Landcover Analysis Of Delhi
A Project report submitted in partial fulfillment
of the requirements for the award of the degree
of Master of Science (Data Science) of CHRIST
(Deemed to be University)
December - 2024
1
TABLE OF CONTENTS
● Abstract
● Introduction
● Problem Statement
● Scope
● Objective
● Study Area
● Data and Methodology
● Results
● Comparative Study
● Discussion
● Challenges
● Conclusion
● Recommendations
● References
● Contribution
ABSTRACT
2
The next paragraph gives a series of astonishing facts based on Delhi's land cover over a five-
year period from 2019 to 2024, with a focus on urbanization and its effects on natural resources.
It examines the alteration of several types of land cover, including urban areas, agricultural land,
tree cover, freshwater sources, and land in dry regions, and assesses the extent to which these
changes affect the environment and sustainability of the capital city region.
The study aims at identifying significant patterns linked to the conversions and variations of land
use and land cover, observing the level of modification of urbanised regions, and providing
information on the extent of reduction or increase of various forms of natural.
The primary source of data for the investigation was satellite photos obtained from Earth
Explorer. QGIS was used for data preprocessing and geographical analysis. Meanwhile, many
advanced algorithms, including Multilayer Perceptron, Random Forest, and K-Means clustering,
were used to classify and identify the patterns. The resulting results were validated using various
accuracy assessment tools, such as a confusion matrix and ground truthing. The research
indicates that one of the most significant findings of the analysis is that urbanisation has
increased by more than 15% as the popularity of the topic grows.
INTRODUCTION
Land cover analysis is an important tool for understanding the complicated relationship between
human activities and environmental change. It explains how man-made pressures like
urbanisation, deforestation, and industrialisation affect natural landscapes. Rapid urbanisation
has been a prominent global trend in recent decades, particularly in rising economies like India.
Urbanisation regularly converts woods, agricultural regions, and bodies of water into built-up
areas, with major environmental consequences. These consequences include habitat loss, reduced
biodiversity, urban heat island effects, and overexploitation of natural resources such as
groundwater and fertile soil.
Delhi, the capital of India, is a good example of this trend, having undergone dramatic change in
recent years due to population growth, economic activity, and infrastructure development. With
an estimated population of over 32 million, Delhi has become one of the most populous
metropolises in the world. The rapid pace of urban growth in the region is putting enormous
pressure on its limited natural resources. This has led to serious environmental problems such as
shrinking agricultural land, loss of vegetation cover, reduced freshwater sources and increased
levels of air and water pollution. Moreover, declining forest cover and water bodies increase
Delhi’s vulnerability to climate change, contributing to rising temperatures, erratic weather
patterns and poor air quality.
The relationship between urbanisation and environmental deterioration in Delhi emphasises the
importance of rigorous, data-driven evaluations of land cover change. These studies will not only
aid in the identification of geographical and temporal trends, but will also serve as a foundation
for informed decision-making and long-term urban planning. By researching how land cover
changes over time, policymakers and planners can identify crucial locations for intervention and
develop policies to limit the harmful effects of sprawl.
3
PROBLEM STATEMENT
Rapid urbanisation in Delhi has resulted in major changes in land cover between 2020 and 2024,
as well as increased demand for natural resources such as forests, agricultural fields, and water
bodies. These changes are posing additional issues, including biodiversity loss, limited
freshwater sources, urban heat islands, and rising environmental demands. These changes not
only threaten the natural balance of the region, but also raise concerns about climate, resource
use, and socioeconomic disparities.
The problem is the lack of thorough, data-driven analysis to identify and understand these
changes. Without such research, policymakers and urban planners cannot develop long-term
solutions to mitigate the negative impacts of urbanization. This study fills this gap by
investigating spatial and temporal patterns of land cover change, assessing environmental
impacts, and proposing practical recommendations for sustainable urban development in Delhi.
SCOPE
The primary goal of this research is to investigate changes in land cover inside the limits of Delhi
over a five-year period (2020-2024). It investigates changes over time and space in urban areas,
agricultural land, tree cover, bare soil, and water bodies. The study analyses and validates
environmental implications such as resource stress, urban heat islands, and biodiversity loss
using QGIS and machine learning approaches. The findings are intended to guide sustainable
urban design and policy actions to mitigate the challenges caused by rapid urbanisation.
OBJECTIVE
The main objective of this study is to assess and quantify land cover changes in the Delhi region
from 2020 to 2024. By identifying land use patterns and trends and changes in the countryside,
the purpose of the study is to assess their environmental impact and to provide recommendations
for sustainable urban development.
