The Indian Geographical Journal

Volume 95 (2) December - 2020, pp. 282-291

ISSN 0019-4824

AUTOMATED SURFACE WATER EXTRACTION USING

DIFFERENT SPECTRAL INDICES FROM THE SATELLITE
IMAGE FOR ERNAKULAM DISTRICT
Dhanusree M., Narmada K., Dipjyoti Gogoi and Bhaskaran G.
Centre for Water Resource Management, University of Madras, Chennai - 600 025
E-mail: [email protected]; [email protected]

Abstract
Water is an indispensable part of ecosystems and valuable, substantial reserves for social
survival and human development. Remote sensing has been successfully applied as it
facilitates to harness the natural resources effectively. The present study has been carried
out in the Ernakulam District of Kerala to detect waterbodies and determine the surface
area changes of waterbodies from 1991 to 2019. Remote sensing techniques have been
used as an essential tool for understanding spatial changes. The change in surface
waterbodies can be identified using these techniques. The Normalized Difference Water
Index (NDWI), Modified Normalized Difference Water Index (MNDWI), Automated Water
Extraction Index (AWEI), and Water Index 2015 (WI2015) were developed for the detection
of water features from Landsat imagery. These indices were used to identify waterbodies of
Ernakulam District. This study provides a scientific way to understand waterbodies using
remote sensing techniques.

Keywords: Automated surface water extraction, Remote sensing, Indices, Waterbodies

Introduction

Water stabilizes ecosystems and maintains climate variation, carbon cycling, and
so on. Hence, its accurate and automatic extraction is a challenge. Remote sensing
technology is widely used for detecting changes. It is used for different purposes, such as
land use/cover change, disaster monitoring, forest and vegetation change, urban sprawl,
and hydrology. Remote sensing data is generally applied to these studies. Observing
changes using remote sensing is extensively used in various applications such as land use/
land cover changes, disaster monitoring, forest and vegetation change, urban sprawl, and
hydrology. Trustworthy knowledge about surface water's spatial allocation is critically
significant in numerous scientific methods such as estimating present and future water
resources, wetland inventory, watershed study, surface water examination and
management, flood mapping, and environment monitoring (Desmet, 1996). Remote sensing
satellites at various spatial, spectral, and temporal resolutions give an enormous volume of
data that have become prime sources, mainly used to detect and extract surface water and
its changes in recent decades. The Normalized Difference Water Index (NDWI), Modified
282
Dhanusree M. et al. The Indian Geographical Journal, 95 (2) December - 2020

Normalized Difference Water Index (MNDWI), Automated Water Extraction Index (AWEI)
was generated for the extraction of water features from Landsat imagery.
Remote sensing satellites at different spatial, spectral, and temporal resolutions
give an immense quantity of data that have become primary sources, and it is mainly used
for identifying and extracting surface water and its changes in recent decades (Xu, 2006).
This study contributes a precise way to know about the waterbodies by using remote
sensing techniques. Hence it would be useful for the decision-making process and help
administrators and experts formulate suitable plans for sustainable water resource
management. The paper focuses on the current condition of the waterbodies of Ernakulum
district due to the accelerated expansion in urbanization and to evaluate the reduction of
waterbodies using landuse and land cover and water indices methods from 1992 to 2019.
Study Area

The study is in Ernakulam District, Kerala to identify waterbodies and to analyse the
waterbodies' surface area changes from 1991 to 2019. It holds an area of 2407 sq.km. It is
bordered in the west by the Arabian Sea, in the north by Thrissur and Alappuzha, Kottayam
districts in the south, Idukki District in the east. The district is well joined with a sound
system of roads and railways. One of the rapidly growing city Cochin is located in this
district. The accelerated urban extension around Cochin results in the invasion of water
resources. In this situation, the demand for research about the waterbodies and their
surface area variations is essential. The total area of Ernakulam District is 2407 sq.km. One
of the growing port city, cochin is the headquarters of the district. The district has a great
system of roads and railways. The longest flowing river Periyar is flowing through this area.

