Monitoring Surface Water Dynamics Under Climate Variation

1. Background & Study

Due to the increased impact of climate change and subsequent changes in the global
and local water cycles, the measurement of changes that have taken place in surface
waters has become a priority area of research. Surface water sources such as rivers,
lakes, reservoirs, swamps, and streams are very important sources of fresh water.
These sources help to promote ecosystems, agricultural systems, and industrial
systems, produce electricity by generating hydro, power, and support the survival of
living organisms such as human beings. However, the effects of climate change have
increasingly caused changes in the nature of different water sources. Changes in the
amount of rainfall as well as the increased intensity of both rainy and dry seasons due
to higher temperature that accelerates the rate of evaporation have resulted in more
changes in the sources of surface waters. Consequently, different locations have
experienced the problem of either drought or flood.

If one possesses the knowledge of how surface water bodies are formed, it will
definitely be advantageous in the proper management of water resources. This, in turn,
will help to keep track of the future level of water resources. Besides this, it aids in
the understanding of the impact of global warming. It is also useful in drafting plans
on how to cope with the changes.

Traditionally, monitoring of hydrological resources has been through earth

observation. Besides that, observation for monitoring is done through observation
points. In addition, it includes the use of manual observation. It should be noted that
the above methods involve accurate data.

However, there are some limitations in this particular field. Some of these limitations
include an extension of time in the release of data. Besides that, the data may not be
readily available to everyone.

Nevertheless, satellite monitoring and remote sensing have greatly changed the way
we monitor the environment. That's because a huge amount of data is generated on the
Earth's surface. Today's satellites can deliver a temporal sequence of images that are
clear, accurate, and taken at specific intervals. These images can provide information
about both the land and the water at a wider scale. As the satellite records can show
historical trends compared to the current trends in surface waters, it has been found as
a very effective alternative method as compared to the conventional one. Especially
when data on the surface is not accessible.

Free satellite data and advancement in remote sensing over the last few years have
made it possible to effectively monitor the surface water. As the water spectral indices
derived from the satellite images play a crucial role in the identification of water from
other land cover types. Among these, the Normalized Difference Water Index (NDWI)
has been generally accepted due to its simplicity and effectiveness in enhancing open
water features while suppressing the effects of vegetation and soil. The use of such
indices facilitates the consistent detection of water bodies over different time periods.
Besides the detection of surface water, the monitoring of the temporal changes of
water bodies is an equally important aspect. Climatic variation leads to seasonal as
well as long, term changes in water extent, which single, date observations are unable
to depict accurately. Multi, temporal satellite images make it possible to quantify
changes of surface waters distribution over time and consequently, one can detect
such changes as water bodies' growth, drought, or disappearance. Such a temporal
viewpoint is crucial for the interpretation of the influence of climatic variability on
water systems.

Monitoring surface water from space instruments has its main benefits compared to
the classic ways of water monitoring that are mainly on the ground, namely:

covering a large area at once, performing repeated measurements, and being relatively
[Link] of these expediencies, space, based water measurements are highly
effective in the areas of the world where on, ground measuring instruments are very
few or do not exist at all. Furthermore, space images facilitate uninterrupted
monitoring, which is indispensable for early warning of flood and drought events.

In this context, the monitoring of surface water changes under climate change through
remote sensing and the use of information technology has gained significant
momentum.

Satellite data can be combined with image recognition that works by itself without
human assistance to locate the spots of a change in surface water bodies such as rivers,
lakes, and reservoirs and analyze it in depth. This results in a more water, related issue
management than before, a better understanding of the impacts of climate change, and
a shift towards a more sustainable development planning in line with coming water,
related problems.

1.2 EXISTING SYSTEM

1. Automated systems are currently not possible

In most of the conventional methods of water level monitoring, the analysis involves
some manual or fully manual analysis of remotely sensed images, which tends to
require manual delineation of water bodies and comparison of images through time.
There exist automated systems for image segmentation and classification, but they
tend to need constant adjustments according to different environments, sensors, and
geographical positions, which limits scalability and repeatability of analysis through
time series and larger geographic scales for obtaining information regarding changes
in dynamics of surface water due to climatic changes.