Specifically, the purpose of the study:
4
● Analyze the scale of urbanization and the spatial amount of the Delhi during the study
period.
● Identify areas with serious losing agricultural land, vegetation and water body.
● Consider land conversion in infertile and dry areas thanks to urban expansion and
resource development.
● Assess the rate and causes of deforestation and its impact on overall environmental
pressures.
● Study of urban heat island patterns resulting from increased urbanization.
● Assess changes in freshwater supplies and assess their implications for resource
management.
● Gain insight into the impact of these changes on environmental sustainability,
biodiversity and resource availability.
● Inform city planners and political decision -making strategies for balanced urban
development, environmental protection and flexibility.
The study uses advanced geographical space tools and machine learning algorithms to analyze a
wide range of Delhi -alternating Earth coverage, and emphasizes the urgent need for sustainable
intervention measures to address the problems caused by urbanization.
STUDY AREA
Geographical Context
The capital of India, Delhi, is located in the northern part of the country. It covers an area of
1,484 square kilometers and is home to over 20 million people. Bounding with Harjan to the
west and the Utaris east, Delhi is located along the Yamuna River, which is the key but an
increasingly contaminated source of water. Most of the city, flat terrain experiences extreme
seasonal temperature variations.
Main Functions:
● Yamuna River: An essential source of freshwater that faces challenges with
contamination and shrinking volume.
5
● Green Space: including organic hot spots such as Lodhi Gardens and Delhi Ridge, they
are threatened by rapid urbanization.
● Industrial Area: This is important for the economy of the region, but it is important to
promote environmental degradation.
Maps
The study includes a map of Delhi showing administrative boundaries, water bodies, vegetation
zones and urban areas, giving a clear overview of the city's land cover.
DATA AND METHODOLOGY
Data Used
The data for this study comes primarily from Landsat satellite imagery (collection 8 or 9)
acquired using Earth Explorer. Data collected by seven satellite constellations over a five-year
period (2024-2024) was used to examine land cover changes in Delhi.
These bands hold spectral information that helps differentiate between different types of land
cover, such as urban areas, forests, water bodies, and agricultural land. They include visible,
near-infrared, and short-wave infrared data, offering a detailed perspective of the ground surface.
Preprocessing
Satellite data processing included several critical measures to prepare data for classification and
analysis:
6
● Cutting: The original satellite images were reduced to the deli borders to concentrate the
analysis of the geographical area in question. This process removed unnecessary parts of
the images and ensured that only this area of interest was considered in the further
analysis.
● Visualization: To visually interpret the ground cover data, 7 bands were combined using
a combination of RGB 4-3-2 bands. This method uses an almost infrared (NIR) band in
red, red (R) green and green (g) blue. This combination helps in clearly visualizing the
land surface features as different land cover types (such as water, urban and vegetation)
can be easily distinguished by their unique colors in this specific band combination. The
resulting images were visually interpreted to observe different land cover types across
Delhi.
Creating Training Dataset
For land cover classification, a training dataset is required. To build this dataset, ROC polygons
(regions of interest) were manually drawn on satellite images to identify representative areas for
each land cover class. The training dataset consists of five primary land cover classes:
● Crops: agricultural land and fields.
● Water: rivers, lakes, and other bodies of water. Forest: Forest areas and green areas.
● City: Built-up area, including residential, commercial and industrial areas. Barren: Areas
with little vegetation, including deserts and cities.
These regions are labeled and used to train a machine-learning model for classification.
Landcover Classification
Using machine learning techniques to classify land cover in Delhi, Use the following algorithm
for classification:
● Multilayer Perceptron (MLP): A type of neural network suitable for identifying patterns
in data. MLP was used to classify satellite images based on a training dataset.
● Random Forest (RF): An ensemble learning method that uses multiple decision trees to
improve classification accuracy. RF is used to identify the type of field coverage, taking
into account the different properties of the images.
● Support vector machine (SVM): Supervision learning algorithm used to classify by
finding the optimal superplane that separates different rural categories for classification.
These models were trained using the previously created training data sets, and each algorithm
was classified for rural coverage according to the properties of spectral properties.
Tools and Software
The analysis utilized QGIS (Quantum GIS), a robust open-source geographic information system
that offers various tools for data visualization, processing, and analysis. The Semi-Automatic
Classification Plugin (SCP) in QGIS was employed for the subsequent tasks:
7
● Preprocessing: Clip the satellite imagery to the study area and visualize the data using
RGB band combination. Classification: The SCP plugin helps implement and run
machine learning classification algorithms (MLP, RF, and SVM).
