CHANGE DETECTION ALGORITHMS IN

URBAN EXPANSION: A LITERATURE

STUDY

Submitted in partial fulfillment of the requirements

of the degree of
M. Tech. in Urban Systems

By

Jasim Qamar
22M0266

Supervisor
Prof. Arnab Jana
Prof. Eswar Rajasekaran

Center for Urban Science and Engineering (CUSE)

Indian Institute of Technology Bombay

April 2023
 Declaration

I declare that this written submission represents my ideas in my own words and where
other’s ideas or words have been included, I have adequately cited and referenced the original
sources. I also declare that I have adhered to all principles of academic honestly and integrity
and have not misrepresented or fabricated or falsified any ideas, data, facts or sources in my
submission. I understand that any violation of the above will be cause of disciplinary action
by the institute and evoke penal action from the sources which have thus not been properly
cited or from whom proper permission has not been taken when needed.

Date: 28.04.2023 Jasim Qamar

Place: Mumbai 22M0266

v
 Abstract

Urban expansion impacts local ecosystems and monitoring land use changes through
remote sensing and geospatial technology is crucial for effective management of natural
resources. However, the most effective change detection technique is still up for debate and
there is a lack of a comprehensive overview of urban domains for researchers to select
appropriate methodologies. This report focuses on the use of remote sensing-based data for
monitoring and analyzing urban land change, including the use of Landsat TM, ETM+ & OLI
and SPOT-5 sensors for detecting changes in land use, vegetation cover, surface temperature,
and moisture in urban areas. The report begins with an overview of the importance of
urbanization and the need for accurate and up-to-date information on urban land change. It
then discusses the advantages and limitations of remote sensing-based change detection, as
well as the different types of data sources and methods available for monitoring urban land
change. The report discusses different categories of change detection algorithms: algebra,
transformation, classification, and object-based. Each category has its advantages and
limitations, and the selection of an appropriate technique depends on the study area and the
type of change being detected. The report also provides a review of the current state of
knowledge in urban land change science, highlighting gaps in knowledge and suggesting five
strategies to reduce these gaps. The strategies include studying urban land change in a wider
geographic area, focusing on smaller and medium-sized cities, monitoring urban areas more
frequently, examining within-urban and volumetric change, and improving the
reproducibility, replicability, and comparability of results. The report also discusses the
challenges of creating a "best practices guide" for urban land change science due to the
complexity and diversity of methods available and the types of changes being measured.

Keywords: Urban expansion; Remote sensing; Change detection: Algorithm

vii
 Table of Contents

Contents

Declaration ...................................................................................................................... v

Abstract ........................................................................................................................ vii

Table of Contents ...........................................................................................................ix

Nomenclature .................................................................................................................xi

Chapter 1 Introduction .................................................................................................. 13

1.1 Aim ................................................................................................................. 15

1.2 Objectives ....................................................................................................... 15

Chapter 2 Urban expansion conceptualization using remote sensing ........................... 16

2.1 Unidirectional land transitions ........................................................................ 16

2.2 Comparing qualitative and categorical change ............................................... 17

2.3 Innovations altering remote sensing of urban expansion ................................ 18

2.3.1 Global scale products ................................................................................ 18

2.3.2 Online platforms ....................................................................................... 18

2.3.3 Modern data science tendencies ............................................................... 19

2.3.4 Indices ....................................................................................................... 19

Chapter 3 Methodology................................................................................................. 21

3.1 Selection of Research ...................................................................................... 21

3.2 Database search and screening........................................................................ 22

3.3 Metrics ............................................................................................................ 22

3.4 Geography ....................................................................................................... 23

Chapter 4 Data............................................................................................................... 24

4.1 Remote sensing imagery classification ........................................................... 24

ix
 4.1.1 Landsat TM, ETM+ & OLI sensor imagery ............................................ 24

4.1.2 SPOT-5 ..................................................................................................... 26

4.1.3 DMSP-OLS night-time image .................................................................. 27

4.2 Ancillary data ................................................................................................. 29

Chapter 5 Algorithm ..................................................................................................... 30

5.1 Algebra ........................................................................................................... 30

5.2 Transformation ............................................................................................... 32

5.3 Classification .................................................................................................. 34

5.4 Object based Image analysis .......................................................................... 36

Chapter 6 Case Study: Spectral indices-based object-oriented classification .............. 38

6.1 Study Area ...................................................................................................... 38

6.2 Objective of the study..................................................................................... 39

6.3 Results ............................................................................................................ 40

6.4 Summary ........................................................................................................ 41

Chapter 7 Analysis........................................................................................................ 42

7.1 Studies mostly track urban land transformation across short- to medium-term
timeframes 42

7.2 Algorithm development studies and algorithm application studies ............... 43

7.3 Accuracy assessment ...................................................................................... 43

Chapter 8 Discussion and Way Forward ...................................................................... 46

References .................................................................................................................... 49

x
 Nomenclature

RS Remote sensing

LULC Land use/land cover

GIS Geographic Information Systems

OBIA Object-Based Image Analysis

GUF Global Urban Footprint

GRUMP Global Rural-Urban Mapping Project

GHSL Global Human Settlements Layer

MODIS Moderate Resolution Imaging Spectroradiometer

GUIs Graphical User Interfaces

ENVI Environment for Visualizing Images

ERDAS Earth Resources Data Analysis System

NDVI Normalized Difference Vegetation Index

NDBI Normalized Difference Built-up Index

NDWI Normalized Difference Water Index

MSAVI2 Modified Soil Adjusted Vegetation Indices

MNDWI Modified Normalized Difference Water Index

TM Thematic Mapper

ETM+ Enhanced Thematic Mapper Plus

OLI Operational Land Imager

SPOT 5 Satellite pour l'Observation de la Terre

HRG High-Resolution Geometry

CNES Centre National D’Etudes Spatiales

DMSP-OLS Defence Meteorological Satellite Program- Operational Line Scanner

xi
DEMs Digital Elevation Models

CVA Change Vector Analysis

PCA Principal Component Analysis

KT Tasselled Cap

GS Gramm-Schmidt

EM Expectation-Maximization method

ANN Artificial Neural Networks

NCT National Capital Territory

WGS84 World Geodetic System 1984

PCM Percentage Correctly Matched

GLCM Gray Level Co-occurrence Matrix

AVHRR Advanced Very High-Resolution Radiometer

MBI Morphological Building Index

DRR Disaster Risk Reduction

URA Urban Risk Assessment

xii
 Chapter 1
Introduction

A history of urban expansion indicates that cities are among the planet's most dynamic
regions. Despite their importance for the local economy, urban expansion impacts the local
ecosystem in significant ways (Yuan et al., 2005a). As a result of the expansion of
urbanization, natural forms of land cover have been transformed into urban areas (Alqurashi
et al., 2016).

The world population has rapidly increased from 244 million in 1900 to 7 billion in
2011 and is expected to exceed 9.3 billion by 2050 (WUP2011_Report, n.d.). In the last few
decades, urban areas have grown worldwide. One of the most obvious causes of this growth is
the increase in the population. Therefore, environmental concerns associated with urban
growth must be analyzed and monitored carefully, in order to effectively manage land use.
Also, the influences exerted by urban growth may give rise to varying consequences in the
different dimensions (e.g., physical and socioeconomic) of cities (Y. Chen et al., 2017).

Having accurate and up-to-date information on land cover change is essential for
comprehending both human influences and the environmental effects of such changes, which
can then help with managing natural resources (Afifi & Darwish, 2018). An adequate amount
of data about the Earth's surface is required to analyze and monitor land cover changes.
Remote sensing (RS) is the most common method used for this purpose. Land use/land cover
(LULC) data from remote sensing are excellent sources of updated information, allowing fast,
accurate, and efficient analysis, modeling, and interpretation (Araya & Cabral, 2010).

LULC changes are best determined at different scales using geospatial technology, based on
satellite remote sensing data (RS) and Geographic Information Systems (GIS). Understanding
and predicting LULC changes depend on tracking and analyzing their spatial patterns (F.
Ahmad et al., 2017).