Fig. 1. Ernakulam District, Kerala

283
Dhanusree M. et al. The Indian Geographical Journal, 95 (2) December - 2020

Database and Methodology

The study was carried out using Landsat 4-5 TM for 1992(on 6 March) and Landsat
8 for 2019(on 23 March). These data are collected from the United States Geological
Survey (USGS). The satellite data downloaded from the USGS earth explorer is pre-
processed in Erdas Imagine. Supervised classification and Change detection techniques
were used to find out landuse change. Indices like NDWI, AWEI, MNDWI techniques were
used to extract waterbodies from satellite images.

Fig. 2. Methodology of the Study

Results and Discussion

Landuse and Land cover
Application of land use /land cover changes helps plan and manage an area, as up-
to-date knowledge is needed to understand the prevailing area of land and its land cover
and identify land-use changes from year to year. This understanding will help in
conservation, diverse practices, and developmental pressures. Land cover is the physical
element at the surface of the earth. Land covers include grass, asphalt, trees, bare ground,
water, etc. In 1992, the area was having a considerable amount of water resources. The
settlements are found around Cochin, which is a coastal belt. Aluva and Vypen have high
settlements in the year 1992. The total built-up area accounts for about 12,500 hectares.
284
Dhanusree M. et al. The Indian Geographical Journal, 95 (2) December - 2020

Nearly 150,000 hectares are under agricultural land, which includes plantation
crops and paddy fields. Forest areas cover around 72,000 hectares, which is located in the
eastern part. Barren lands occupy nearly 53,000 hectares. In 2019, waterbodies were
reduced. Many small waterbodies, which were present in 1992 were missing in the 2019
map. Built-up land covers the areas near the coastal belt, and barren lands have turned into
residential areas. Built-up areas cover nearly 55,000 hectares of land. Cochin international
airport, which is built in 1994, is located near a tank. Agriculture area has been lessened
due to human intrusions, and many farming areas are converted into built-up land. Nearly
125,000 hectares of land was covered by agriculture in 2019. Forest land was reduced from
1992 due to deforestation activity. Barren land has reduced from 1992 to 2019; the main
cause being the urban growth, where these areas were transformed for economic or
recreational purposes. Moreover, 8,000 to 10,000 hectares of barren land have been
reduced. Table 1 shows the increasing and decreasing trend of various land uses.

In 2019, waterbodies were reduced. Many small waterbodies, which were present
in 1992 were missing in the 2019 map. Built-up land covers the areas near the coastal belt,
and barren lands have turned into residential areas. Built-up areas cover nearly 55,000
hectares of land. Cochin international airport, which is built in 1994, is located near a tank.
Agriculture area has been lessened due to human intrusions, and many farming areas are
converted into built-up land. Nearly 125,000 hectares of land was covered by agriculture in
2019. Forest land was reduced from 1992 due to deforestation activity. Barren land has
reduced from 1992 to 2019; the main cause being the urban growth, where these areas
were transformed for economic or recreational purposes. Moreover, 8,000 to 10,000
hectares of barren land have been reduced. Table 1 shows the increasing and decreasing
trend of various land uses.
Comparison of landuse and Land cover (1992-2019) and Accuracy Assessment
For accuracy assessment, a total of 50 random sample points was created for each
classified image and estimating kappa statistic using accuracy assessment module
available in ERDAS IMAGINE. From the results, the overall classification accuracy as
obtained from the classified images was observed to be 96.90% and 92.01% with kappa
statistic 0.945 and 0.89 for the years 1992 and 2019 respectively. From the table, it shows
that many land use classes have changed. Some of the areas which were earlier under
waterbodies have been transformed into settlements. Almost 32,000 hectares of farming
land have been converted into settlements. The areas near Aluva and Alappuzha have
changed into urban centres. Approximately 450 hectares of forest areas have transformed
into settlements. Over 9500 hectares of barren lands have transformed into settlements.
Table 1. Land Use Land Cover (1992 to 2019)
Class Number Class Name 1992 (Area in Hectare) 2019 (Area in Hectare)
1 Waterbodies 23989 21551.2
2 Settlements 12457.7 54763.7
3 Agriculture 144126 125093
4 Forest 72134.8 59690
5 Barren Land 53162.9 45336.3

285
Dhanusree M. et al. The Indian Geographical Journal, 95 (2) December - 2020

Fig. 3. Landuse Land Cover of Ernakulam District (1992)