2. Accuracy in the detection of shallow or mixed water pixels.

Existing mapping methods require rigid values or indices based on reflectance, failing
to provide a distinctive difference in areas that contain deep water, shallow water,
vegetation under water, saturated soil areas, or areas containing cloudy or sediment-
dense water masses. By changing under different climatic conditions like flooding
events, drought conditions, or enhancements in evaporation levels, spectral properties
tend to change erratically, thus impairing accuracy in their measurement and
distinction simultaneously. Consequently, there has been a high propensity towards
misclassifying shallow and transition areas in water mass determinations.

3. High processing time for multi-date image comparison.

Surface-water analyses cannot be done without the consideration of remote scenes
captured on various dates to identify the changes with the passage of time. Current
methods in remote systems involve various preprocessing tasks for each date to
remove noise, identify the cloudy region, determine the threshold, and perform
atmospheric correction. Aligned processing with manual comparisons between the
various data sources introduces complexities in the processing tasks, causing delay in
the output of the results. Delays in processing can limit the ability to react to changes
in water resources, such as the rapid increase in flood levels.

4. Generic difficulty in analyzing the influence of climate on water.

Most of the current models are designed to isolate the static extent of water and not
the trend related to the climate forcing elements like rainfall anomalies, temperature
rise, and evapotranspiration variations. In the absence of integrated analysis, the
relationship between hydrological and climatic elements can be difficult. Additionally,
the lack of automatic temporal trend analysis results in fragmented data, making the
determination of water body contraction and expansion, as well as future risk, difficult.

5. Dependence on manual interpretation of satellite images.

Ambiguous or unclear areas in satellite images now require analysts to manually
analyze them. Analysts also have to enhance classification results by determining
water boundaries. It should be remembered that these processes involve a human
element, leading to inconsistencies in the results obtained by analysts, especially
where several analysts with varying levels of understanding analyze various study
periods.

6. Insufficient visualization and tracking of data.

The outputs are typically in the form of static images or separate classification maps
for each date. The isolated output makes it impossible to see the temporal changes in
the water systems and gain insights into the changes in the water hydrological level as
a consequence of the climatic changes. The lack of visualization pipelines or the
absence of a temporal series visualization feature would make the long-term
monitoring a cumbersome task, and the necessary yearly or seasonal variability would
remain unseen.

CHALLENGES IN EXISTING SYSTEMS

1. Today, the current methods for monitoring the amount of water

Current water resource monitoring systems are heavily dependent on the traditional
methods of remote sensing, whereby human intervention takes place in the system
from data selection, processing, classification, to validation. Processes like the
selection of the appropriate indices for spectral analysis, establishment of the
threshold levels, and error correction in the classification process have to be
performed manually in the system. Thus, the partial automation of the system makes
the processing of the large quantities of satellite data inefficient, and the system will
not be capable of continuous operation and near-real-time monitoring of the water
resources.
2. Low accuracy for detecting shallow water pixels and mixed pixels.
Current methods encounter difficulties in correctly detecting shallow water features
and mixed pixels, for which water is amalgamated with vegetation, ground, and built-
up pixels or features in a single pixel. This is primarily because shallow water and the
surrounding earth cover spectral similarities, mainly when the water has high levels of
turbidity or is occupied by aquatic vegetation. As a result, small, temporal, or
fragmented water features are either inaccurately classified or entirely overlooked.

3. High processing time for multi-date image comparison:

For studying the dynamics of water, it is essential to analyze satellite images taken on
different dates. Traditional methods entail several preprocessing steps, such as
atmospheric correction and pixel-wise comparisons, at each interval. The process is
quite intensive in terms of processing and memory requirements. The longer the time
interval for which the monitoring system is designed, the longer will be the processing
time, and thereby the system will not be very effective for real-time monitoring.