Since the 1970s, remote sensing data have been routinely used to detect changes in land cover
and use. According to the level of image processing used, change detection approaches can be
divided into three categories: pixel-level, feature-level, and object-level. For detecting
13
changes in multi-temporal imaging, pixel-level change detection analyses data pixel by pixel.
However, using a user-defined heterogeneity criterion, object-based image analysis (OBIA)
separates the image into meaningful segments (groups of pixels) known as object primitives.
Before classifying the images, segmentation is required. Additionally, a single land cover
class is allocated to each segment. These segments were produced based on spectral
characteristics as well as geographical and contextual information, such as shape, texture, etc.
The classifier can distinguish between spectrally similar but spatially dissimilar land cover
classes thanks to this spatial information. As a result, the user can recognize things made up
of pixels with a variety of characteristics. It allows the creation of classes and thematic maps
with strong connections to the real world (Bhatt et al., 2018).

Characterizing and tracking urban land change has undergone a radical transformation thanks
to remote sensing. Previously solely accessible through ground surveys and conventional
mapping methods, data and information about changes in urban land structure, shape, and
extent are now routinely gathered and produced using a constellation of government and
commercial airborne and satellite sensors. Remote sensing has revolutionized urban land
change measurement, but it is equally obvious that the most effective change detection
technique is still up for debate. Numerous techniques are in use now, and many more are
created every day. This multiplicity is required because the methodologies must be
compatible with the urban contexts under consideration and the kinds of urban land
transformations taking place. However, it might be challenging for user communities to sort
through the extensive research on urban land change detection methods (Reba & Seto, 2020).

There isn't a complete overview of urban domains that enables researchers to assess and select
the most appropriate methodologies for their topic, understand potential overlaps and
synergies, or apply findings from one discipline to another. In addition, there is no description
of the kinds of spatial data utilized for urban algorithm analyses or the procedures for
selecting parameters (Casali et al., 2022).

The demand for precise and timely information on land use changes, particularly in
metropolitan areas, is increasing as urbanisation proceeds. By comparing various data sets
(such as satellite photos) acquired at various dates to identify changes in the land cover and
land use, change detection algorithms can provide such information.

14
1.1 Aim

The study will focus on to evaluate the effectiveness of different change detection techniques
in detecting urban expansion.

1.2 Objectives

The following are the article's key objectives:

• To study about the different techniques such as supervised classification, unsupervised

classification, and object-based image analysis in detecting land use and land cover
changes that are associated with urban expansion.
• To provide a comprehensive review of the different algorithms available for detecting
specific types of urban land change and to provide guidance on which algorithm is
most appropriate for different types of change.
• To evaluate the strengths and weaknesses of each algorithm in terms of accuracy,
computational requirements, and data requirements.

Following the introduction, Chapter-2 presents concepts for understanding Urban expansion
through the lens of Remote sensing, and Chapter-3 specifies the methodological framework of
conducting the literature study. Subsequently, Chapter-4 presents different data sources
available in the literature for RM and Chapter-5 summarizes different remote sensing
algorithms that are used in urban change detection. Following that, Chapter-6 presents an
Indian case study that detects changes in highly urbanized areas using satellite imagery over
time and Chapter-7 presents an analysis of the literature. Finally, Chapter-8 discusses the
findings and presents a road map for choosing the best possible algorithm and also discusses the
way forward for my MTP (Master’s Thesis Project) in Urban risk assessment.

15
 Chapter 2
Urban expansion conceptualization using remote sensing

The algorithms we use to assess and track urban land change directly depend on how
we conceptualize urban land cover change. Here, we give a succinct summary of some of the
important conceptualizations in the literature as well as the several forms of urban land
change that have been quantified.

2.1 Unidirectional land transitions

Many urban land change detection studies assume a unidirectional change from non-
urban to urban land cover (i.e., from natural vegetation or forest to intensive agriculture and
urban areas), as this is a common pattern in many urban areas. However, this assumption is
not always valid, so to better understand urban land cover changes, it is important to use a
range of data sources and analytical techniques that can capture both unidirectional and multi-
directional changes in land cover over time. In recent years, Remote Sensing based Land
use/Land cover change detection efforts have also increasingly focused on unidirectional
transitions yet more so by taking any one type of LULC transition per study (Pandey et al.,
2018). Since numerous LULC shifts may take place throughout time, LULC transitions are
multidirectional. For instance, a forest region may be cleared for farming, then the land's
usage might alternate between farming and fallow land until the area is eventually put to use
for urban purposes. It is still feasible for vegetation to grow even after urban growth. These
changes in LULC may be due to variation within a particular land use class (such as urban or
agricultural), but they may also be the result of land use transitions, which can entail many
land use states without a clear terminal state. These structural changes, respectively, inside
and between LULC categories have seldom been conceptualized.

16
 Figure 1: (a) Panel A shows a common conceptualization of unidirectional land cover transitions with a
hypothetical land cover trajectory (left) (b) Panel B presents a newly proposed multidirectional conceptualization of land
cover transitions and hypothetical trajectory (left) (Reba & Seto, 2020).

Accurate links between land change science and economic and social impacts can be created
with an in-depth understanding of the subtle changes in urban land cover over time (Pandey et
al., 2018).

2.2 Comparing qualitative and categorical change

The kinds of changes that will be monitored can be used to group urban land change detection
techniques. Some changes to land cover are categorical in character, i.e., they represent
significant shifts between different land use or cover categories. Other modifications,
including adjustments to usage, density, or intensity, indicate more subtle alterations within
categories. Measuring the extent of change in these circumstances is crucial. It is crucial to
distinguish between assessing categorical change and the size of changes within land cover
classes since the most suitable techniques and chances of success differ. It is often simpler to
assess categorical changes, such as the transition from farmland to urbanism, than more subtle
longitudinal changes, such as the shift from low-intensity to high-intensity urban land.

17
2.3 Innovations altering remote sensing of urban expansion

2.3.1 Global scale products

Although methods for identifying and monitoring urban land change are the main subject of
this study, our capacity to map urban land change has evolved as a result of the expanding
global urban land product inventory for one time period. Urban land change study at scale is
possible because to initiatives like the Global Urban Footprint (GUF), Global Rural-Urban
Mapping Project (GRUMP), GlobalLand30, and the Global Human Settlements Layer
(GHSL), which were developed by many research organisations worldwide. We now have
several time slices of global urban products from GHSL and GUF that can greatly help urban
land change at regional and global scales, even if these goods are not accessible for regular
time intervals. These products have significantly improved in resolution and accuracy
compared to older global products, such as the MODIS 500 m resolution global urban
product, which may be used to study at-scale urban land change. These global products are
helpful for analysis at the global and regional levels, but less for tracking localized land use
changes. Products at the global level typically lack the precision required to quantify land
cover characteristics locally because they try to encompass a wide range of urban land change
scenarios.

2.3.2 Online platforms

Early remote sensing analyses were command-line driven. This amount of coding was
necessary to perform the analysis. Graphical user interfaces (GUIs), with their user-friendly
dropdown menus and minimal coding requirements, such as ENVI, IDRISI, and ERDAS
Imagine, arose as both the size of the user community and software development increased.
Online systems combine these two strategies in certain ways.

For instance, the increasingly popular Google Earth Engine platform has fundamentally
changed how we track and describe the urban land transformation in three crucial ways:
Access to a multi-petabyte collection of satellite images; the ability to analyze sizable
geospatial datasets utilizing cloud computing; and the integration of various forms of

18
remotely sensed imagery and geospatial datasets from public and private sources are all
provided by this company. Google Earth Engine may be used with only rudimentary Python
or JavaScript skills.

As a result, a wide spectrum of users, even individuals with little technical know-how or no
prior expertise with high-performance computing, can utilise the platform. Many of the
processing chains are not open access with inspectable and alterable code, and have been
criticised for their "black box" character, even if users are obliged to build and submit their
own scripts.

2.3.3 Modern data science tendencies

Finally, early analyses of urban remote sensing focused mainly on spectral, spatial, or
temporal changes separately, although today it is more typical to simultaneously evaluate
spectral, spatial, and temporal trends. An in-depth and more complex knowledge of urban
transformation trajectories for a specific location may be achieved by integrating all of these
methodologies collectively. For instance, detailed time series analyses can reveal patterns of
nonlinear development and redevelopment over time, while a deeper understanding of urban
spatial form can reveal relationships between various intra-urban land uses, such as the
coexistence of residential and commercial areas within a city.

2.3.4 Indices

There are various indices that can be used to measure urban land change, such as the
Normalized Difference Vegetation Index (NDVI), the Normalized Difference Built-up Index
(NDBI), and the Normalized Difference Water Index (NDWI). These indices are based on the
spectral characteristics of different land cover types and can be used to differentiate between
urban areas and other land cover types.

