2009–2014 map of Kotwali PS, it Ramganj and OP Bhagwaanganj was found to be low.
was found that murder and attempt to murder both were
not found in the category of hotspot area. Home 4.9 Christian Ganj PS
breaking at night mainly occurred near Information After studying about 2009–2014 map of Alwar gate PS,
centre circle and Ajmer tower. it was found that murder and attempt to murder,
kidnapping were not found in the category of hotspot
Two wheeler thefts mainly occurred at outside JLN area but criminal comes at Anasagar lake for hiding
hospital, K.C. complex, Swami complex, Bhesa dead body so installation of CCTV camera is required
complex, Ajmer tower, Amar plaza, Prabhat cinema, at this place.
HDFC bank, Near G.P.O., near P.N.B. Kachhari road,
P.R. marg, near Jagdamba bar and near Mango masala. Home breaking during day occurred at Panchsheel
Four wheeler thefts mainly occurred at J.L.N medical nagar, UIT, main Christian ganj area and night time
college and outside Swami complex. Other theft mainly crimes took place in main Panchsheel A and B block,
occurred at JLN Hospital, PNB (Kachhari road) and 100 RPSC Colony, Aanandnagar and in 500 meters buffer
meter distance to PS at near P.R. marg. zone area of Christian ganj OP. Two wheeler theft
occurred at Anasagar chaupati, Reliance fresh, Miraj
Proximity of crime with Kotwali PS and OP Madar gate mall , Sagarvihar colony and Shastrinagar.
was very low because no hotspot area was found in 100
meter buffer zone. Hotspots are found in 100 meter Proximity of crime with outpost police station was
buffer zone of OP Kesar bagh. found to be high because Hotspot areas were found in
500meters buffer zone of OP Christian ganj, OP
4.7 Alwargate PS Shastrinagar and OP Haribhau Updhaynagar. No
After studying about 2009–2014 map of Alwargate PS, hotspot was found in buffer zone of Christian ganj PS.
it was found that murder and attempt to murder both
were not found in the category of hotspot area.
Kidnapping were mainly occurred near Kundannagar.
Home breaking during day occurred at Gandhi nagar
and night time crimes took place in main Alwar gate
area near Srinagar road.

Robbery was mainly seen in the area of Martin bridge

and Aamka talab. Two wheeler thefts occurred at main
Alwargate area, Convent school, Shiv temple, LIC
parking and 9 no. petrol pump. Four wheelers were
mainly stolen from Heena garden and near Shrinagar
road (main Alwargate area). Other theft area occurred at
Martin bridge and near P.N.B.

Hotspots were found in 500 meters buffer zone of

Alwargate PS and proximity of crime with outpost
police station was low, crime occurred 500 meters away
from the OP Nakamadar, OP J.P. nagar and OP Nareli. Figure 6: Crime map of Ajmer city (2009 - 2014) with
Some crime cases were found at Nagra, Bihariganj, municipal boundary of Ajmer city
Gokulnagar, Jadugar ,Vinaynagar and Madarpura.
Some examples of crime wise map
4.8 Ramganj PS
After studying about 2009–2014 map of Ramganj PS, it Different crimes maps of each PS were generated for
was found that murder and attempt to murder both were identification of crime type of hotspot area and
not found in the category of hotspot area. Kidnapping compared hotspots of different crimes.
mainly occurred near Pahadganj and Shubhas nagar.
Rape and kidnapping at Shashibasti area. Home Similarly, crime maps of other crime types such as
breaking during day occurred at Ajaynagar, Satguru murder, attempt to murder, rape, kidnapping, robbery,
colony and night time crimes took place in Chandverdai, home breaking during day and night, two wheeler and
Jhulelal colony and Ajaynagar. four wheeler thefts and other theft for all nine PS of
Ajmer city were generated. Crime data were classified
 101

Journal of Geomatics Vol 10 No. 1 April 2016

into four time periods of the day viz. 06:00 to 12:00,
12:00 to-18:00, 18:00 to 22:00, and 22:00 to 06:00. Fajemirokun, F., O. Adewale, T. Idowu, A. Oyewusiand
and B. Maiyegun (2006). A GIS approach to crime
These types of maps were very useful in understanding mapping and management In Nigeria: A case study of
the relationship between the geographical location, was Victoria islandlagos, shaping the change. XXIII
crime type and timing. The respective PS has started Congress Munich, Germany, October 8 – 13, 2006.
using this relationship knowledge against the criminals
to reduce the crime. This knowledge has been obtained Gupta, R., K. Rajitha, S. Basu and S.K. Mittal (2012).
though GIS. Application of GIS in crime analysis: A gateway to safe
city.7-9 February, 2012, India Geospatial Forum.
Crime rate per ten-thousand population, defined as
Jaishankar, K.S., Shanmugapriya and V. Balamurugan
number of crime in area (2004). Crime mapping in India: A GIS implementation
crime rate = X10000
population of the area in Chennai city policing. Journal of Geographic
Information Science, Volume 10, No. 1, 20-34.
for Ajmer city was 37 per ten-thousand population.
Kumar, D.L.B., K. Selvavinayagam and S. Suresh Babu
Based on the analysis, some suggestions regarding (2014). Assessment of crime and its mapping using
crime were given to the police department in terms of remote sensing and 3D geo-spatial model for Chennai
where different types of crimes are occurring and city. IJSETR Vol 9(8), 418-425.
whether these hotspot areas were covered by police
force and CCTV camera. These suggestions will help to Patel, K., P. Thakkar, L. Patel and P. Chandresh (2014).
decrease crime rate. GIS based decision support system for crime mapping,
analysis and identify hotspot in Ahmedabad city.
5. Conclusion International Journal of Modern Engineering Research
Vol. 4 (1), 32 – 35.
In this study of crime mapping, year-wise and crime
wise maps of the PS were generated. This was followed Polvi., N., T. Looman, C. Humphries and K. Pease
by in-depth analysis to identify hotspot of overall (1990). Repeat break and enter victimisation: Time
criminal activities and to understand relationship course and crime prevention opportunity. Journal of
between geographical location, crime types and time of Police Science and Administration 17:8-11.
the crimes. Crime rate for Ajmer city was 37 per ten-
thousand population. Thangavelu, A., S.R. Sathyarai and S. Balasubramanian
(2013). Assessment of spatial distribution of rural crime
After conducting the whole research, finally we mapping in India: A GIS perspective International
conclude that police department should know which Journal of Advanced Remote Sensing and GIS, Volume
type of crimes are increasing and in which direction so 2, Issue 1, pp. 70-85.
as to reduce crime rate and to establish law and order in
the city. Velasco, M. and R. Boba (2005). Manual of crime
analysis map production. Journal of Community
References Oriented Policing Services, US Department of Justice,
November.
Chainey, S. and J. Ratcliffe (2005). GIS and crime
mapping. A Journal of John Wiley and Sons, Ltd. Wilson, J.Q. and G.L. Kelling (1982). Broken windows:
The police and neighborhood safety. The Atlantic
Cohen, L.E. and M. Felson (1979). Social change and Monthly 249(3): 29-38.
crime rate Trends: A routine activities approach.,
American Sociological Review 44:588-608.
 102