4. Climate effect on changes in water difficulty of analysis.

Currently, most systems used for water resource monitoring are essentially concerned
with the automatic extraction of water features from satellite images alone, without
considering the integration with climate factors like precipitation, temperature,
evaporation rates, and drought levels. As a result, it is hard to develop relationships
between the changes that are identified in water and the climate factors that cause
these changes. This hinders the system’s ability to determine the climate change
impacts on water resource levels.

5. Dependence on manual interpretation of satellite images.

A major issue in current solutions is that expert systems are largely dependent upon
manual interpretation to carry out the assessment of water bodies and verification of
results. Typically, a satellite image analysis is performed manually by experts to
remove misclassifications and enhance accuracy, which proves to be a resource-
intensive and inefficient process. At the same time, human interpretation comes with
uncertainties because different experts tend to produce conflicting outcomes based
upon personal expertise and interpretations.

6. Lack of visualization and tracking.

The existing methods are usually producing static maps or reports about a certain time
period and are not capable of offering standardized visualization support. The fact that
there is no standardized approach to long-term tracking makes it complicated to
visualize trends or make comparative assessments about the historical changes that
have taken place within the water resources or to spot long-term processes about
expansion or contraction.

problem statement
The focus of surface water dynamics under climate variability is to develop a system
that can monitor and manage changes in surface water extent from satellite pictures.
The system should be able to automatically detect and analyze changes in surface
water with little human intervention. It is quite possible if image processing
techniques combined with basic machine learning algorithms are used to
automatically separate water from other surface features. This guarantees consistency
and rapid processing of huge areas.
Apart from water feature extraction, the model also uses climatic variables such as
rainfall, temperature and drought indices, thus enabling it to determine the impact of
these variables on surface water fluctuations. By setting the relationships between the
climatic variables and the alterations of the water area, the model gives a better
understanding of how climatic variations and events influence water levels.

It is a matter of fact that this proposed system will support the enhancement of the
visualization and long, term tracking capabilities for the decision, making process in
water resources management for the more sustainable and positive handling of surface
water resources despite the climatic variation changes.

PROPOSED SYSTEM

The proposed system offers an automated tool for tracking changes in surface water
through the integration of remote sensing and climate data. Reliable Earth
Observation satellites provide multi, temporal satellite images that show the changes
over time. The Normalized Difference Water Index (NDWI) is applied to firstly
highlight water features and secondly to suppress non, water features so as to facilitate
the delineation of surface water bodies.

In order to significantly improve the accuracy of the system in such challenging

situations like shallow and/or turbid water as well as different land covers, the authors
resort to machine learning classification methods. The authors try to separate water
pixels from vegetation, soil, and man, made surfaces using spectral and spatial
features.

The system starts by performing multi, temporal analysis to identify the expansion,
contraction, and variation of water levels after automated change detection. The
system then integrates rainfall and temperature data to determine statistically how
fluctuations in the climate influence aquatic surface levels. The results of the analysis
are presented in the system through thematic maps and temporal graphs.

OBJECTIVES
The primary goal of the project would be the investigation of multi-temporal satellite
image information with the purpose of automatically extracting and analyzing the
changes in the water bodies on the Earth’s surface. In addition, the project will
evaluate the effects of climate variability on water extent through the application of
climate information and machine learning methods.
 For pre-processing satellite images to extract water bodies accurately.
 To identify and assess water regions with NDWI, basic machine learning
techniques, as well as different time periods.
 To evaluate the climate data in order to interpret the variations in water area.
 To be able to view the results in a graphical manner via maps and graphs.
Methodology

1. Sentinel-2
The sentinel-2 satellite offers high-resolution multispectral satellite photos with
visible and near-infrared channels, which is critical for the observation of surface
water. The satellite photos are free and can be used for long-term observation of water
for different climatic conditions.
2. Clipping to Upper Lake Clipping:
Clipping helps in isolating the study area from the satellite image based on the
boundary of Upper Lake. The purpose of this process is to eliminate unwanted
information from the image.
3. NDWI Calculation
Normalized Difference Water Index (NDWI)” is derived from the green and NIR
bands. NDWI accentuates the difference in reflectance between water and vegetation,
with positive pixels indicating the presence of water.
4. Generating water mask