NDVI, for example, is a measure of vegetation cover and can be used to detect
changes in urban green spaces. NDBI, on the other hand, measures the presence of built-up
areas, such as roads, buildings, and other infrastructure. NDWI measures the presence of

19
water bodies, which can also be used to detect changes in urban land cover. Using these
indices, urban land change can be monitored and quantified over time.

This information can be used to develop strategies for sustainable urban development,
such as identifying areas that require conservation or redevelopment. Overall, remote sensing
using indices provides a powerful tool for measuring urban land change and monitoring the
impacts of urbanization on the environment.

Several research (Bhatt et al., 2018; X. L. Chen et al., 2006) has examined the surface
temperature obtained from the thermal infrared band after extracting land use/cover
information from remote sensing pictures of various time periods using these various urban
indices (NDVI, NDWI, MSAVI2, MNDWI, and NDBI)

This Chapter discusses the importance of conceptualizing urban land cover change and
using a range of data sources and techniques to capture both unidirectional and
multidirectional changes in land cover over time. It also emphasizes the need to distinguish
between assessing categorical change and the size of changes within land cover classes, as
well as the innovations altering remote sensing of urban expansion, including global scale
products, online platforms, and modern data science tendencies.

20
 Chapter 3
Methodology

This section describes the process and methods that have been followed in this
literature study. Methodologically, the study process falls into the conventional three steps of
a literature review process: i) planning the review by developing eligibility criteria; (ii)
identifying relevant literature through a database search, screening and selection; (iii)
conducting the review and analyzing it.

3.1 Selection of Research

Change detection techniques and urban expansion are broad fields. Following are the
criteria used to select papers that are relevant for our analysis:

• Papers mainly used change detection algorithms to detect urban land use/land cover
change. The study included supervised, unsupervised algorithm and object-based
image analysis methods.
• Papers with a focus on urban scales ranging from counties to neighbourhoods.

• Papers used geospatial datasets, i.e., data series, vector, or raster datasets when they
are analyzed in conjunction with a geographical location stored by coordinates or by
indexes. For example, the study included papers that used satellite images in
combination with feature datasets as references for land use and land cover.
• The study didn't include publications concerned with remote sensing, the identification
of geographical features and objects in remote sensing photos, image processing,
image classification, computer vision, or photographs of city streets.
• Papers are published in journals or peer-reviewed conferences and written in English.

21
3.2 Database search and screening

To identify an initial pool of literature for this study, ScienceDirect was used to ensure
the highest academic standards and validity of the articles, and for its broad and multi-
disciplinary coverage. ScienceDirect is Elsevier’s leading platform of peer-reviewed scholarly
literature. Elsevier's exacting publishing guidelines ensure high-quality papers. Eminent
editorial boards oversee journals, and publications are extensively peer-reviewed.

The phrases "urban areas”, “cities" and "urban environment" are frequently used
interchangeably in literature. Therefore, the study included each term in the search. Moreover,
it also included the keywords ‘Urban expansion’, ‘land use change’ to aim for papers with a
spatial analytical component. As a result, the study used as keywords "urban remote sensing",
"Land expansion algorithm,", "Urban change detection algorithm" for the database search.
(Search string: "urban" AND "remote sensing" AND "change detection" AND "algorithm")

The research papers selected were highly cited according to ScienceDirect statistics to
ensure inclusion of publications with high impact. Also, the study focuses research papers
from 2000 to 2022 to ensure that the most recent trends and developments are covered. I
collected this set of papers on March 2023.

3.3 Metrics

Following the selection of the research papers based on the qualifying requirements, a review
of the existing literature is conducted. The key information selected in the papers are: title,
authors, year of publication, objective of the study, location of the study, methodology, data
reported, imagery type, algorithm, and accuracy assessment. The study examined each
publication to gather the data, which was then organized in a tabular format for mapping of
themes and methodologies. If a paper did not provide any information about a specific detail,
it is reported as missing.

22
Title Author DOI Year Location(Geography) Objective Imagery type Algorithm Accuracy
To analyze land-use
changes of Dalian City in
Analysis of the expansion of the built-up Chang Li-ping https://doi.org/10.100 Liaoning Province of China multi-temporal TM images in
2002 Dalian City, China
area of Dalian city Zhang Shu-wen 7/s11769-002-0045-5 over ten years by means of 1986, 1996 and 2000
the correlation analysis
method.

To derive land cover and

7 States of United States
Change detection with heterogeneous Kathleen M. Bergena land-cover change
(Illinois, Indiana, Iowa,
data using ecoregional stratification, Daniel G. Browna https://doi.org/10.101 information for forest,
2005 Michigan,
statistical summaries and a land James F. Rutherforda 6/j.rse.2005.03.016 urban, and agriculture
Minnesota, Missouri and
allocation algorithm Eric J. Gustafsonb categories over a seven-
Wisconsin)
state region in the U.S.
to examine the changes in
land use/ cover pattern in
a rapidly changing area of
the PRD region in relation
Remote sensing image-based analysis of Xiao-Ling Chen
Pearl River to urbanization since the
the relationship between Hong-Mei Zhao https://doi.org/10.101
2006 Delta (PRD) in Guangdong 1990s and then to Landsat TM and ETM+ images
urban heat island and land use/cover Ping-Xiang Li 6/j.rse.2005.11.016
Province, southern China investigate the impact of
changes Zhi-Yong Yin
such changes on the
intensity and spatial
pattern of the UHI effect
in the region.
a new urban change
Object-oriented Urban Dynamic AN Kai
https://doi.org/10.100 Haidian District of Beijing, detection approach, urban Landsat Thematic Mapper (TM)
Monitoring —A Case Study of Haidian ZHANG Jinshui 2007 object-oriented approach 84.83%
7/s11769-007-0236-1 China dynamic monitoring based Satellite Imagery
District of Beijing XIAO Yu1
on objects, is introduced

to detect land-use changes

PCA‐based land‐use change detection and J. S. Deng in an urban environment hybrid classifier
https://doi.org/10.108
analysis K. Wang based on Landsat-7 ETM+ & SPOT-5 (XS) High- combining unsupervised
0/0143116080195016 2008 Hangzhou City, China 89.54%
using multitemporal and multisensor Y. H. Deng principal-component resolution geometry (HRG) and supervised
2
satellite data G. J. Qi analysis (PCA) and hybrid classification
classification methods.
DENG Jin-Song to detect and analyze the
https://doi.org/10.101
Urban Land Use Change Detection Using WANG Ke dynamics of land use Landsat-7 ETM+ & SPOT-5 (XS) High- supervised
6/S1002- 2009 Hangzhou, China 90.67%
Multisensor Satellite Images LI Jun changes in the city proper resolution geometry (HRG) classification
0160(08)60088-0
DENG Yan-Hua of Hangzhou.
National land cover map, Landsat
TM, LISS-III (summer) and SPOT object-oriented approach,
Analysis and Modeling of Urban Land urban land use change (spring) satellite Shannon‘s Entropy & 92.51%(2000)
Yikalo H. Araya 1 https://doi.org/10.339 Setúbal and Sesimbra,
Cover Change in Setúbal and Sesimbra, 2010 analysis and modeling was images. Also digital elevation Markov chain (MC) and and
Pedro Cabral 0/rs2061549 Portugal
Portugal carried out model (DEM), major road Cellular Automata (CA) 87.68%(2006)
networks, and high spatial modelling methods
resolution aerial photographs

Tab 1: Sample showing how the metrics are gathered

3.4 Geography

All of the cities stated in the original study were included in the analysis to assess the
geographic scope of the research. For each individual case study, the country is counted even
if a study covers many study cities in a single nation. Similar to this, if research takes cities
from several nations into account, each nation is considered separately. Because many studies
incorporate numerous places, there are more geographic locations in the assessment than there
are individual publications.

This chapter explains the methodology used in the literature review, including the
criteria for selecting relevant research papers, the database search and screening process,
metrics used to analyze the papers, and the geographic scope of the study.