● Visualization: QGIS provides functionality to visually classify land cover maps and
interpret the results spatially. These tools can efficiently process large satellite data sets
and create accurate land cover classifications.
Validation
An important part of land cover classification is validating the results to assess their accuracy.
Unfortunately, this study could not validate the classification results due to the lack of ground
truth data, which is essential for comparison. Ground truthing typically involves matching
classification results with established land cover maps or field observations. As these data were
not available, no formal validation steps were performed for this study. However, we performed
a visual inspection of the classification results to ensure that they were consistent with the
recognized land cover characteristics and trends of Delhi.
RESULTS
Landcover classification was conducted annually using MLP and Random Forest (RF)
classifiers. The extent of various landcover classes, such as urban, water, crops, barren land, and
forests, was examined to identify changes over the years. Below is a detailed breakdown for each
classifier.
Categories/Classes
MLP Classifier Results
The following table summarizes the area (in square meters) covered by each landcover class as
classified by the MLP algorithm:
Yea Urban Water Crop Barren Forest
r
2020 151,682,400 18,959,040 53,533,800 151,682,400 56,669,400
8
2021 453,756,600 44,325,000 188,646,300 99,159,300 68,115,300
2022 576,231,300 22,252,500 289,546,200 259,485,300 330,442,200
2023 654,920,100 127,419,300 36,050,050 648,131,400 42,633,900
2024 1,005,425,100 28,893,600 84,339,900 74,898,900 275,584,500
Interpretation (MLP)
1. Urban Areas: Urban sprawl has increased significantly from 2020 to 2024, with the area
more than doubling over the five-year period.
2. Water Bodies: A slight reduction in water bodies is visible, particularly after 2021,
where the area shows a consistent decline.
3. Crop and Forest Lands: There is an increasing trend in crop areas up to 2022 but a
decline afterward. Forested areas also fluctuate significantly, showing a loss of green
cover between 2020 and 2023, followed by partial recovery in 2024.
2020
2021
9
2022
2023
2024
10
Random Forest Classifier:
RF Classifier Results
The following table summarizes the area (in square meters) covered by each landcover class as
classified by the RF algorithm:
Yea Urban Water Crop Barren Forest
r
11
2020 422,184,600 24,096,600 426,000,000 426,000,000 377,000,000
2021 603,909,000 53,237,700 396,000,000 55,100,000 359,000,000
2022 787,565,700 27,799,200 168,000,000 120,000,000 366,000,000
2023 926,383,500 128,000,000 98,700,000 300,000,000 200,000,000
2024 1,082,323,800 60,380,100 110,000,000 133,000,000 91,400,000
Interpretation (RF)
1. Urban Areas: Similar to MLP, urban areas show consistent growth over the years,
reaching the highest levels in 2024.
2. Water Bodies: A slight increase in the water body area is observed in 2023, but there is a
reduction in 2024.
3. Crop and Forest Lands: The forest area sees a decline after 2020 and continues to
shrink significantly by 2024. Crop areas also show a notable drop after 2021.
4. Barren Lands: Unlike MLP, RF results indicate a more consistent decrease in barren
land after 2020, with minor increases in 2024.
2020
2021
12
2022
2023
13
2024
Landcover Distribution and Temporal Changes
Year-wise Landcover Distribution (2020 to 2024)
14
For each year, landcover classification was performed using MLP and Random Forest (RF)
classifiers. The area covered by different landcover classes, including urban, water, crop,
barren, and forest, was analyzed to detect changes over the years. Below is the detailed
distribution for each classifier.
MLP Classifier Results
The following table summarizes the area (in square meters) covered by each landcover class as
classified by the MLP algorithm:
Yea Urban Water Crop Barren Forest
r
2020 151,682,400 18,959,040 53,533,800 151,682,400 56,669,400
2021 453,756,600 44,325,000 188,646,300 99,159,300 68,115,300
2022 576,231,300 22,252,500 289,546,200 259,485,300 330,442,200
2023 654,920,100 127,419,300 36,050,050 648,131,400 42,633,900
2024 1,005,425,100 28,893,600 84,339,900 74,898,900 275,584,500
Interpretation (MLP)
1. Urban Areas: Urban sprawl has increased significantly from 2020 to 2024, with the area
more than doubling over the five-year period.
2. Water Bodies: A slight reduction in water bodies is visible, particularly after 2021,
where the area shows a consistent decline.
3. Crop and Forest Lands: There is an increasing trend in crop areas up to 2022 but a
decline afterward. Forested areas also fluctuate significantly, showing a loss of green
cover between 2020 and 2023, followed by partial recovery in 2024.