The binary water mask is produced based on the threshold of the values of NDWI.
The pixels representing water will be shown in white, and the pixels representing non-
water in black.
5. Tile Creation (512 / 128)
The images from the water mask are broken down into tiles that are either 512x512
pixels or 128x128 pixels in dimension. Tiles are beneficial in that every image that is
processed is uniform in size, which improves data handling and learning in machine
learning algorithms.
6. Dataset for Machine Learning
The tiles generated are classified based on a formatted data set, which consists of both
water samples and non-water samples. The data set is used as input in the training of
the machine-learning algorithm.
7. Training Models
The trained machine learning model will then utilize the prepared dataset to enable
detection of water features as well as the change in the extent of the surface waters
over time.
SOFTWARE AND HARDWARE REQUIREMENTS

The main tools used for the development of this project include the Python
programming language, along with several image processing, geospatial, and machine
learning libraries. The system processes satellite images to compute the Normalized
Difference Water Index (NDWI) for detecting surface water bodies and generates
binary water masks to analyze surface water dynamics under climate variation.

The implementation is carried out using Jupyter Notebook for interactive

development, visualization, and analysis of satellite-derived water features.

The following are the hardware and software requirements for the project
implementation

SOFTWARE REQUIREMENTS

• Operating System: Windows 10 or Linux (Ubuntu 20.04+)

Supports the development and execution of Python scripts and geospatial processing
tools.

• Programming Language: Python 3.8 or higher

Used for satellite image processing, NDWI computation, water mask generation, and
data preparation.

• Libraries and Packages: NumPy, Pandas, OpenCV, Rasterio, GDAL, Matplotlib

Enable numerical computation, satellite image handling, raster data processing, and
visualization of NDWI and water masks.

• Machine Learning Libraries (Optional): Scikit-learn, TensorFlow, PyTorch

Used for future extensions such as automated water segmentation and prediction.

• Development Environment: Jupyter Notebook / VS Code

Provides an interactive environment for coding, testing, and result visualization.

• GIS Software: QGIS

Used for spatial visualization, clipping of satellite images, and analysis of surface
water extent.

HARDWARE REQUIREMENTS
• Processor: Intel i5 or equivalent (Intel i7 recommended)
Provides sufficient speed for processing large satellite images and NDWI computation.

• RAM: Minimum 8 GB (16 GB recommended)

Ensures smooth processing of high-resolution satellite imagery and image tiling
operations.

• Storage: At least 10 GB of free space (SSD preferred)

Required for storing satellite images, processed outputs, tiles, and project files.

• Graphics: Optional GPU (NVIDIA)

Can accelerate future deep learning–based water segmentation and analysis.

• Internet Connection: Required

Used for downloading satellite data, Python libraries, and accessing online data
repositories.

1.9 ORGANIZATION OF THE PROJECT

The project has been divided into various chapters

Chapter 1: Introduction
The introduction of Chapter 1 is about the use of satellite data for monitoring surface
water dynamics. The significance of surface water monitoring is described in the
context of climate variation and environmental management. This chapter covers the
background of the study, the existing system, the challenges in the existing system,
the problem statement, the proposed system, the objectives, an overview of the
methodology, and the software and hardware requirements. The study area and the
scope of the project are also described in this chapter.

In chapter 2, there is a detailed literature review about surface water monitoring

methods and remote sensing that are hence relevant to each other. Chapter 2
elaborates on several methods to determine water such as water detection by spectral
index, GIS analysis, unsupervised and supervised learning algorithms, and deep
learning algorithms. It also throws light on the pros and cons of the present methods
and identifies the research gap resulting in this study.

Chapter 3 lays out a methodology and system architecture for the monitoring of
changes in surface water. This chapter narrates how the work was carried out
revolving around the subject matter, the system architecture put to work for
monitoring the changes of surface water. It tells how the author obtained satellite
images, the preprocessing steps taken, the choice of spectral bands, and the derivation
of the Normalized Difference Water Index (NDWI). Also, it depicts how the author
got water masks and image tiles ready for the analysis. Besides that, the entire flow of
the proposed system has been succinctly illustrated.