23
 Chapter 4
Data

4.1 Remote sensing imagery classification

4.1.1 Landsat TM, ETM+ & OLI sensor imagery

Landsat-5 TM (Thematic Mapper) and Landsat-7 ETM+ (Enhanced Thematic Mapper

Plus) sensors are commonly used for change detection studies due to their ability to capture
multispectral data at a moderate spatial resolution of 30 meters (USGS Website). Almost all
studies (Afifi & Darwish, 2018; Agapiou et al., 2015; F. Ahmad et al., 2017; S. Ahmad et al.,
2016; Alqurashi et al., 2016; DENG et al., 2009) use Landsat TM (Thematic Mapper) and
ETM+ (Enhanced Thematic Mapper Plus) bands for change detection in urban areas. Both
sensors have seven spectral bands, and each band has a specific wavelength range that
corresponds to different surface features and materials.

For urban change detection, the visible and near-infrared bands (bands 1-4) are
particularly useful for detecting changes in vegetation cover, while the mid-infrared and
thermal bands (bands 5-7) can help to identify changes in land use, surface temperature, and
moisture.

Here is a summary of how the Landsat TM and ETM+ bands can be used for urban
change detection:

• Band 1 (0.45-0.52 μm): This band can be used to distinguish between built-up areas,
vegetation, and water bodies.
• Band 2 (0.52-0.60 μm): This band is useful for detecting vegetation cover and changes
in vegetation density.
• Band 3 (0.63-0.69 μm): This band can help to identify urban features such as roads
and buildings, as well as changes in water quality.

24
 • Band 4 (0.76-0.90 μm): This band is particularly useful for detecting changes in
vegetation cover and health.
• Band 5 (1.55-1.75 μm): This band can help to identify changes in soil moisture,
vegetation, and land use.
• Band 6 (10.4-12.5 μm): This band is sensitive to changes in surface temperature,
which can be useful for detecting urban heat islands and changes in land use.
• Band 7 (2.08-2.35 μm): This band can be used to detect changes in vegetation health
and moisture content.

By analyzing the changes in these spectral bands over time, it is possible to detect and
monitor urban changes such as land use conversion, new construction, and changes in
vegetation cover.

Landsat-8 OLI provides high-quality data in multiple spectral bands, with a spatial
resolution of 30 meters for most bands, and a revisit time of 16 days. This data can be used to
identify changes in land cover, such as urban expansion, deforestation, and agricultural
practices. The Landsat-8 OLI sensor provides high-resolution multispectral data with
improved radiometric calibration, spectral resolution, and signal-to-noise ratio compared to
previous Landsat sensors.

Landsat-8 Operational Land Imager (OLI) imagery has been used for change detection
analysis since its launch in 2013 (USGS website). As a result, most studies (F. Ahmad et al.,
2017; Alqurashi et al., 2016; Bhatt et al., 2018; Cao et al., 2019; L. Chen et al., 2018) that
have been conducted since 2013 have utilized Landsat 8 OLI imagery for change detection,
especially for applications requiring a higher spatial resolution. It has been particularly useful
for monitoring changes in urban areas, where rapid expansion and development can have
significant impacts on the environment and the economy. Overall, the Landsat 8 OLI sensor
has provided an important tool for change detection studies, and it continues to be a valuable
resource for researchers, analysts, and policymakers.

However, it's important to note that earlier Landsat sensors, such as Landsat 5 and 7,
are still in use, and their data are still being used for change detection studies, particularly in
cases where longer time-series data are required. Also, the availability and quality of Landsat
data depend on various factors such as cloud cover, seasonal variations, and other factors that

25
may limit the use of Landsat 8 OLI imagery in some cases. Nonetheless, the Landsat 8 OLI
sensor has become a valuable resource for change detection studies and has significantly
improved the accuracy and reliability of these studies.

4.1.2 SPOT-5

SPOT 5 (Satellite pour l'Observation de la Terre) was a commercial Earth-imaging

satellite from CNES (Centre National D’Etudes Spatiales), the French Space Agency. It was
launched on 4 May 2002 and ceased operations after a malfunction in March 2015 (European
space agency website). SPOT-5 (XS) High-Resolution Geometry (HRG) has two HRG
instruments, each with a swath width of 60 km and a resolution of 2.5 meters. This makes it
an excellent tool for high-resolution remote sensing applications, such as change detection.
(Deng et al., 2008) made an attempt to use multisensor data with similar spatial resolution
from different image modes (SPOT-5 XS and Pan) as the primary source for multitemporal
change detection.

The high spatial resolution of SPOT-5 (XS) HRG allows for the detection of small
changes that may not be visible with lower-resolution sensors. Additionally, the stereo
capability of the sensor can be used to create digital elevation models, which can aid in the
identification of changes. Overall, SPOT-5 (XS) HRG is a valuable tool for change detection
and other remote sensing applications that require high spatial resolution and accurate data.

26
 Figure 2: Color composition image of Landsat-7 enhanced thematic mapper (ETM) in 2001 (a); color composition
image of SPOT (systeme pour l’observation de la terre or earth observation satellite)-5 in 2003 (DENG et al., 2009)

4.1.3 DMSP-OLS night-time image

DMSP-OLS (Defense Meteorological Satellite Program - Operational Line Scanner)

night-time imagery can be used for change detection analysis, particularly for monitoring
changes in urban areas, as well as natural disasters such as wildfires and floods. DMSP-OLS
night-time imagery is often used for change detection analysis because it provides a unique
view of the Earth's surface that is not available with other types of imagery. Specifically, the
night time imagery can reveal changes in urban areas and human settlements, which are often
illuminated by artificial lights at night. This makes it particularly useful for monitoring
urbanization and changes in land use patterns. It's worth noting that DMSP-OLS night-time
imagery has some limitations, such as low spatial resolution and low sensitivity to small
changes. Therefore, it should be used in conjunction with other data sources and techniques to
ensure accurate change detection analysis. The DMSP-OLS has been photographing
nightlights since the early 1970s, and from 1992 to 2008, yearly worldwide composites of
temporally stable night time lights were created using more modern digital data (Small et al.,
2011).

(Agapiou et al., 2015) used a radiance calibrated Defence Meteorological Satellite

Program Operational Line Scanner (DMSP-OLS) night-time data set for 2010. A direct

27
comparison of this dataset with urban areas of 2010, as classified from the Landsat sensor,
was applied to evaluate the potential use of night-time images.

Figure 3: Sample of DMSP-OLS stable light product (2013) for the Yangtze River Delta, China. The red circles
denote eight sample prefectural-level cities (Zheng et al., 2020).

DMSP-OLS night-time imagery can also be useful for monitoring natural disasters
such as wildfires and floods, as these events often produce changes in the patterns of night-
time lights. By comparing images taken before and after the event, analysts can identify areas
of change and assess the impact of the disaster. Additionally, DMSP-OLS night-time imagery
has the advantage of being collected continuously over time, which allows for long-term
monitoring and analysis of changes. This is particularly important for studying slow changes
in land use patterns, such as urbanization and deforestation. Overall, the use of DMSP-OLS
night-time imagery for change detection analysis can provide valuable insights into the
dynamic nature of the Earth's surface and help inform policy and management decisions
related to urbanization, natural resource management, and disaster response (Huang et al.,
2014).

28
4.2 Ancillary data

Ancillary data in change detection refers to any additional data or information that can
help improve the accuracy and reliability of detecting changes between two or more datasets.
Ancillary data can provide important contextual information that can help analysts better
understand the changes they are detecting. For example, if we are comparing two satellite
images to detect changes in land use, ancillary data such as maps of urban growth or census
data on population density could provide valuable additional information that can help
distinguish between natural changes and changes caused by human activity.

Another example is the use of topographic maps or digital elevation models (DEMs)
in change detection for identifying changes in water bodies such as lakes or rivers. By
comparing a recent image with a topographic map or DEM, analysts can identify whether a
change in the water level is due to natural factors such as rainfall or snowmelt or due to
human intervention such as damming or diversion of water. In summary, ancillary data in
change detection provides additional context and information that can help improve the
accuracy and reliability of detecting changes between datasets.

This chapter discusses the use of remote sensing imagery classification for urban
change detection, specifically with Landsat TM, ETM+, OLI, and SPOT-5 sensors. It
describes the spectral bands of these sensors that are useful for detecting changes in urban
areas, and highlights the importance of Landsat 8 OLI in providing high-resolution data for
change detection studies. Finally, it mentions the advantages of SPOT-5's high spatial
resolution for detecting small changes in urban areas.