Journal of Geomatics Vol 10 No. 1 April 2016

On the quality of orthometric correction determination
Raaed Mohamed Kamel Hassouna
Department of Civil Engineering, Faculty of Engineering in Shebin El-Kom, MenoufiaUniversity
Shebin El-Kom, Menoufia, Egypt
Email: [email protected]

(Received: Aug 23, 2015; in final form: Feb 21, 2016)

Abstract: In the present study, a comparison of accuracy and precision between two formulae for assessing Helmert's
orthometric correction was carried out. Two test levelling lines were used for comparison. For the two lines, the actual
levelling and gravity data were used as reference (real world) data, whereas respective UCPH2002 derived gravity values
and SRTM30-interpolated elevation data were executed as erroneous input information. Practically, one formula proved
to be less sensitive to the propagation of biases and random errors of the input data, thus yielding a remarkably accurate
and precise orthometric correction. So, it is recommended to use this formula, especially when using interpolated end
benchmarks elevations in the computation of the height correction. Furthermore, the determination of realistic
uncertainties of the height corrections could help model them as observed parameters during the adjustment of levelling
loops.

Keywords: Orthometric correction, Error analysis, Uncertainty, Accuracy

1. Introduction OC model. While an accurately computed OC value is

directly related to its nature as a correction for a
Orthometric height (OH) is of a great importance for systematic error, it could be claimed that the precision
engineering and geophysical applications. Orthometric (or uncertainty) of OC is less stringent. However, such
height differences are computed through the addition of uncertainty expresses the spread of the computed OC
the orthometric correction (OC) to the levelled height about its computed value.
differences along the levelling lines. Such correction
compensates for the error arising from the non-parallel Motivated from the above, the objective of the current
geo-potential surfaces. So, via the application of OC, study is to compare the above two Helmert OC
spirit levelling loop closures would theoretically vanish formulae, regarding the accuracy and precision of the
(Sanso and Vaniček, 2006). For this purpose, gravity resulting OC. Such investigation will comprise two
information should be available at appropriate sections levelling lines in Egypt with observed elevations and
along the levelling route (Heiskanen and Moritz, 1967). gravity data. The first line runs along the Nile valley,
while the second is located in the Western desert. The
The impact of the accuracies of both gravity data and two lines exhibit relatively moderate and mountainous
benchmarks' heights, on the computed OC, was terrain roughness, respectively. The qualities of geo-
previously studied by Filmer and Featherstone (2011). potential model derived gravity and digital terrain
In this respect, besides Helmert OC, other two types of model (DTM) derived benchmark elevations are used
height corrections were considered. Also, the use of for a practical comparison among the two formulae over
global geo-potential models derived gravity in Helmert the levelling lines. In this respect, the UCPH2002 geo-
OC assessment was investigated by Filmer et al. (2013) potential model (Howe and Tscherning, 2002) is used,
and Hassouna (2013). Regarding Helmert OC accuracy, up to degree and order 90. Also, the SRTM30 global
such studies considered the common OC formula elevation model (USGS, 2006) is utilized.
derived by Heiskanen and Moritz (1967). Also, no
explicit investigation has been carried out on the OC 2. OC error analysis: comparative algorithms
precision, as expressed by its variance.
The OC along a spirit levelling line, AB, is commonly
Hwang and Hsiao (2003) introduced a new formula for expressed by (Heiskanen and Moritz, 1967)
the computation of Helmert OC. However, the qualities
of OC, as computed from the two formulae, have never k  gi  g  g 
been compared. Such comparison could flag the
specific formula that is more accurate and/or more
OC   
i 1   0
 1 
 i

n   A
 0
 1  H  
 A  0
B
 1 HA ,

   
precise. Obviously, an accurate OC model would be
more efficient if the input gravity and/or benchmark
(1)
elevations exhibit some biases. A good example for
gravity bias is the omission error of a geo-potential
model, from which gravity data are derived. On the where
other hand, a more precise OC implies an optimal
propagation of the input data random error through the

© Indian Society of Geomatics

 103

Journal of Geomatics Vol 10 No. 1 April 2016

n i the geometric height increment relevant to the OC n i

 , (4a)
th
i levelling section, g i 0
k the number of levelling sections,
OC gi
gi the observed gravity relevant to the ith section,   1, (4b)
n i 0
0 the normal gravity at geodetic latitude 45° on
OC HA
the WGS-84 ellipsoid, which can be computed  , (4c)
according to Moritz (1980), g A 0
HA the elevation of the start benchmark A, OC HB
 , (4d)
HB the elevation of the end benchmark B, g B 0
gA the mean gravity along the plumb line at A OC gA
  1,
g 
(4e)
A
 g A  0.0424H A , H A 0

gB OC g 
the mean gravity relevant to B   B
 1 , (4f)
 0 
g B
 g B  0.0424H B ,  H B  
g A & g B the observed gravity at A and B, respectively.  g A   g A  0.0424 H A , (4g)

 g B   g B  0.0424 H B . (4h)
Equivalently, but alternatively expressed, the OC could
be formulated as follows (Hwang and Hsiao, 2003) So, Eq. (3) may be re-written as follows
k g  g  g 
 n i
k k
H
OC   i  1 n i   A  1 H A . (2)  OC    g i    i  1n i  A  g A 
g   B
g  0 i 1   0  0
i 1  B  i 1