4. Barren Lands: The barren land area shows sharp fluctuations, peaking in 2023 before
reducing in 2024.
RF Classifier Results
The following table summarizes the area (in square meters) covered by each landcover class as
classified by the RF algorithm:
15
Yea Urban Water Crop Barren Forest
r
2020 422,184,600 24,096,600 426,000,000 426,000,000 377,000,000
2021 603,909,000 53,237,700 396,000,000 55,100,000 359,000,000
2022 787,565,700 27,799,200 168,000,000 120,000,000 366,000,000
2023 926,383,500 128,000,000 98,700,000 300,000,000 200,000,000
2024 1,082,323,800 60,380,100 110,000,000 133,000,000 91,400,000
Interpretation (RF)
1. Urban Areas: Similar to MLP, urban areas show consistent growth over the years,
reaching the highest levels in 2024.
2. Water Bodies: A slight increase in the water body area is observed in 2023, but there is a
reduction in 2024.
3. Crop and Forest Lands: The forest area sees a decline after 2020 and continues to
shrink significantly by 2024. Crop areas also show a notable drop after 2021.
4. Barren Lands: Unlike MLP, RF results indicate a more consistent decrease in barren
land after 2020, with minor increases in 2024.
Comparison of Classifiers
● Urban Classification: Both MLP and RF show increasing urban sprawl, but RF tends to
classify more area as urban compared to MLP.
● Water Bodies: RF generally detects slightly more water body coverage than MLP.
● Crop and Forest Lands: MLP shows more fluctuation in these areas compared to RF,
which indicates a more consistent trend.
● Barren Lands: There is greater variability in the barren land classification by MLP,
whereas RF shows a more gradual trend.
Selected Classifier
Based on the classification accuracy and visual inspection of results, Random Forest (RF) was
selected as the preferred model for further analysis due to its ability to handle high-dimensional
data and provide stable results.
Comparative Analysis (2020–2024)
16
The comparative analysis across years reveals some critical landcover changes in Delhi:
Urbanization Trends:
● From the data, it is evident that urban areas have shown a consistent and
significant increase in coverage from 2020 to 2024 for both classifiers (MLP and
RF).
● For instance, in 2020, MLP estimated urban coverage at 151,682,400 hectares,
which rose dramatically to 1,005,425,100 hectares by 2024. Similarly, RF
predictions showed an increase from 422,184,600 hectares in 2020 to
1,082,323,800 hectares in 2024.
● This pattern clearly indicates rapid urbanization, likely at the expense of green
cover, barren land, and potentially waterbodies.
● Such a significant rise in urban areas could point to increased infrastructure
development, population growth, and changes in land use patterns.
Waterbody Changes:
● Waterbody areas have shown a mixed trend with fluctuations across the years for both
classifiers.
● For example, RF estimates water coverage as 24,096,600 hectares in 2020, decreasing to
6,038,010 hectares in 2024. MLP also shows a decline, with water areas dropping from
189,590,400 hectares in 2020 to 28,893,600 hectares in 2024.
● These fluctuations could indicate seasonal variations, potential drying up of waterbodies,
or misclassification issues. However, the general trend suggests a decline in waterbody
areas, which could be a result of urban sprawl, climate change, or improper water
management.
17
Forest Cover:
● Forest areas have experienced a decrease in both MLP and RF classifications, though the
rate of decline differs between the two methods.
● For instance, RF estimates show forest cover at 377,000,000 hectares in 2020,
decreasing to 91,400,000 hectares in 2024. MLP shows a similar reduction, though the
numbers vary.
● This decline could point to deforestation for agriculture, urbanization, or other land use
activities. The consistent reduction highlights a critical environmental concern and the
need for conservation measures.
Barren Land:
● Barren land areas have fluctuated significantly year-to-year. For example, RF estimates
barren areas at 426,000,000 hectares in 2020, decreasing to 133,000,000 hectares in
2024. MLP also shows variability, but the values are generally lower.
● This pattern could indicate either reforestation, land reclamation, or changes in land
management practices. However, the drop in barren land alongside the rise in urban areas
suggests a shift toward development rather than natural regeneration.
18
Crop Areas:
● Crop areas, as predicted by MLP, have shown significant growth from 53,533,800
hectares in 2020 to 84,339,900 hectares in 2024. RF, on the other hand, shows
fluctuating trends with no clear pattern.
● The increase in crop areas could indicate improved agricultural practices, expansion into
previously barren land, or better land classification techniques in recent years.