29
 Chapter 5
Algorithm

The choice of an appropriate method or algorithm for a specific research endeavour is

crucial yet challenging since digital change detection is impacted by geographical, spectral,
thematic, and temporal restrictions, and because many change detection strategies are usable.
The change detection techniques are categorised into four categories for simplicity of
understanding:

5.1 Algebra

The algebra category comprises background removal, change vector analysis (CVA),
vegetation index differencing, image rationing, image differencing, and image regression.
These techniques locate changes but do not provide information on the nature of change
(Yuan et al., 2005b). These methods all use thresholds to identify the areas that have changed,
which is a common feature. These techniques, with the exception of CVA, are quite easy to
use, understand, and put into practice, but they are unable to offer comprehensive change
information matrices. The difficulty in choosing appropriate thresholds to detect the modified
areas is a drawback of the algebra category. In this category, choosing appropriate picture
bands or vegetation indices and choosing appropriate thresholds to identify the altered regions
are crucial for the change detection outcomes.

(Deng et al., 2008) utilised band selected from each new image to compare with each
other using change-detection approaches such as image differencing in extracting and
quantifying classes to detect and monitor land-use change in Hangzhou City, one of the most
developed regions in China. A more comprehensive CVA method that incorporates texture
information into the standard spectral-based CVA was utilised in order to accurately identify
the LULC alterations in rural–urban fringe area (He et al., 2011). Change Vector Analysis

30
approach (An et al., 2007) was utilised to compute Euclidean distance based on the shared
object with the CVA in order to extract the urban change information.

Tab 2: Summary of Algebra change detection techniques. (The five levels indicate the complexity of the change
detection techniques, from simplest 1 to the most complex 5.) (Lu et al., 2004)

The most often applied change detection technique in the algebra-based change
detection category is image differencing(Lu et al., 2004). It has been demonstrated that visible
red band image differencing is effective for change detection in semi-arid and dry settings, but
it is unclear if this is true in other environments, such as moist tropical areas. Since results
rely on the features of the study regions and the image data utilised, various writers have
come to different conclusions regarding which methodology, among the vegetation index
differencing, image ratioing, image regression, and CVA methods, offered the best results.

31
Due to its low capacity to identify changes, the background subtraction approach was rarely
utilised.

5.2 Transformation

The PCA (Principal component analysis), KT (Tasselled cap), Gramm-Schmidt (GS),

and Chi-square transformations are within the transformation category. These approaches
have the benefit of eliminating data redundancy between bands and emphasising various
information in derived components. They are unable to offer comprehensive change matrices,
and instead rely on the user to choose thresholds to recognise regions that have changed.
Another drawback is how challenging it is to read and label the change information on the
altered images (Tewkesbury et al., 2015).

(Agapiou et al., 2015) applied Principal component analysis (PCA) to detect any
significant changes under the period examined (1984–2010) to comprehend the pattern of
urban development in the future, paying particular attention to locations near historic
monuments.

(Deng et al., 2008)For extracting and quantifying classes to detect and track land-use
change in Hangzhou City, one of the most developed locations in China, a new technique
combines PCA with hybrid categorization. The main goal of PCA was to keep as much
variance in the data set as feasible while reducing the dimensionality of a data set made up of
several connected variables.

PCA and KT are the techniques most frequently employed to identify information that
has changed or remained the same in the transformation category. Many change detection
applications seem to benefit from the KT approach. The fact that KT transform coefficients
are independent of the image scenes, in contrast to PCA, gives it an edge over the latter. Due
to their greater complexity as compared to PCA and KT approach, the GS and Chi-square
techniques are utilised considerably less frequently in actual practise. Additionally, the
majority of commercial remote sensing image processing software does not support GS and
Chi-square approaches.

32
 Tab 3: Summary of Transformation techniques for Change Detection. (The five levels indicate the complexity of
the change detection techniques, from simplest 1 to the most complex 5.) (Lu et al., 2004)

33
5.3 Classification

Post-classification comparison, spectral-temporal combined analysis, expectation-

maximization method (EM) change detection, unsupervised change detection, hybrid change
detection, and ANN are all included in the classification category. These techniques are based
on categorized pictures, where high training sample data quality and quantity are essential to
achieve accurate classification results. The capacity to provide a matrix of change information
and lessen the external influence of atmospheric and environmental variables across the multi-
temporal pictures are these technologies' main advantages. However, choosing suitably large
and high-quality training sample sets for image classification can be challenging, especially
for the classification of historical picture data.

The capacity to provide a matrix of change information and lessen the external influence of
atmospheric and environmental variables across multi-temporal pictures are these
technologies' main advantages. However, choosing suitably large and high-quality training
sample sets for image classification can be challenging, especially for the classification of
historical picture data. Unsatisfactory change detection findings are frequently the
consequence of the time-consuming and challenging process of developing highly accurate
classifications, particularly when high-quality training sample data are not available.

Post-classification, which necessitates comparison of separately produced classed pictures, is

by far the most obvious method of change detection. Data from two dates are individually
categorised using the post-classification comparison approach, which reduces the issue of
atmospheric correction normalisation and sensor variances across various dates (Yuan et al.,
2005b). In order to create difference maps from which ''from-to'' change information may be
derived, post-classification comparison approaches were employed to put forth a strategy for
identifying and evaluating the danger of urban development and its impact on agricultural
land in the El-Beheira governorate (Afifi & Darwish, 2018).

While performing buffer assessments along the metro lines and stations, (S. Ahmad et al.,
2016) uses maximum likelihood supervised classification to fill up the research gaps by
describing if and how an urban rail infrastructure, along with the complementing policies that
follow, results in a change in land cover. (Afifi & Darwish, 2018) Cross tabulation was used

34
to build the two algorithms of supervised maximum likelihood and post-classification change
detection for monitoring urban sprawl and change detection.

In order to compare the maximum-likelihood classification approach with other classification

methods, (S. S. Chen et al., 2015) investigated both supervised and unsupervised image
classification techniques, including k-means, Mahalanobis distance, and maximum likelihood.
The maximum-likelihood classification test result was not only reliable but also rather
straightforward to understand and put into practise.

A layer arithmetic operation is frequently used in hybrid change detection to identify changed
elements, then supervised or unsupervised direct classification is used to provide meaning to
the altered features. For example, (Seto et al., 2002)after creating a CVA to show the
radiometric change's magnitude and direction, individual land cover transitions were labelled
using supervised classification. While (Doxani et al., 2012)used a knowledge-based
classification to filter and categorise the data after first applying a MAD transform to
highlight altered regions in VHR imaging. These examples demonstrate a research trend that
aims to employ many stages of change comparison to address specific issues. This tendency is
expected to persist as workflows get more intricate.

In order to classify image data, either supervised or unsupervised, these classification

algorithms often need a lot of training sample data. Improved classification outcomes are
achieved by image modification, vegetation indices, sophisticated classification techniques,
modelling, and the integration of various data sources. The difficulty of categorising historical
image data consistently has a significant impact on the outcomes of change detection, despite
the fact that post-classification comparison is a regular strategy employed in practise. The
benefits of threshold and classification approaches are combined in the hybrid change
detection method. Detecting altered regions involves threshold techniques like image
differencing, followed by classification and analysis of change areas. Due to the challenge of
recognising and labelling change trajectories, the spectral-temporal mixed change detection
approach and unsupervised change detection method are less frequently used.

35
5.4 Object based Image analysis

Object-based image analysis (OBIA) is a technique that involves grouping individual

pixels into meaningful image objects based on their spectral and spatial characteristics. OBIA
can be useful for change detection, as it allows for a more detailed analysis of the changes that
have occurred between two images. To use OBIA for change detection, the first step is to

Figure 4: General Methodology adopted for change

detection using OBIA (Bhatt et al., 2018).

acquire two or more images of the same area taken at different times. These images should be
pre-processed to correct for any differences in radiometric and atmospheric conditions. Then,
using OBIA software, the images are segmented into objects based on their spectral and
spatial characteristics. Once the images have been segmented, the objects can be compared to
identify changes between the two images. For example, the area covered by each object can
be compared, and any significant changes in object size can be identified as potential areas of
change. Similarly, changes in the spectral properties of objects can be used to identify
changes in land cover or land use (Bhatt et al., 2018).

36
 (Bhatt et al., 2018) compared unsupervised classification technique such as K-means
with OBIA for change detection in highly urbanized areas and finds that OBIA's classification
accuracy ranges from 82.2 to 90.1%, which is high enough to qualify it as the most popular
developing approach. Although k-means provides the maximum accuracy attainable is just
78%.