Based on the different functional models in Eqs. (1) and

HB g  g 
(2), one could in general expect different error  g B   A  1  H A   B  1  H B .
propagation characteristics. The following two sub- 0  0   0 
   
sections will explore both the systematic and random
(5)
error analysis features for OC as estimated from Eqs.
(1) and (2).
On the other hand, the total differential of Eq. (2) can
be expressed by
2.1 OC systematic error analysis

In the sense of Schofield and Breach (2007, Eq. 2.6

OC   OC  g i   OC ni  OC  g A (6)
k k
therein), it might be recognized that the resultant
systematic error in OC could be assessed simply via the i 1 g i i 1 ni g A
total differential. Thus, applying the total differential
operator to Eq. (1), it follows that the resultant  OC  g B  OC  H A ,
g B H A
systematic error in OC,  OC , may be expressed as
with
k
OC k
OC OC
 OC    gi   n i  gA  OC n i
i 1 g i i 1 n i g A  , (7a)
g i gB
OC OC OC
gB  HA  HB , OC  gi 
g B H A H B   1 , (7b)
(3) n i  gB 
where  g i , n i ,  g A ,  g B ,  H A and  H B
denote the systematic errors in the respective input OC HA
 , (7c)
quantities. The partial derivatives and  g A ,  g B in g A gB
Eq. (3) can be derived as follows (Filmer and
Featherstone, 2011) OC 1  k

g B
 2   g i n i  H A g A  ,
 (7d)
gB i 1
 104

Journal of Geomatics Vol 10 No. 1 April 2016

2
 gA 
2 2
OC  2OC  
k  ni  k 
 2g i    gg i

1  2ni  
HA 
 2g A 
  1 . (7e) 


 


H A i 1  g B  i 1    gB 
 gB 
B

2 2
 1 k 
Accordingly, Eq. (6) may be formulated as follows

2  g i ni  H A g A    2 g B   gg A 1  2H A .
k
n i g  H k
 B   g  i 1  
 OC    g i    i  1 n i  A  g A 
 B
(12)
i 1 gB i 1  gB  gB
1  k  g  Also, using the law of error propagation, it follows that
2 
g i n i  H A g A   g B   A  1  H A .
g B  i 1   gB   2 g A   2 g A  (0.0424)2  2 H A , (13a)
(8)

2.2 OC random error analysis  2 g B   2 g B  (0.0424)2  2 H B . (13b)

The impact of random errors on OC follows from the 3. Input data and input errors for the test levelling
application of the law of variance-covariance lines
propagation to Eqs. (1) and (2). So, neglecting the error
covariances and applying the error propagation The current study considers two levelling lines in Egypt
principle to Eq. (1), the OC variance,  2OC , is with observed elevations and gravity data. The first line
expressed by (e.g. Ghilani and Wolf, 2006) (Line I) runs along the Nile valley, while the second
(Line II) is located in the Western desert. Line I and line
2 2 2 II exhibit relatively moderate and mountainous terrain
k   k    
 2OC    OC   2 g i    OC   2ni   OC   2 g A  roughness, respectively. Figure 1 depicts two post maps
i 1  g i  i 1  ni  
g A  for the gravity points along both lines. While the
available terrestrial gravity and elevation data are used
2 2 2 as real world reference values in the current
 OC   OC   OC 
   gB 
2    HA 
2    HB ,
2 investigation, the input "erroneous" gravity and height
 g B   H A   H B  information (input into the error analysis algorithms)
     
(9) are derived from the UCPH2002 geo-potential model
and the SRTM30 terrain model, respectively. First, geo-
where  g i , n i ,  g A ,  g B ,  H A and  H B
potential model derived surface gravity was derived as
the sum of the UCPH2002-synthesized surface gravity
stand for the error standard deviations relevant to the disturbances relative to WGS-84 and the respective
input quantities, which reflects their precisions. The normal gravity values (Forsberg and Tscherning, 2008;
partial derivatives in Eq. (9) are the same derived in Eq. Filmer and Featherstone, 2011; Filmer et al., 2013;
(4). So, Eq. (9) can be written as follows Hassouna, 2013). On the other hand, the SRTM30 grid
was used to interpolate respective elevation data at the
2
k  n 2 k g   
2 points of lines I and II, via the B-spline interpolation
 2OC     i   2 g i     i 1  2ni   H A   2 g A  algorithm (Cimmery, 2010).
i 1  0  i 1 0  0   
Table 1 lists the different features of the two levelling
2 2 2 lines under consideration. Table 2 shows the statistics
HB     

 


 2 g B   g A 1  2H A   g B 1  2H B . of the differences among the derived "erroneous"
 0   0
  0
 gravity and benchmark elevation values; and the
(10) respective reference values over the investigated lines.

4. Numerical error analyses comparison

Similarly, applying the law of variance-covariance
propagation to Eq. (2), Regarding OC accuracy, as given by Eqs. (5) and (8),
the section-wise varying  g i in Table 2 were used to
2 2 2 represent the respective gravity biases, while  g A and
 2OC    OC 
 2 g i    OCn   2ni   OC 
  2g A 
k k
   g B were supposed to be the gravity biases at the start
  g 
i 1  g i  i 1  i   A 
and end benchmark, respectively. While such gravity
2 2 biases represent the geo-potential model omission error,
 OC 

 g


 2 g B   OC
H 
  HA,
 2 it could be comparable to gravity biases that could arise
 B   A  from the interpolation of the respective gravity values
(11) from a specific gravity data base. In the same manner,
 H A and  H B (in Table 2) were used as the elevation
which after using Eq. (7) gives
biases relevant to the start and end benchmarks. Such
 105

Journal of Geomatics Vol 10 No. 1 April 2016

biases might seem realistic for the actual lines I and II, gravity discrepancies,   g , as representing  g i over
provided that the typical surveyor, although could i
perform a precise spirit levelling, might not have access each levelling line, as given in Table 2.
to the official elevations for all benchmarks in a
levelling network. Accordingly, a way out is to use Similarly, the spatial precision of the elevations
interpolated elevations for the assessment of OC discrepancies in Table 2,  H , was used to represent
(Filmer and Featherstone, 2011). i
both  H and  H for each levelling line. Such
A B
values were used to evaluate the OC precision,  OC ,
using Eqs. (10) and (12) for the two lines. Also, as far
as the aim of the current work is to compare the
accuracy and precision of two formulae in furnishing
the OC, the very pessimistic biases and standard
deviations in Table 2 could help test the sensitivity of
the two formulae to systematic and random error
propagation. In particular, interpolated benchmark
elevations and synthesized or interpolated gravity data
could exhibit such bad qualities. It should be
emphasized that spirit levelling is characterized by high
quality height increments. So, the height increments,
n i , were considered errorless over the whole
computations.