Fig. 4. Landuse Land Cover of Ernakulam District (2019)

Table 2. Error Matrix (1992 and 2019)

1992 2019
Classes Names Producers User Producers User
Accuracy Accuracy Accuracy Accuracy
Agriculture 93.33% 94% 75% 98%
Barren Land 95% 95% 88.24% 88.24%
Forest 96.31% 95.31% 88.33% 88.33%
Settlements 85.71% 88.89% 90% 81.12%

Waterbodies 94.29& 82.50% 97.73% 93.48%

Overall Classification 96.90% 92.01%
Accuracy
Kappa Statistics 0.945 0.89

286
Dhanusree M. et al. The Indian Geographical Journal, 95 (2) December - 2020

Table 3. Change Detection (1992 to 2019)

Class Name Waterbodies Settlements Agriculture Forests Barren Lands
Waterbodies 0 1603.53 1889.46 19.98 637.74
Agriculture 935.55 32827.5 0 3052.62 12260.1
Forest 119.97 455.76 8508.15 0 13169.7
Barren Lands 535.77 9519.57 177665.5 6684.93 0
Water Indices
The use of remote sensing for water body extraction began many years ago, and
then different types of water indices have been developed (Zhang et al. 2018). Some of the
important water indices are the normalized difference water index (NDWI) (McFeeters
1996) and the Modified Normalized Difference Water Index (MNDWI) (Xu, 2006). The
earlier one takes the green and the N.I.R. band for water body evaluation. Yet, NDWI could
not efficiently extract water as it mixed up built-up features with water area to be considered
and consequently exaggerated water (Xu 2006). The modified normalized difference water
index (MNDWI) was developed to succeed in this problem. It uses the green and the mid-
infrared band instead of the near-infrared band and is found to be more accurate. In recent
years, new water indices have been developed which use more than two bands to improve
spectral differentiation and water feature delineation. Automatic Water Extraction Index
(AWEI) developed by (Feyisa et al. 2014). Fisher et al. (2016) developed Water Index2015
(WI2015), which includes five bands.
Although newly formed, AWEI is widely used along with MNDWI for water body
extraction and mapping (Guo et al. 2017b). Other purposes are to facilitate the analysis of
the change in waterbodies and support research into the long-term trends of water extent
(Fisher et al. 2016). In this respect, these three indices are chosen for the present study to
verify the correctness of these indices under a specific landscape situation and water
surface area changes in Ernakulam District.
Normalized Difference Water Index (NDWI)
(NDWI) was first suggested by (S. K. McFeeters, 1996) to identify surface waters in
wetland environments and to allow for the measurement of surface water dimensions.
NDWI for T.M. and O.L.I. sensor is defined as the equation (Verpoorter, 2012)

Fig. 5. NDWI 1992 Fig. 6. NDWI 2019

287
Dhanusree M. et al. The Indian Geographical Journal, 95 (2) December - 2020

NDWI = (Green – NIR) / (Green + NIR)

Where, Band2=Green, band4=NIR for Landsat 4-5 TM.
Band3=Green, band5=NIR for Landsat 8 OLI/TIRS
Modified Normalized Difference Water Index (MNDWI)

McFeeters (1996) proposed the Normalized Difference Water Index (NDWI). The
Normalized Difference Water Index (NDWI) results in water features with positive values.
On the other hand, vegetation and soil typically come out at zero or negative values
(McFeeters 1996). The modified normalized difference water index (MNDWI) can be
expressed as follows:

MNDWI = ( −SWIR)/( +SWIR)

Where, Band2=Green, band5=SWIR for Landsat 4-5 TM

Band3=Green, band6=SWIR for Landsat 8 OLI/TIRS

Fig. 7. MNDWI (1992) Fig. 8. MNDWI (2019)

Automatic Water Extraction Index (AWEI)

The Automatic Water Extraction Index (AWEI) was developed by Feyisa et al.
(2014). The purposes for developing of AWEI were (a) to automatically suppress
classification noise from shadows and other non-water dark objects and to improve the
precision of surface water mapping, (b) to investigate the robustness of the new method
under diverse environmental conditions as well as to evaluate the relative accuracy in
comparison with existing classification procedures. The selection of spectral bands, their
arithmetic combinations, and the coefficients for the development of AWEI equations was
based on the rational inspection of various land features' reflectance characteristics. Based
on environmental and landscape conditions, Feyisa et al. (2014) have developed equations.