(Alqurashi et al., 2016) In this work, the categorization of LULC maps was carried out
using the OBIA approach. The total accuracies in all five cities ranged from 82% to 96%,
indicating extremely reliable findings to estimate LULC changes. In the study, both OBIA
and conventional pixel-based methods worked effectively. This could be because of the
medium spatial resolution, like that of Landsat, which offered about identical scales for both
pixels and objects.

This chapter discusses different categories of change detection techniques: algebra,

transformation, classification, and object-based. Each category has its advantages and
limitations, and the selection of an appropriate technique depends on the study area and the
type of change being detected.

37
 Chapter 6
Case Study: Spectral indices-based object-oriented
classification

The detection of changes in land use and cover, vegetation, and flood levels can be
done using remote sensing data. This study aims to develop a way to detect changes in highly
urbanized areas using satellite imagery over time. The study compares the effectiveness of
object-oriented classification to unsupervised algorithms like k-means for change detection.
The study area chosen is the National Capital Territory of Delhi, which represents urban
conditions in Asia. The study uses Landsat 5 and 8 imageries from 1993-2014 and three
spectral indices to characterize built-up areas, open water, and green vegetation. Object-based
classification is used to segment and classify the imagery, which reduces complexity and
produces highly accurate results. The study found that the object-oriented approach achieved
an accuracy of up to 90.1%, which is better than the unsupervised clustering algorithm k-
means (Bhatt et al., 2018).

6.1 Study Area

The study focuses on the National Capital Territory (NCT) of Delhi, located in
northern India. The total area of NCT of Delhi is 1483 sq.km., and it shares borders with Uttar
Pradesh and Haryana. The maximum length of NCT is 51.90 km with an extreme width of
48.48 km. Over the past six decades, NCT of Delhi has experienced a significant population
growth, increasing from 1.74 million in 1951 to 16.7 million in 2011. The study uses Landsat
5 TM and Landsat 8 OLI data from May 1993 to May 2014, with a part/row of 146/40 and
datum WGS84. The study selected this time of year because it is summer, and the foliage of
green vegetation is minimal, and water bodies have the smallest spread, which helps to
highlight built-up areas in comparison to other urban land use classes (Bhatt et al., 2018).

38
 Figure 5: Study area, i.e., NCT of Delhi, India (Bhatt et al., 2018)

6.2 Objective of the study

The objective of the research was to create a technique for detecting changes in highly
urbanized regions by utilizing satellite imagery over time. The study investigates the
efficiency of object-oriented classification as compared to unsupervised algorithms, such as k-
means, for detecting changes. To capture a broad range of pattern variations, the research
obtained a sequence of satellite images taken by Landsat 5 (TM) and Landsat 8 (OLI)
between 1993 and 2014. The study utilizes three spectral indices, specifically the Normalized
Difference Built-up Index (NDBI) to characterize built-up areas, the Modified Normalized
Difference Water Index (MNDWI) to identify open water bodies, and the Modified Soil
Adjusted Vegetation Index 2 (MSAVI2) to symbolize green vegetation (Bhatt et al., 2018).

Figure 6: Indices image generated for the Landsat imagery of study area for the year 1993. (a) Raw image, (b)
MNDWI, (c) MSAVI2 and (d) NDBI (Bhatt et al., 2018)

39
6.3 Results

Employing spectral indices significantly decreases data correlation and redundancy

between original multispectral bands, thus reducing spectral confusion between the different
land-use classes. During the object-based image analysis (OBIA), not only were the spectral
indices MNDWI, NDBI, and MSAVI2 created from the multispectral image, but also spatial
and contextual features were generated. These spatial, spectral, and contextual features were
then analyzed and utilized during the classification process. Figure below displays the
classified image created through object-based theory, as well as the image produced using an
unsupervised classification method such as k-means (Bhatt et al., 2018).

Figure 7: Classified images for the Landsat imagery of study area. 1993 (a, e), 2000 (b, f), 2008 (c, g), 2014 (d, h)
(Bhatt et al., 2018).

Additionally, the study compared the results of the k-means classification with those
obtained through OBIA. The comparison was based on the Percentage Correctly Matched
(PCM), which measures the percentage at which the k-means results match with the OBIA
results, with the latter being taken as the reference.

40
 Tab 4: Percentage wise distribution of classes and PCM evaluation for k-means(Bhatt et al., 2018)

The table shows that during the study period, there was a significant increase in built-
up area from 31% to 55.69%, while vegetation areas decreased from 59.5% to 38.01%. There
was little change in the water body area and a slight decrease in barren land. The rise in built-
up area can be attributed to the growth of industrial areas, educational institutions, IT hubs,
and corresponding infrastructure development in the NCT of Delhi. The OBIA method
achieved an accuracy of classification ranging from 82.2% to 90.1%, which is considered
good, while k-means only achieved a maximum accuracy of 78%.

6.4 Summary

In summary, the study used time-series satellite imagery and object-based image
analysis (OBIA) to detect changes in the urbanization of NCT Delhi from 1993 to 2014.
Spectral indices such as NDBI, MNDWI, and MSAVI2 were utilized along with texture-
based features like GLCM to reduce data dimensionality and decrease the dependency on
spectral characteristics. The results showed a clear increase in built-up area and a decrease in
vegetation over time, likely due to urban development. The OBIA approach produced a
significantly higher overall accuracy (90.1%) compared to the unsupervised technique of k-
means, and was recommended as a promising semi-automated tool for classification of
remotely sensed satellite data in urban areas. The use of OBIA with traditional spectral
indices provides dimensionality reduction and produces classified maps that have a great
correspondence with reality. An expert system based on OBIA proved to be relevant in
extracting accurate change information from remotely sensed imagery.

41
 Chapter 7
Analysis

This analysis suggests that there are two main approaches to change detection studies:
place-based and algorithm-based. Place-based studies are focused on a specific location and
aim to understand the urban land change within that area. Algorithm-based studies, on the
other hand, are more focused on the technical and methodological advances of change
detection methods and are less concerned with the specific geographic context. Place-based
studies may be more useful for policymakers and urban planners, as they provide detailed
information on the urbanization process in a particular location. Algorithm-based studies,
however, may be more useful for researchers and developers looking to improve the
effectiveness of change detection methods. In both types of studies, the location is an
important factor, but it is approached differently depending on the focus of the research.

7.1 Studies mostly track urban land transformation across short-

to medium-term timeframes

It is true that many studies focus on tracking urban land transformation over short to
medium-term timeframes, such as five to ten years. This is often due to the availability of data
and the need to understand current trends and patterns in urban development. (Yuan et al.,
2005b) used multi-date post-classification comparison change detection algorithm to
determine changes in land cover in four intervals, 1986–1991, 1991–1998, 1998–2002, and
1986–2002.

(An et al., 2007) used cloud-free Landsat TM images with the pixel size of 30m×30m
acquired on 1997-05-16 and 2004-5-19 to extract the urban change information using object-
oriented method with the spectral and texture information.

42
 In order to determine land cover and land-cover change information for forest, urban,
and agricultural categories across a seven-state region in the United States, 2000 AVHRR
satellite data was coupled with a 1980 national-scale land-cover categorization based on aerial
photo interpretation (Bergen et al., 2005).

Overall, the timeframe of urban land transformation studies depends on the research
question being asked and the availability of data. While short- to medium-term studies are
common, longer-term studies can also provide valuable insights into the history and evolution
of urban areas.

7.2 Algorithm development studies and algorithm application

studies

Research papers focused on algorithm development frequently introduce new methods

or techniques, but they can also offer innovative combinations of existing frameworks,
strategies, or tweaks to current methods. These new techniques are often compared
quantitatively to similar established techniques that use similar imagery or measure similar
types of change. These studies usually involve a test case or experiment that concentrates on a
specific study area. On the other hand, algorithm application papers often utilize established
methods or series of steps to gain an in-depth understanding of urban change in a specific
region of interest, typically concerning urban expansion. Studies that bridge both algorithm
and application groups typically start with a case study and then develop a method to answer a
specific question about a particular location. They often employ a suite of previous or
modified techniques to provide insight into a specific area.

7.3 Accuracy assessment

Evaluating the accuracy of a remote sensing algorithm is a critical factor. This

assessment involves measuring the ability of the method to detect change. Therefore, it is
essential to not only consider the location and methodology used to measure urban land
change but also the effectiveness of the approach.