During the assessment of the OC quality features, using

Eqs. (5), (8), (10) and (12), the reference gravity,
elevations and height increment values were used as the
evaluation points of the respective partial derivatives.
Also, it should be kept in mind that the input biases and
uncertainties of gravity and benchmark elevations were
marginally treated. In other words, as will be depicted
by Tables 3 and 4, if the gravity accuracy or precision
is emphasized, the benchmark elevations are assumed
errorless and vice versa. Such separation in error
analysis could enable individual investigations of the
effect of gravity and benchmark elevations' qualities on
that of the OC.
It is worth mentioning that  g A 
&  g B and

 g A
& g B  were appropriately computed from
Figure 1: Post maps for Lines (I) and (II) Eqs. (4g & 4h) and (13), respectively, taking into
account whether or not the gravity or benchmark
elevations are kept errorless.
Table 1: Different features of the two levelling
profiles In the above sense, Tables 3a and 3b summarize the
accuracies of OC, as computed from Eq. (5) and Eq. (8).
No. of Mean Also, Tables 4a and 4b show the respective items, but
H Length regarding the precisions of OC determination obtained
Line gravity elevation
(m ) (km ) via Eqs. (10) and (12).
sections (m )
I 93 -24.68 34.85 280.4
5. Discussion and concluding remarks
II 78 -391.82 268.18 369.2
Tables 3a and 3b show that in general, the accuracies of
The observational noises of the observed (reference) OC determination by the two formulae deteriorate as
gravity data for the test lines are in the order of 1 mgal. the terrain roughness increases. Table 3a implies that
Also, the error budget for the reference elevations of the Eqs. (1) and (2) yield equally accurate OC values, if one
start and end benchmarks could be as worse as a few deals with highly accurate (i.e. nearly errorless)
centimeters. However, such small uncertainties might benchmark elevations. Table 3a shows also that in such
not represent those of any eventual interpolated gravity case, geo-potential models derived gravity could safely
or elevation values, which are greatly affected by the be used to assess quite accurate OC values in case of
local gravity and terrain signal roughness. So, it was moderate terrain roughness, as that represented by line
decided to use the spatial standard deviations of the I. This is easy to conclude from the respective two zero
 106

Journal of Geomatics Vol 10 No. 1 April 2016

values for  OC . Numerically, this can be attributed to Eq. (1), compared to that assessed by Eq. (2). Regarding
the insignificance of the non-vanishing terms in Eqs. (5) line I, such deterioration is less pronounced. So, in
and (8) for line I. On the other hand, Table 3b shows a general, Eq. (2) yield more accurate OC magnitudes, if
dramatic deterioration of OC accuracy, if computed via interpolated benchmark heights are to be used along
with highly accurate gravity data.

Table 2: Statistics of the discrepancies among the erroneous and reference quantities for the two lines

Max .
 g
Min .
Mean
g g
Line Item Unit
g
i g i
g i
A B

I 4.65 8.33 -17.46 22.70 11.50 20.78

 gUCPH 2002  g observed  ( mgal )
II 23.79 13.17 3.00 50.43 10.35 17.95

Line Item Unit

Mean  H Min . Max .
H H
H H H
i
A B
i i i

I -2.26 1.51 -5.71 1.83 -3.34 -2.37

 H SRTM 30  H observed  (m )
II -2.66 32.90 -133.39 200.11 14.94 -8.50

Table 3a: Comparison among the OC accuracies Tables 4a and 4b. So, Eq. (2) is more precise than Eq.
(gravity accuracy emphasized) (1), when interpolated benchmark elevations are used.
 gi  gA  gB  OC  OC So, generally, for a high quality OC determination, it is
Line ( mm ) ( mm ) recommended to use Eq. (2). Such results should be
( mgal ) ( mgal ) ( mgal ) understood to hold to the investigated interpolated
(Eq. 5) (Eq. 8) benchmark elevations and synthesized gravity values
I Section- 11.50 20.78 0.0 0.0 along levelling lines. Besides being more economic,
wise such levelling data sources may be the only available
II varying 10.35 17.95 -4.6 -4.6
tool for assessing OC. The use of the SRTM30 and the
Table 3b: Comparison among the OC accuracies UCPH2002 low resolution models in the current study
(elevation accuracy emphasized) was for the sake of handling an instance worst case. A
HA HB  OC  OC future work may test other high resolution DTMs and
Line ( mm ) ( mm ) geo-potential models for representing erroneous data.
(m ) (m )
(Eq. 5) (Eq. 8)
A further related criterion could be the assessment of the
I -3.34 -2.37 2.0 0.6
values and qualities of OCs for height differences along
II 14.94 -8.50 -40.2 -4.7 profiles, which are extracted from DEMs. Such terrain
models may be derived from satellite imagery or
Table 4a: Comparison among the OC precisions photogrammetry. In this respect, the published
(gravity precision emphasized) uncertainty of elevations would often be of the order of
 gi g g  OC  OC that discussed in the current study. Accordingly, Eq. (2)
Line A B
( mm ) ( mm ) is expected to show better error propagation
( mgal ) ( mgal ) ( mgal ) characteristics.
(Eq. 10) (Eq. 12)
I 8.33 0.5 0.5 Finally, it could seem rational and innovative to model
II 13.17 8.3 8.3 the OCs as biases over levelling circuits in vertical
networks. In this sense, OCs over levelling lines might
Table 4b: Comparison among the OC precisions be treated as observed weighted parameters during the
(elevation precision emphasized) adjustment of levelling nets (e.g. Ghilani and Wolf,
2006). Of course, the weights of such parameters should
H and H B
 OC  OC lean on the respective error standard deviations, and
hence, the importance of OC precision may arise. This
A
Line ( mm ) ( mm )
(m ) in turn could optimally help pick out the random loop
(Eq.10) (Eq. 12) closures parts during the adjustment.
I 1.51 3.1 0.3
II 32.90 78.7 11.3 Acknowledgment