AWEI=4×( – 1)−(0.25× +2.75× 2)

288
Dhanusree M. et al. The Indian Geographical Journal, 95 (2) December - 2020

These equations were based on the spectral reflectance value form the Landsat 4-
5 TM imageries, where band1=Blue, band2=Green, band4=NIR, band5=SWIR1, and
band7=SWIR2. And from the Landsat 8 OLI/TIRS imageries, band2=Blue, band3=Green,
band5=NIR, band6=SWIR1, band7=SWIR2. These equations aim to put non-water pixels
below zero (0) and water pixels above zero (0).

Fig. 9. Automatic Water Extraction Index (a) 1992 (b) 2019

The four spectral water indices (NDWI, MNDWI, AWEI) are applied to the lake
water area to highlight the differences between water and non-water areas. NDWI water
areas are generally having values greater than zero. So this index is useful to extract the
surface water. MNDWI images are classified into water and non-water using a threshold of
0 (S. K. McFeeters, 1996). It separates water and non-water objects well, with water areas
generally having values higher than zero and vegetation areas having a substantial
negative value. The optimal threshold value for the AWEI is recommended by (G. L. Feyisa,
2014) varies from −0.15 to 0.045. In this study, the threshold value was accepted 0 due to
provide consistency between all applied indices. Landsat imagery is the most popular data
source for calculating water indices due to its suitable spectral bands and medium spatial
resolution. It is clear from maps that water bodies are easily discriminated from all of the
three water index images. It is also noted that MNDWI images have higher contrast in this
case. All three indices show the extraction and detection of water bodies. This study has
discussed the change in the landuse and landcover of Ernakulam District using the
geoinformatics techniques. Around 2400 hectares of water bodies have been lost in the last
27 years. Settlement areas have increased 5 times in the study area replacing the water
bodies, agricultural land, forest cover and the barren lands. Supervised classification
method was used to classify the image and accuracy assessment has shown overall
accuracy to be 96.9% and 92.1% for the years 1992 and 2019 respectively. Different
indices have been used to understand the most efficient method to extract waterbodies.
NDWI and MNDWI show better results than AWEI.

However, NDWI could not achieve likely accuracy in water feature evaluation with
built-up surroundings. The estimated water pixels in that area mixed with built-up noise
indicated that pixels from built-up features might have a positive value and are classified as

289
Dhanusree M. et al. The Indian Geographical Journal, 95 (2) December - 2020

water. Examining the spectral reflectance of water features, vegetation, and built-up areas,
authors found that the reflectance pattern of built-up and water features are similar in the
green and N.I.R. band in T.M. image. This is why the enhanced water presence in the
NDWI-image is often mixed with built-up land noise.in this condition use of MNDWI is used.
The MNDWI presented with three results: (a) water obtained more refined positive value
than NDWI due to greater consumption of SWIR (b) built-up areas displayed negative
values (c) soil and vegetation also produced negative values as reflects more SWIR light
than NIR. In AWEI a default threshold zero (0) was used at the first step; however, there
were possibilities that due to the variation among scene contrast and brightness, acquisition
time and space, the default threshold might not give a proper extraction of water.

Conclusion

This paper is focused on the temporal change of water bodies of Ernakulam District
Kerala, using remote sensing techniques and geographic information systems. Five
techniques are applied to detect the change of the surface water bodies by using ERDAS
Imagine and ArcGIS software. It has been clearly explained in the paper, that the water
resources, especially in the urban areas, are 'mismanaged' because of human interference
in the name of development. Extend of the city limit, an increase in the population and
many anthropogenic activities have been carried out near the water bodies, which has
worsened the water bodies condition more. The analysis results reveal that the water
bodies have shown very major changes in their surface area coverage over the past
decades. If such a decreasing trend in water bodies, especially lake, continues, the lake will
likely lose its entire water surface shortly. Further, a new approach was introduced for
surface water change detection based on NDWI, MNDWI, and AWEI. The study
demonstrated high performance of the proposed approach in simultaneously detecting the
surface water changes between two time periods. The result shows better change detection
using NDWI, MNDWI, and does not work well with AWEI. For smaller water bodies, mid
resolution Landsat image pixels were unable to detect the water, whereas it is vice versa for
the larger ones. In conclusion, the water extraction indices like NDWI, MNDWI has been
proven to be effective in detecting the water surface changes in Ernakulam. Accordingly,
the methods may prove useful in studying other surface waters in the world as well as flood
monitoring.