43
 At the outset, the intention of the study was to evaluate and contrast the reported
accuracies of all the algorithms included in the study. However, this task presented
challenges. While 96% of the studies we examined included some form of accuracy
assessment, the assessments were frequently not readily comparable due to variations in
criteria, assessment techniques, and reporting methods. For example, certain research papers
have presented accuracy results using different measures, such as overall accuracy, user
accuracy, producer accuracy, and class-specific accuracy (e.g., for urban or forest areas), as
well as kappa statistics based on the Congalton or Congalton and Green methods. Meanwhile,
other studies have used other indicators such as the percentage of pixels that have changed or
remained the same, or positive, negative, and no-change accuracy assessments (Yuan et al.,
2005b).

There is a significant difference in the level of rigor used to assess accuracy, ranging
from studies that simply state that classified images were compared to on-the-ground status
without providing details, to studies that use detailed methodologies, sampling strategies, and
analysis of results. Qualitative visual validation methods are increasingly being used, often
involving visual comparison of sample pixels with Google Earth. While this approach allows
for ease of assessment and validation of remote areas, it lacks quantitative validation values at
the pixel level(Reba & Seto, 2020).

One alternative method for evaluating the effectiveness of an algorithm is to compare

it with an existing method. Comparison-based studies involve directly comparing and
contrasting methods using the same criteria and accuracy assessments. These studies can help
identify the strengths and weaknesses of different methodological approaches. For example,
(Im et al., 2008) conducted a study where they compared 5 different methods at the pixel and
object levels, using object and neighbourhood correlation images. They also compared a
decision tree classification algorithm with a nearest neighbour method. In another study,
(Tang et al., 2013) compared three different change vector analysis (CVA) methods:
morphological CVA, parcel-based CVA, and MBI-based CVA (morphological building
index).

This chapter discusses the two main approaches to change detection studies - place-
based and algorithm-based - and their respective benefits for policymakers, urban planners,
researchers, and developers. The timeframe of urban land transformation studies varies based
on the research question and data availability, with short to medium-term studies being

44
common. Accuracy assessment is also crucial in evaluating the effectiveness of remote
sensing algorithms, but variations in criteria and reporting methods make comparisons
challenging.

45
 Chapter 8
Discussion and Way Forward

Although there are significant gaps in knowledge regarding urban land change, there
can be five strategies that can help reduce or close these gaps. Firstly, studies on urban land
change should cover a wider geographic area, particularly in Global South cities like India
and Nigeria, where urban populations are projected to double by 2050. Secondly, there needs
to be more studies focused on smaller and medium-sized cities, which make up a significant
portion of the world's total population. Developing algorithms to detect micro-urbanization
poses new challenges due to the scale of land changes. Thirdly, monitoring urban areas more
frequently is necessary to understand urban change dynamics on a fine timescale. Fourthly,
more studies should examine within-urban and volumetric change, especially changes in
urban form and structure. Lastly, it is essential to improve the reproducibility, replicability,
and comparability of results. One approach is to consistently create and document clear
workflows that outline the series of steps, tasks, and methods used to implement an algorithm
(Reba & Seto, 2020).

This analysis provides guidance for selecting an appropriate method. One

recommendation is for users to decide whether methodology or geographic area is more
important. For example, an application-specific method may be different from a geography-
specific algorithm. The number of available images and the technical capacity of users should
also be taken into account. Simpler methods may be more appropriate for certain contexts
unless there is a clear benefit to using a highly complex method. Areas with persistent cloud
cover may not be suitable for very high temporal frequency methods (Reba & Seto, 2020).

According to the literature studied, there are gaps in algorithm development that can
be filled by the developer community. However, many user needs can be met by using
existing algorithms rather than developing something new. There is a wide range of methods
available for users who want to monitor growth. Initially, the goal was to create a
comprehensive "best practices guide" for urban land change science information users.
However, it was difficult to create such a guide for several reasons. One reason is that
algorithms have become more complex and specific as remote sensing-based change detection
science has developed. Another reason is that the types of changes urban remote sensing

46
scientists are measuring have become more diverse, from impervious surface to three-
dimensional building changes to destruction due to hazards or natural disasters. As a result, it
is challenging to recommend a few "best" methods for a specific type of change and even less
so for a particular geography (Reba & Seto, 2020).

Furthermore, accuracy assessments of different methods and types of changes may not
be directly comparable, as they often use different validation techniques. For example,
comparing binary change information with "to-from" class changes, where urban land change
is a small part of the overall land change, is challenging. Thus, it was not possible to
quantitatively compare accuracy assessments of all algorithms analyzed in a meaningful way.
Comparing methods through studies can help refine decision-making for both algorithm
developers and users, but these decisions are not always reported or explained, and providing
such explanations could be valuable for others facing similar decisions and trade-offs.

Remote sensing-based change detection can be a useful tool for assessing urban risks,
as it enables the identification and monitoring of changes in urban areas that may increase the
risk of natural disasters, environmental degradation, and other hazards. However, while
change detection provides valuable information about the physical changes occurring in urban
areas, it does not necessarily provide a complete picture of the risks and vulnerabilities
associated with those changes. To move from change detection to urban risk assessment, it is
important to consider the broader socio-economic and environmental context in which those
changes are occurring. This may involve incorporating additional data sources, such as
demographic and land-use data, and engaging with local communities to understand their
perceptions of risk and vulnerability.

One approach to urban risk assessment that incorporates both physical changes and
socio-economic factors is the Disaster Risk Reduction (DRR) framework. This framework
involves a multi-hazard approach that considers the likelihood and potential impact of various
hazards, as well as the vulnerability and exposure of the population and infrastructure. DRR
can be applied at different spatial scales, from individual buildings to entire cities, and can
inform a range of interventions, from infrastructure development to community engagement
and capacity-building. To implement the DRR framework using remote sensing-based data, it
is necessary to identify the specific hazards and vulnerabilities of the urban area in question
and develop indicators that capture those factors. These indicators can then be integrated with
change detection data to provide a comprehensive assessment of urban risks.

47
 Moving forward, the aim of my Master’s thesis will be to focus on learning about
urban risk assessment and applying it in the Indian context. This is an important and timely
topic given the increasing urbanization and vulnerability to natural disasters in India. Urban
risk assessment is a crucial tool for understanding and mitigating the risks associated with
urbanization, and it involves identifying and evaluating the hazards, vulnerabilities, and
capacities of urban areas.

To carry out my research, I will familiarize myself with the concepts, methods, and
tools used in Urban Risk Assessment (URA). This will involve conducting a comprehensive
literature review to understand the current state of knowledge and best practices in the field. I
will also be learning about the various types of hazards that are prevalent in urban areas, such
as floods, earthquakes, and landslides, and the factors that contribute to vulnerability, such as
poverty, inadequate infrastructure, and social inequalities.

Once I have a solid understanding of the theory and methods of urban risk assessment,
I can begin to apply them in the Indian context. This will involve collecting data on the
hazards, vulnerabilities, and capacities of specific urban areas in India and using this
information to develop risk profiles and maps. I will also identify and evaluate the
effectiveness of existing risk reduction measures and propose new interventions to mitigate
the risks associated with urbanization in India. Overall, my Master’s thesis on urban risk
assessment in the Indian context has the potential to make a valuable contribution to the field
of urban planning and disaster risk reduction. By applying the concepts and methods of urban
risk assessment in a real-world context, you can help to improve the resilience of urban areas
in India and contribute to the broader goal of sustainable and equitable urban development.