Three unknown reviewers are acknowledged for their

Regarding OC uncertainty, almost all the above critical review of the manuscript.
comments relevant to Tables 3a and 3b still apply for
 107

Journal of Geomatics Vol 10 No. 1 April 2016

References Engineering in Shebin El-Kom, Menoufia University,
Egypt,36(4), pp. 475-479.
Cimmery, V. (2010). User guide for SAGA (version
2.0.5). Volumes 1 and 2, August, 2010, SAGA System Heiskanen, W.A. and H. Moritz (1967). Physical
for Automated Geoscientific Analyses. Available: Geodesy. W.H. Freeman and Company.
http:// www. saga-gis.org.
Howe, E. and C.C. Tscherning (2002). Analysis of one
Filmer, M.S. and W.E. Featherstone (2011). Error month of CHAMP state vector and accelerometer data
propagation for three common height-system for the recovery of the gravity potential. Poster
corrections to differential levelling. Journal of Spatial presented at the XXVII General Assembly of the
Science 56(1), pp. 39-58. European Geophysical Society, Nice, France, April 21-
26.
Filmer, M.S., C. Hirt and W.E. Featherstone (2013).
Error sources and data limitations for the prediction of Hwang, C. and Y.S. Hsiao (2003). Orthometric
surface gravity: a case study using benchmarks. Stud. corrections from levelling, gravity, density and
Geophys. Geod., 57(1), pp. 47-66. elevation data: a case study in Taiwan. Journal of
Geodesy, 77(5-6), pp. 279-291.
Forsberg, R. and C.C Tscherning (2008). An overview
manual for the GRAVSOFT geodetic gravity field Moritz, H. (1980). Geodetic reference system 1980.
modelling programs. 2nd ed., National Space Institute Bulletin Géodésique, 54(3), pp. 395-405.
(DTU-Space), Denmark, August.
Sanso, F. and P. Vaniček (2006). The orthometric
Ghilani, C. and P. Wolf (2006). Adjustment height and the holonomity problem. Journal of
computations: Spatial data analysis. 4th ed., John Wiley Geodesy, 80(5), pp. 225-232.
& Sons, Hoboken, New Jersey.
Schofield, W. and M. Breach (2007). Engineering
Hassouna, R. (2013). Geo-potential models as a tool of surveying (6th ed.). Elsevier.
densifying gravity for orthometric correction
computation. Engineering Research Journal, Faculty of USGS (2006). SRTM30. Available: http://dds.cr.usgs.
gov/srtm/.
 v

Journal of Geomatics Vol 10 No. 1 April 2016

Guidelines for "National Geomatics Award for Excellence”

This award has been instituted to recognize outstanding and conspicuously important contribution in promoting geomatics
technology and applications at the country level. The contributions should have made major impact on the use of this
technology for national development.

Areas of contribution considered for the award are:

1. Geographical Information System

2. Global Positioning System
3. Photogrammetry
4. Digital Cartography

The award shall consist of Rs. 50,000/- in cash, a medal and citation.

Eligibility

Any citizen of India, engaged in activities related to geomatics technology and its applications is eligible for this award. The
prize is awarded on the basis of work primarily done in India.

The age limit for awardees is 45 years or above as on June 30 of the year of award.

Selection

A duly constituted Award Committee will evaluate all nominations received. The committee shall consist of eminent experts
in the field of geo-spatial technology, to be identified by the Executive Council, ISG. The committee shall forward selected
name/s to ISG – EC for approval and announcement. Apart from those persons, whose nominations have been received, the
Committee may consider any person or persons who, in their opinion, have made outstanding contributions to development
of geo-spatial technology and applications.

The award can be withheld in any year if, in the opinion of the committee, no candidate is found suitable in that particular
year.

Presentation of the Award

The award shall be presented during the Annual Convention of ISG. Local Hospitality shall be taken care by ISG & Air fare
(low cost) may be reimbursed if awardees request for it.

How to make Nomination

The nominations can be proposed by Head of a major research institute/ centre; Vice-Chancellor of a university; Secretary
of Government Scientific Departments; President of a National Academy, President, Indian Society of Geomatics / Indian
Society of Remote Sensing / Indian National Cartographic Association / ISG fellow or two life members of the society with
more than 10 year old membership.

A candidate once nominated would be considered for a total period of two years. Nomination should be sent in the prescribed
format to Secretary, ISG.

The last date for receiving nominations shall be September, 31 or otherwise extended.
 vi

Journal of Geomatics Vol 10 No. 1 April 2016

Nomination Format for "National Geomatics Award for Excellence”

1. Name of the Nominee

2. Postal Address
3. Academic Background (Bachelor degree onwards)
4. Field of Specialisation
5. Important positions held (in chronological order)
6. Professional Experience including foreign assignments.
7. Important Awards / Honours
8. Important Publications/Patents: (A set of ten most important publications to be enclosed with this form)
9. Contributions of Nominee based on which the nomination is sent (in 1000 words, also provide a statement
in 50 words which may be used for citation.):
10. Other Relevant Information:

Proposer:

Signature
Name
Address
Phone/ Fax
E-mail
Life Membership No. (in case of ISG Member):

Place & Date

Endorsed by (in case nomination is by 2 ISG Life members)

Signature
Name
Address
Phone/ Fax
E-mail
Life Membership No. (in case of ISG Member):

Place & Date

(The proposer should give a brief citation of the nominee's work)
 vii

Journal of Geomatics Vol 10 No. 1 April 2016

National Geomatics Award

National Geomatics Award to be given each year: a) for original and significant contribution, b) for innovative applications
in the field of Geomatics. Each award comprises a medal, a citation and a sum of Rs 25,000/- The guidelines for these awards
are available on ISG website.

ISG Chapter Award for Best Performance

The best chapter award will be given to an active chapter of Indian Society of Geomatics, which has made significant
contribution to further the mandate and goal of the society. The award consists of a citation and medal

President’s Appreciation Medal for Contribution to the ISG

This award will be given to a member of the society, who has made noteworthy contribution to the growth of the ISG (its
main body or any chapter). The Award consists of a Medal and a Citation.