Classifier Comparisons:
● MLP vs RF: Across all years, both MLP and RF classifiers show similar trends in urban
expansion and forest decline. However, their estimates of absolute areas vary
significantly. For instance, RF tends to classify larger portions as urban and forest
compared to MLP.
● For 2024, RF predicts a higher urban area (1,082,323,800 hectares) compared to MLP
(1,005,425,100 hectares), while MLP shows higher barren areas for most years. These
differences could be attributed to variations in the algorithms’ sensitivity to feature
boundaries or training data.
● Selecting a classifier for final analysis depends on its performance metrics like accuracy,
precision, and recall on validation datasets.
19
DISCUSSION
Interpretation of Results:
The observed changes in landcover distribution over the years reflect significant human-induced
and environmental transformations:
● Urbanization: The rapid increase in urban areas is likely driven by population growth,
industrialization, and infrastructure development. The data suggests that natural
landscapes, such as forests and barren lands, have been converted into urban spaces.
Policy decisions to prioritize urban expansion over conservation might have accelerated
this trend.
● Deforestation and Loss of Green Cover: The consistent decline in forest areas indicates
increasing land clearance for agriculture, urban construction, and industrial zones. This
could result in reduced biodiversity, habitat loss, and diminished ecosystem services like
carbon sequestration and water cycle regulation.
● Decline in Waterbodies: The reduction in waterbody areas highlights potential issues
such as urban encroachment, overuse of water resources, and lack of effective water
management policies. Seasonal or climatic factors, such as decreasing rainfall or rising
temperatures, may have also contributed to shrinking water coverage.
● Barren Lands: Fluctuations in barren land areas suggest changing land use dynamics,
possibly including reclamation for agriculture or urban development.
Impacts on the Environment
● Biodiversity Loss: The reduction in forest cover and waterbodies adversely affects flora
and fauna, disrupting ecological balance.
● Climate Impacts: Urbanization and deforestation contribute to increased greenhouse gas
emissions, urban heat islands, and altered rainfall patterns.
● Hydrological Cycle: Decreasing waterbody areas may lead to reduced groundwater
recharge and increased vulnerability to droughts.
CHALLENGES
During the analysis, several challenges were encountered:
1. Color Changes in Clustering Results:
While performing K-Means clustering, the colors assigned to clusters were inconsistent,
making it difficult to interpret the results visually across different iterations.
2. Inaccuracy of Clustering Results:
The clustering approach did not produce accurate or meaningful separation of landcover
classes, leading to unreliable insights. As a result, we transitioned to classification
techniques such as Random Forest and MLP for better precision.
3. Absence of Ground Truth Data:
The lack of a ground truth file posed a significant challenge in validating the
20
classification results. This limited the ability to assess the accuracy of the models and the
reliability of the derived landcover maps.
Despite these challenges, the study was able to provide valuable insights into the landcover
changes in Delhi using the available resources and advanced classification techniques.
CONCLUSION
This study reveals significant landcover changes in Delhi from 2020 to 2024, with rapid
urbanization leading to a decline in forest cover, barren lands, and waterbodies. These findings
emphasize the urgent need for sustainable urban development and conservation efforts. Using
advanced classification techniques like MLP and RF, the analysis provides a foundation for
informed decision-making. Future work could involve higher-resolution data, longer timeframes,
and socio-economic overlays to better understand these trends.
RECOMMENDATIONS
1. Promote sustainable urban planning with green spaces and efficient infrastructure.
2. Enforce afforestation programs and protect existing green cover.
3. Preserve and rejuvenate waterbodies through conservation policies.
4. Integrate environmental goals into urban development plans.
5. Raise public awareness and engage communities in conservation efforts.
These steps can help Delhi achieve a balance between development and environmental
sustainability.
REFERENCES
● Landsat 8 and 9 Satellite Imagery. Retrieved from Earth Explorer.
● QGIS 3.30 with Semi-Automatic Classification Plugin (SCP).
CONTRIBUTIONS
1. Alex [2348307]: Classification for the year 2023, Documentation (interpretation)
2. Catharin Nivitha[2348314]: Classification for the year 2022, Documentation
(preprocessing, interpretation, problem statement, comparative study)
3. Hitesh Saini [2348324]: Classification for the year 2021, Documentation (interpretation,
abstract, problem statement)
21
4. Sree Vani Bandi [2348360]: Classification for the year 2024, Documentation
(interpretation, comparative study)
5. Yashashwy Alok [2348467]: Classification for the year 2020, Documentation
(interpretation)
22