References

1. Amare Sisay (2016), "Remote Sensing Based Water Surface Extraction and Change
Detection in the Central Rift Valley Region of Ethiopia," American Journal of
Geographic Information System 2016, 5(2) pp 33-39, DOI: 10.5923/j.ajgis.20160502.01
2. Coppin, P.R. and Bauer, M.E. (1996), “Change Detection in Forest Ecosystems with
Remote Sensing Digital Imagery” Remote Sensing Reviews, pp 207-234.
3. Desmet, P.J.J., Govers, G(1996), "A GIS procedure for automatically calculating the
USLE LS factor on topographically complex landscape units," J. Soil Water Conserv.,
pp, 427–433

290
Dhanusree M. et al. The Indian Geographical Journal, 95 (2) December - 2020

4. Durmaz F, Karakaya N and EvrendilekF (2016), “Spatiotemporal Change Detection

Analysis of Turkish Lake Water Surface Area in Response to Anthropogenic Ecosystem
Disturbances Using LongTerm Landsat TM/ETM+ Data," Journal of Ecosystem
&Ecography, ISSN:2157-7625 JEE, Volume 6 • Issue 2, DOI: 10.4172/2157-7625.
5. Feyisa, G.L., Meilby, H., Fensholt, R., & Proud, S.R. (2014), "Automated Water
Extraction Index: A new technique for surface water mapping using Landsat imagery,"
Remote Sensing of Environment, pp 23-35
6. Fisher, A., Flood, N., & Danaher, T. (2016), "Comparing Landsat water index methods
for automated water classification in eastern Australia," Remote Sensing of
Environment, pp 167- 182
7. Gregoire, M., Stephane, D., Emmanuel, T. and Lionel B. (2010), "Change detection in
remote sensing observations," ISTE Ltd and John Wiley & Sons Inc, pp 95-142.
8. GulcanSarp, Mehmet Ozcelik (2016), "Waterbody extraction and change detection
using time series: A case study of Lake Burdur, Turkey," Journal of Taibah University
for Science 11 (2017) 381-391
9. Guo, Q., Pu, R., Li, J., & Cheng, J. (2017), "A weighted normalized difference water
index for water extraction using Landsat imagery," International Journal of Remote
Sensing, pp 5430- 5445
10. Lu, D., Mausel, P., Brondizio, E. and Moran, E. (2004) “Change detection techniques”,
Int J Remote Sens.
11. Lu, S., Wu, B., Yan, N., Wang, H(2011), “Waterbody mapping method with HJ-1A/B
satellite imagery”, Int. J. Appl. Earth Obs. Geoinf., pp 428–434.
12. McFeeters, S. (2013), “Using the normalized difference water index (NDWI) within a
geographic information system to detect swimming pools for mosquito abatement”, A
practical approach. Remote Sens., pp 3544–3561.
13. McFeeters S.K. (1996), “The use of the normalized difference water index (NDWI) in
the delineation of open water features”, International Journal of Remote Sensing, pp
1425–1432.
14. Rouse, J.J.W., Haas, R., Schell, J., & Deering, D. (1974), “Monitoring vegetation
systems in the Great Plains with ERTS”, In Proceedings of the Third Earth Resources
Technology Satellite Symposium. Greenbelt, Ontario, Canada, pp 309–317
15. Singh, A. (1989), “Digital change detection techniques using remotely-sensed data”
International Journal of Remote Sensing, pp 989-1003.
16. Verpoorter C., Kutser, T. and Tranvik L. (2012), “Automated mapping of water bodies
using Landsat multi-spectral data", Limnol. Oceanogr., pp 1037–1050
17. Xu, H. (2006), “Modification of normalised difference water index (NDWI) to enhance
open water features in remotely sensed imagery”, International Journal of Remote
Sensing, pp 3025-3033