Prof. Kakani Nageswara Rao Endowment Young Achiever Award

Indian Society of Geomatics instituted a new award from year 2013 named “Prof. Kakani Nageswara Rao Endowment Young
Achiever Award”, to encourage young researchers/scientists/academicians pursuing research in the field of geospatial
technology/applications. The award carries a cash prize of Rs.10,000/- along with a citation.
 viii

Journal of Geomatics Vol 10 No. 1 April 2016

NATIONAL GEOMATICS AWARDS 2016

Indian Society of Geomatics has instituted two National Geomatics Awards to be given each year for a Original and
significant contribution, b Innovative application(s) in the field of Geomatics. Each award comprises a medal, a citation and
a sum of Rs. 25,000/-.

The guidelines for the award are as under

Areas of contribution considered for the award

1. Geographical Information System

2. Global Positioning System

3. Photogrammetry

4. Digital Cartography

Eligibility

Any citizen of India engaged in scientific work in any of the above-mentioned areas of research is eligible for the award.

The awards are to be given for the work largely carried out in India.

 First award will be given for original contribution in the field of Geomatics supported by publications in a refereed
journal of repute.

 Second award will be given for carrying out innovative application(s).

 The contribution for the first award should have been accepted by peers through citation of the work.

 Work based on the applications of existing technologies will not be considered for the first award.

 The work should have made impact on the overall development of Geomatics.

How to Send Nomination

Nominations should be sent in the prescribed format, completed in all aspects to the Secretary, Indian Society of Geomatics,
Space Applications Centre Campus, Ahmedabad 380 015 by August 31, 2016.

Selection Process

An expert committee, consisting of at least three members, constituted by the Executive Council of the Indian Society of
Geomatics, will scrutinize the nominations and recommend the awardees’ names to the Executive Council. The Council will
decide on the award based on the recommendations.
 ix

Journal of Geomatics Vol 10 No. 1 April 2016

FORMAT FOR AWARD NOMINATION

1. Name of the Candidate:

2. Present Position:
3. Positions held earlier (chronological order):
4. Academic qualifications (Bachelor’s degree onwards):
5. Names of at least three Indian Scientists/Technologist in the area as possible referees *:
6. Brief write up on the work (500 words) for which award is claimed:
7. Publication(s) on the above work (reprint(s) to be enclosed):
8. List of other publications of the candidate:
9. Citation of the work for which award is claimed:
10. Impact of the work (for which award is claimed) on the development in the field of Geomatics (500 words):
11. Whether the work has already won any award? If so, give details:

The Applications in the above format (five copies) should be submitted (by Registered Post or Speed Post) to

The Secretary, Indian Society of Geomatics,

Space Applications Centre Campus,
Ahmedabad-380015

so as to reach by August 31, 2016.

*ISG is, however, not bound to accept these names and can refer the nomination to other experts/peers
 x

Journal of Geomatics Vol 10 No. 1 April 2016

INDIAN SOCIETY OF GEOMATICS

FELLOWS
Sh ri Pramod P. Kale, Pune
D r G e o rg e J o s e p h , A h me d a b a d
D r A . K . S. G o p a l an , H y d e ra b a d
D r Pri t h v i s h N a g , V a ra n a s i
D r B a l d e v Sa h a i, A h me d a b a d
Sh ri A . R . D a s g u pt a , A h me d a b a d
D r R . R . N a v a l gu n d , B a n g a lu ru
Sh ri R a j e s h M a t h u r, N e w D e l hi
D r A j a i, A h me d a b a d
Pr o f P. V e n k a t a ch a l a m, M u mb a i

ISG - PATRON MEMBERS

P-1 Director, Space Applications Centre (ISRO), Jodhpur Tekra Satellite Road, Ahmedabad - 380 015
P-2 Settlement Commissioner, The Settlement Commissioner & Director of Land Records-Gujarat, Block No. 13, Floor 2, Old Sachivalay, Sector-10,
Gandhinagar – 382 010
P-3 Commissioner, Mumbai Metropolitan Region Development Authority, Bandra-Kurla Complex, Bandra East, Mumbai - 400 051
P-4 Commissioner, land Records & Settlements Office, MP, Gwalior - 474 007
P-5 Director General, Centre for Development of Advanced Computing (C-DAC), Pune University Campus, Ganesh Khind, Pune - 411 007
P-6 Chairman, Indian Space Research Organization (ISRO), ISRO H.Q., Antariksha Bhavan, New BEL Road,Bengaluru 560 231
P-7 Director General, Forest Survey of India, Kaulagarh Road, P.O. I.P.E., Dehra Dun – 248 195
P-8 Commissioner, Vadodara Municipal Corporation, M.S. University, Vadodara - 390 002
P-9 Director, Centre for Environmental Planning and Technology (CEPT), Navarangpura, Ahmedabad - 380 009
P-10 Managing Director, ESRI INDIA, NIIT GIS Ltd., 8, Balaji Estate, Sudarshan Munjal Marg, Kalkaji, New Delhi - 110 019
P-11 Director, Gujarat Water Supply and Sewerage Board (GWSSB), Jalseva Bhavan, Sector – 10A, Gandhinagar - 382 010
P-12 Director, National Atlas & Thematic Mapping Organization (NATMO), Salt Lake, Kolkata - 700 064
P-13 Director of Operations, GIS Services, Genesys International Corporation Ltd., 73-A, SDF-III, SEEPZ, Andheri (E), Mumbai - 400 096
P-14 Managing Director, Speck Systems Limited, B-49, Electronics Complex, Kushiaguda, Hyderabad - 500 062
P-15 Director, Institute of Remote Sensing (IRS), Anna University, Sardar Patel Road, Chennai - 600 025
P-16 Managing Director, Tri-Geo Image Systems Ltd., 813 Nagarjuna Hills, PunjaGutta, Hyderabad - 500 082
P-17 Managing Director, Scanpoint Graphics Ltd., B/h Town Hall, Ashram Road, Ahmedabad - 380 006
P-18 Secretary General, Institute for Sustainable Development Research Studies (ISDRS), 7, Manav Ashram Colony, Goplapura Mod, Tonk
Road, Jaipur - 302 018
P-19 Commandant, Defense institute for GeoSpatial Information & Training (DIGIT), Nr. Army HQs Camp, Rao Tula Ram Marg, Cantt., New
Delhi - 110 010
P-20 Vice President, New Rolta India Ltd., Rolta Bhavan, 22nd Street, MIDC-Marol, Andheri East, Mumbai - 400 093
P-21 Director, National Remote Sensing Centre (NRSC), Deptt. of Space, Govt. of India, Balanagar, Hyderabad - 500 037
P-22 Managing Director, ERDAS India Ltd., Plot No. 7, Type-I, IE Kukatpalli, Hyderabad - 500 072
P-23 Senior Manager, Larsen & Toubro Limited, Library and Documentation Centre ECC Constr. Gp., P.B. No. 979, Mount Poonamallee
Road, Manapakkam, Chennai - 600 089.
P-24 Director, North Eastern Space Applications Centre (NE-SAC), Department of Space, Umiam, Meghalaya 793 103
P-25 Progamme Coordinator, GSDG, Centre for Development of Advanced Computing (C-DAC), Pune University Campus, Pune –
411 007
P-26 Chief Executive, Jishnu Ocean Technologies, PL-6A, Bldg. No. 6/15, Sector – 1, Khanda Colony, New Panvel (W), Navi Mumbai – 410 206
P-27 Director General, A.P. State Remote Sensing Applications Centre (APSRAC), 8th Floor, “B” Block, Swarnajayanthi Complex,
Ameerpet, Hyderabad- 500 038
P-28 Director, Advanced Data Processing Res. Institute (ADRIN), 203, Akbar Road, Tarbund, Manovikas Nagar P.O., Secunderabad –
500 009
P-29 Managing Director, LEICA Geosystems Geospatial Imaging Pvt. (I) Ltd., 3, Enkay Square, 448a Udyog Vihar, Phase-5,
Gurgoan- 122 016
P-30 Director, Defense Terrain Research Limited (DTRL), Ministry of Defense, Govt. of India, Defense Research & Development
Organisation, Metacafe House, New Delhi – 110 054
P-31 Chairman, OGC India Forum, E/701, Gokul Residency, Thakur Village, Kandivali (E), Mumbai – 400 101
P-32 Managing Director, ML Infomap Pvt. Ltd., 124-A, Katwaria Sarai, New Delhi – 110 016
P-33 Director, Rolta India Limited, Rolta Tower, “A”, Rolta Technology Park, MIDC, Andheri (E), Mumbai – 400 093
P-34 Director, State Remote Sensing Applications Centre, Aizawl – 796 012, Mizoram
 xi