48
 References

Afifi, A. A., & Darwish, K. M. (2018). Detection and impact of land encroachment in El-Beheira
governorate, Egypt. Modeling Earth Systems and Environment, 4(2), 517–526.
https://doi.org/10.1007/s40808-018-0462-9

Agapiou, A., Alexakis, D. D., Lysandrou, V., Sarris, A., Cuca, B., Themistocleous, K., &
Hadjimitsis, D. G. (2015). Impact of urban sprawl to cultural heritage monuments: The case
study of Paphos area in Cyprus. Journal of Cultural Heritage, 16(5), 671–680.
https://doi.org/10.1016/j.culher.2014.12.006

Ahmad, F., Goparaju, L., & Qayum, A. (2017). LULC analysis of urban spaces using Markov
chain predictive model at Ranchi in India. Spatial Information Research, 25(3), 351–359.
https://doi.org/10.1007/s41324-017-0102-x

Ahmad, S., Avtar, R., Sethi, M., & Surjan, A. (2016). Delhi’s land cover change in post transit era.
Cities, 50, 111–118. https://doi.org/10.1016/j.cities.2015.09.003

Alqurashi, A. F., Kumar, L., & Sinha, P. (2016). Urban land cover change modelling using time-
series satellite images: A case study of urban growth in five cities of Saudi Arabia. Remote
Sensing, 8(10). https://doi.org/10.3390/rs8100838

An, K., Zhang, J., & Xiao, Y. (2007). Object-oriented urban dynamic monitoring - A case study of
Haidian District of Beijing. Chinese Geographical Science, 17(3), 236–242.
https://doi.org/10.1007/s11769-007-0236-1

Araya, Y. H., & Cabral, P. (2010). Analysis and modeling of urban land cover change in Setúbal
and Sesimbra, Portugal. Remote Sensing, 2(6), 1549–1563. https://doi.org/10.3390/rs2061549

Bergen, K. M., Brown, D. G., Rutherford, J. F., & Gustafson, E. J. (2005). Change detection with
heterogeneous data using ecoregional stratification, statistical summaries and a land allocation
algorithm. Remote Sensing of Environment, 97(4), 434–446.
https://doi.org/10.1016/j.rse.2005.03.016

Bhatt, A., Ghosh, S. K., & Kumar, A. (2018). Spectral indices based object oriented classification
for change detection using satellite data. International Journal of System Assurance
Engineering and Management, 9(1), 33–42. https://doi.org/10.1007/s13198-016-0458-7
49
Cao, H., Liu, J., Chen, J., Gao, J., Wang, G., & Zhang, W. (2019). Spatiotemporal patterns of
urban land use change in typical cities in the Greater Mekong Subregion (GMS). Remote
Sensing, 11(7). https://doi.org/10.3390/rs11070801

Casali, Y., Aydin, N. Y., & Comes, T. (2022). Machine learning for spatial analyses in urban
areas: a scoping review. In Sustainable Cities and Society (Vol. 85). Elsevier Ltd.
https://doi.org/10.1016/j.scs.2022.104050

Chen, L., Ren, C., Zhang, B., Wang, Z., & Liu, M. (2018). Quantifying urban land sprawl and its
driving forces in Northeast China from 1990 to 2015. Sustainability (Switzerland), 10(1).
https://doi.org/10.3390/su10010188

Chen, S. S., Yan, Y., Gao, Q., & Liu, D. (2015). Quantifying circular urban expansion patterns of
compact chinese cities: The case of yangtze river delta, China. Environment and Planning B:
Planning and Design, 42(2), 279–299. https://doi.org/10.1068/b120004p

Chen, X. L., Zhao, H. M., Li, P. X., & Yin, Z. Y. (2006). Remote sensing image-based analysis of
the relationship between urban heat island and land use/cover changes. Remote Sensing of
Environment, 104(2), 133–146. https://doi.org/10.1016/j.rse.2005.11.016

Chen, Y., Liu, X., & Li, X. (2017). Analyzing parcel-level relationships between urban land
expansion and activity changes by integrating Landsat and nighttime light data. Remote
Sensing, 9(2). https://doi.org/10.3390/rs9020164

Deng, J. S., Wang, K., Deng, Y. H., & Qi, G. J. (2008). PCA-based land-use change detection and
analysis using multitemporal and multisensor satellite data. International Journal of Remote
Sensing, 29(16), 4823–4838. https://doi.org/10.1080/01431160801950162

DENG, J. S., WANG, K., LI, J., & DENG, Y. H. (2009). Urban Land Use Change Detection Using
Multisensor Satellite Images Project supported by the National Natural Science Foundation of
China (NSFC) (No. 30571112). Pedosphere, 19(1), 96–103. https://doi.org/10.1016/S1002-
0160(08)60088-0

Doxani, G., Karantzalos, K., & Tsakiri-Strati, M. (2012). Monitoring urban changes based on
scale-space filtering and object-oriented classification. International Journal of Applied Earth
Observation and Geoinformation, 15(1), 38–48. https://doi.org/10.1016/j.jag.2011.07.002

He, C., Wei, A., Shi, P., Zhang, Q., & Zhao, Y. (2011). Detecting land-use/land-cover change in
rural-urban fringe areas using extended change-vector analysis. International Journal of

50
 Applied Earth Observation and Geoinformation, 13(4), 572–585.
https://doi.org/10.1016/J.JAG.2011.03.002

Huang, Q., Yang, X., Gao, B., Yang, Y., & Zhao, Y. (2014). Application of DMSP/OLS nighttime
light images: A meta-analysis and a systematic literature review. In Remote Sensing (Vol. 6,
Issue 8, pp. 6844–6866). MDPI AG. https://doi.org/10.3390/rs6086844

Im, J., Jensen, J. R., & Tullis, J. A. (2008). Object-based change detection using correlation image
analysis and image segmentation. International Journal of Remote Sensing, 29(2), 399–423.
https://doi.org/10.1080/01431160601075582

Lu, D., Mausel, P., Brondízio, E., & Moran, E. (2004). Change detection techniques. International
Journal of Remote Sensing, 25(12), 2365–2401.
https://doi.org/10.1080/0143116031000139863

Pandey, B., Zhang, Q., & Seto, K. C. (2018). Time series analysis of satellite data to characterize
multiple land use transitions: a case study of urban growth and agricultural land loss in India.
Journal of Land Use Science, 13(3), 221–237.
https://doi.org/10.1080/1747423X.2018.1533042

Reba, M., & Seto, K. C. (2020). A systematic review and assessment of algorithms to detect,
characterize, and monitor urban land change. In Remote Sensing of Environment (Vol. 242).
Elsevier Inc. https://doi.org/10.1016/j.rse.2020.111739

Seto, K. C., Woodcock, C. E., Song, C., Huang, X., Lu, J., & Kaufmann, R. K. (2002). Monitoring
land-use change in the Pearl River Delta using Landsat TM. International Journal of Remote
Sensing, 23(10), 1985–2004. https://doi.org/10.1080/01431160110075532

Small, C., Elvidge, C. D., Balk, D., & Montgomery, M. (2011). Spatial scaling of stable night
lights. Remote Sensing of Environment, 115(2), 269–280.
https://doi.org/10.1016/j.rse.2010.08.021

Tang, Y., Huang, X., & Zhang, L. (2013). Fault-tolerant building change detection from urban
high-resolution remote sensing imagery. IEEE Geoscience and Remote Sensing Letters, 10(5),
1060–1064. https://doi.org/10.1109/LGRS.2012.2228626

Tewkesbury, A. P., Comber, A. J., Tate, N. J., Lamb, A., & Fisher, P. F. (2015). A critical
synthesis of remotely sensed optical image change detection techniques. In Remote Sensing of
Environment (Vol. 160, pp. 1–14). Elsevier Inc. https://doi.org/10.1016/j.rse.2015.01.006

51
WUP2011_Report. (n.d.).

Yuan, F., Sawaya, K. E., Loeffelholz, B. C., & Bauer, M. E. (2005a). Land cover classification and
change analysis of the Twin Cities (Minnesota) metropolitan area by multitemporal Landsat
remote sensing. Remote Sensing of Environment, 98(2–3), 317–328.
https://doi.org/10.1016/j.rse.2005.08.006

Yuan, F., Sawaya, K. E., Loeffelholz, B. C., & Bauer, M. E. (2005b). Land cover classification and
change analysis of the Twin Cities (Minnesota) metropolitan area by multitemporal Landsat
remote sensing. Remote Sensing of Environment, 98(2–3), 317–328.
https://doi.org/10.1016/j.rse.2005.08.006

Zheng, Q., Weng, Q., & Wang, K. (2020). Correcting the Pixel Blooming Effect (PiBE) of DMSP-
OLS nighttime light imagery. Remote Sensing of Environment, 240.
https://doi.org/10.1016/j.rse.2020.111707

What are the band designations for the Landsat Satellites? (n.d.). Retrieved April 28, 2023,
from https://www.usgs.gov/faqs/what-are-band-designations-landsat-satellites

GHSL - Global Human Settlement Layer. (2016, July 06). Retrieved April 28, 2023, from
https://ghsl.jrc.ec.europa.eu/

Agency, E. (n.d.). Spot 5. Retrieved April 28, 2023, from

https://earth.esa.int/eogateway/missions/spot-5

52
53