Journal of Geomatics Vol 10 No. 1 April 2016

Instructions for Authors

The journal covers all aspects of Geomatics – geodata acquisition, Normally length of a published paper should be about 6-10 pages
pre-processing, processing, analysis and publishing. Broadly this in A4 size including figures. Use of illustrations in colour should
implies inclusion of areas like GIS, GPS, Photogrammetry, be restricted and resorted to only where it is absolutely necessary
Cartography, Remote Sensing, Surveying, Spatial Data and not for enhancing the look of the paper. If the number of
Infrastructure and Technology including hardware, software, colour illustrations exceeds five, authors’ institution may be
algorithms and model. It endeavours to provide an international asked to reimburse the extra cost involved, which at current rates
forum for rapid publication of developments in the field – both in is about Rs. 2500 per coloured figure/diagram/plate/illustration.
technology and applications.
Submission of Manuscript
To begin with the frequency of publication will be six- monthly.
However, depending on the response and interest, frequency of Submissions should be in electronic form via email. The
publication may be reviewed. manuscript may be sent by email to [email protected]. In
exceptional cases hard copy submission in camera ready form may
A manuscript for publication must be based on original research be allowed with the prior permission of the Chief Editor.
work done by the author(s). It should not have been published in Submission in any other form will be returned to the author. To
part or full in any type of publication nor should it be under speed up the review process, authors are advised to provide a list
consideration for publication in any periodical. Unsolicited review of three probable reviewers with their institutional address and e-
papers will not be published. mail IDs.

The Editorial Board or the Indian Society of Geomatics is not Guidelines for Citing References
responsible for the opinions expressed by the authors.
Names of all cited publications should be given in full.
Language No abbreviations should be used. Following procedure is
to be adopted.
The language of the Journal will be English (Indian). However,
manuscripts in English (US) and English (British) are also Journal Publications
acceptable from authors from countries located outside India.
Bahuguna, I.M. and A.V. Kulkarni (2005). Application of digital
Manuscript Format elevation model and orthoimages derived from IRS-1C Pan
stereo data in monitoring variations in glacial dimensions,
Each paper should have a title, name(s) of author(s), and affiliation Journal of the Indian Society of Remote Sensing, 33(1), 107- 112.
of each of the authors with complete mailing address, e-mail (to be referred to in the text as Bahuguna and Kulkarni (2005) or
address, an abstract, four to six keywords, and the text. The text if more than two sets of authors are to be referred to, as (Bahuguna
should include introduction/background, research method, results, and Kulkarni, 2005; Jain et al., 1994)) When more than two
discussion, followed by acknowledgements and references. The authors are to be referred to, use Jain et al. (1994). However, in
main text should be divided in sections. Section headings should References, all authors are to be mentioned.
be concise and numbered in sequence, using a decimal system for
subsections. Figures, images and their captions should be inserted Publication in a Book
at appropriate points of the text. Figures, images and tables should
fit in two column format of the journal. If absolutely necessary, Misra, V.N. (1984). Climate, a factor in the rise and fall of the
figures, images and tables can spread across the two columns. Indus Civilization – Evidence from Rajasthan and Beyond in
Figures and images, however, should not exceed half a page in Frontiers of the Indus Civilization (B.B. Lal and S.P. Gupta: Chief
height. A title should be provided for each Table, Image and Editors) Books and Books, New Delhi, pp. 461-489
Figure. All figures and images should be in 600 dpi resolution and
sized as per column/margin width. Authors must ensure that Papers Published in Seminar/ Symposium
diagrams/figures should not lose easy readability upon reduction Proceedings
to column size. The SI (metric) units and international quantities
should be used throughout the paper. In case measurements are Jain, A., A.R. Shirish, M. Das, K. Das, M.C. Porwal, and P.S. Roy
given in any other system, equivalent measurements in SI (metric) (1994). Remote Sensing and Geographic Information System – An
units should be indicated in brackets. approach for the assessment of biotic interference in the forest
ecosystem. Proceedings. 15th Asian Conference on Remote
Use MS Word with English (UK/US) or English (Indian) Sensing,Bangaluru, November 17-23, 1994, pp. 65-72.
dictionary. The page size should be A4 paper, with 2 cm margin
on all sides. Title, authors and affiliation should be centred. Books
Abstract should be justified across margins. The manuscript text
should be in two columns of 8.2 cm each with a gutter of 6mm Possehl, Gregory L. (1999). Indus Age: The beginnings. Oxford
between them. Use only Times New Roman fonts. Title should be and IBH Publishing Corporation, New Delhi.
12 points bold. Authors and affiliation should be 9 points. All
other text including headings should be 10 points. Heading
numbering scheme should be decimal e.g. 1, 1.1, 1.2.3, etc.
Headings should be in bold.
 xii
Journal of Geomatics Vol 10 No. 1 April 2016

Reviewing
If the authors have used any copyright material in their
Each paper will be reviewed by three peers. Papers forwarded by manuscript, it is understood that they have obtained permission
members of the Editorial or Advisory Boards along with their from the owner of the copyright material and they should convey
comments would get processed faster and may be reviewed by two the same along with the manuscript to the Chief Editor.
referees only.
Certificate of Original Work
Sample format for Authors is available in downloadable form at
ISG website: www.isgindia.org/JOG/Sample_format.doc The authors will also provide a certificate that the paper is an
original work, not published or being considered for publication
Copyright elsewhere.

The copyright of the paper selected for publication will rest with In the event the certificate turns out to be false, the Journal shall
the Indian Society of Geomatics. Corresponding author shall be ban the author(s) from publishing in the Journal for a period of
required to sign a copyright assignment form, on behalf of all five years and inform the same to all other related publications.
authors, once the paper is selected for publication. Authors are,
however, at liberty to use this material elsewhere after obtaining Reprints
permission from the Indian Society of Geomatics.
Authors will be provided soft copy (PDF) of their paper, no hard
copy reprints will be provided.
 xiii
Journal of Geomatics Vol 10 No. 1 April 2016

Journal of Geomatics
Advertisement Rates
1 Issue 4 Issues
Back Cover Page in colour Rs. 25,000 Rs. 80,000
Inside Cover Page in colour Rs. 20,000 Rs. 64,000
Full Page inside in colour Rs. 15,000 Rs. 48,000
Full Page inside in B/W Rs. 10,000 Rs. 32,000

Advertisement Details

Mechanical Details
Double Spread/Center Spread (42 x 29.7) cm
Full page bleed (21 x 29.7) cm
Full page non-bleed (19 x 27.7) cm

Art Requirements

Negatives: Art must be right reading, emulsion, down. Film must be supplied in one piece per color, each identified
by color. Camera-ready art is accepted for black & White adds; however, film is preferred. Electronic Files are also
accepted.

Electronic File Requirements: All material must be received before ad close dates.

Software: Adobe illustrator 9.0 (saved as EPS). Adobe Photoshop CS (saved as EPS or TIFF). Please convert higher
versions down. If you can only supply an IBM format, the file must be in viewable EPS or TIFF format with fonts
embedded as that format.

Colour Ads: Colour separations must be provided, right reading, emulsion down. Please note that files using RGB or
Pantone colours (PMS) must be converted to CMYK before we receive files.
 xiv

Journal of Geomatics Vol 10 No. 1 April 2016

Indian Society of Geomatics (ISG)

(www.isgindia.org)

Membership Application Form

To

The Secretary
Indian Society of Geomatics
39, Basant Bahar - II
Bopal, Ahmedabad – 380058

Sir,

I want to become a Member/ Life Member/ Sustaining Member/ Patron Member/ Foreign Member/ Student Member of
the Indian Society of Geomatics, Ahmedabad for the year ______. Membership fee of Rs. ______/- is being sent to you
by Cash/DD/Multicity Cheque. (In case of DD/ Cheque No. _________, drawn on Bank
________________________________.

I agree to abide by the Constitution of the Society.

Date:

Place:
Signature

1. Name:___________________________________________________________________________________________
2. Address:_________________________________________________________________________________________
_____________________________________________________________________________________ PIN: __________
Phone:________________ Fax: _______________ Email: ____________________________________________________

3. Date of Birth: ____________________________________________________________________________________

4. Qualifications ____________________________________________________________________________________
5. Specialisation: ____________________________________________________________________________________
6. Designation & Organisation.: ________________________________________________________________________
7. Membership in other Societies: _______________________________________________________________________
____________________________________________________________________________________________________
8. Mailing Address: __________________________________________________________________________________
_____________________________________________________________________________________ PIN: __________
Proposed by:
(Member’s Name and No)

Signature of Proposer

For Office Use

ISG Membership No:
ISG Receipt No. : Date:
 xv

Journal of Geomatics Vol 10 No. 1 April 2016

MEMBERSHIP SUBSCRIPTION

Sr. Membership Admission Fee Annual Subscription

No. Category Rs. (Indian)
Rs. (Indian) US$ (Foreign) 300.00
1. Annual Member 10.00 -
2. Life Member
a) Admitted before 45 years of age 2500.00 250.00
b) Admitted after 45 years of age 2000.00 200.00
3. Sustaining Member - 2000.00
4. Patron Member 5000.00 3000.00
5. Student Member 10.00 - 100.00

MEMBERSHIP GUIDELINES

1. Subscription for Life Membership is also accepted in two equal installments payable within a duration of
three months, if so desired by the applicant. In such a case, please specify that payment will be in
installments and also the probable date for the second installment (within three months of the first
installment).
2. A Member of the Society should countersign application of membership as proposer.
3. Subscription in DD or Multicity Cheque should be made out in the name of ‘INDIAN SOCIETY OF
GEOMATICS’ and payable at Ahmedabad.
4. Financial year of the Society is from April 1 to March 31.
5. For further details, contact Secretary, Indian Society of Geomatics at the address given above.
6. ISG has chapters already established at the following places. Ahmedabad, Ajmer, Bhagalpur, Bhopal,
Chennai, Dehradun, Delhi, Hyderabad, Mangalore, Mumbai, Mysore, New Delhi, Pune, Srinagar,
Tiruchirappalli, Vadodara and Visakhapatnam. Applicants for membership have the option to contact
Secretary/Chairman of the local chapter for enrolment. Details can be found at the website of the Society:
www.isgindia.org
7. Journal of the Society will be sent only to Patron Members, Sustaining Members and Life Members.