In Chapter4 is mainly focusing on the component parts of the system which have
been a point of discussion. Chapter 4 is mainly about the implementation of the
system. It informs about the datasets used, the computer programs and technologies
employed, and the execution of NDWI calculation and water mask generation on a
piece of paper. The chapter also covers the discussion of the output visualization
technique and the preparation of the result for a time, based review.

Chapter 5 illustrates the obtained results and discusses them attributing each to the
corresponding facets of the system. It illustrates the NDWI images, the binary water
masks, the visual outputs which are overlaid, and the comparative analysis of the
water surface extent of different years. The study, being a point of discussion, is based
on the observed data of surface water changes in relation to the variations in climate.

Chapter 6 wraps up the project by revisiting the main research outcomes and the
contributions that the study has made. It underscores the success of the NDWI, based
method in revealing the dynamics of surface water in relation to climate changes.
Besides, the chapter suggests the scope of the next developments, for example,
combining the current system with climatic parameters like rainfall and temperature,
making use of deep learning models for further precision, and broadening the system
to multi, season and multi, location analysis.

Overall, such a thematic format allows for a natural and coherent flow of the project
starting with the introduction and the problem statement, then moving through the
methodology and implementation, and finishing with the results and conclusion. It
gives a thorough insight into the purpose, methods, and achievements of the project.

[Link] SURVEY

2.1 Introduction to Surface Water Monitoring

Surface water monitoring is among the major applications of remote sensing and
Geographic Information Systems (GIS). It entails the identification, recording, and
evaluation of water bodies such as rivers, lakes, reservoirs, and wetlands through
satellite images. Surface water is an essential factor for several purposes including
agriculture, provision of drinking water, flood control, maintaining the balance of the
ecosystem, and climate regulation.

Climate change, in particular irregular rainfall, rise in temperature, droughts, and

occurrence of extreme weather events, has greatly impacted the availability and
distribution of surface water. Moreover, conventional field, based methods of
monitoring are very laborious, expensive, and confined to a small area. Remote
sensing, thus provides an effective means for regular and wide, area surface water
monitoring, due to recent developments in satellite technology and image processing
methods.

This chapter compiles a comprehensive literature review on the topic of surface water
detection and monitoring. It also discusses various methodologies that have been
employed in the studies, their advantages, drawbacks, and areas that need further
research.

2.2 Spectral Index-Based Water Detection (McFeeters, 2019)

McFeeters (2019) proposed the Normalized Difference Water Index (NDWI) to

highlight water bodies from satellite images. NDWI makes use of the Green and Near,
Infrared (NIR) bands, as water bodies are known to strongly absorb NIR radiation.
This approach can considerably diminish the presence of vegetation and soil while at
the same time pointing out the water areas.

The research revealed that the NDWI can be used for a wide range of medium,
resolution satellite images, e.g., Landsat and Sentinel, 2. On the other hand, NDWI
might sometimes incorrectly identify shadows and built, up areas as water.

Strengths: The formula is straightforward, water can be enhanced effectively, this

method is popular.

Weaknesses: It is easily influenced by shadows and the selection of the threshold.

2.3 Satellite-Based Surface Water Mapping (Pekel et al., 2020)

Pekel et al. (2020) carried out a detailed research at the global scale of surface water
changes with multi, temporal satellite images. They examined variations in water
bodies over extended periods thereby recognizing seasonal water changes and
permanent water bodies affected by climatic variations.

The investigation emphasized the value of seasonal studies in detecting phenomena

like water retreat, extension, and vanishing. Yet, operating with satellite data over an
extended period needs considerably high computational power and reliable methods
of cloud detection to be development.

Strengths:
Study of long, term trends, effects of climate on water bodies Weaknesses:
High cost of processing, problem of cloud contamination.

2.4 GIS-Based Surface Water Analysis (Goodchild et al., 2021)

Goodchild et al. (2021) highlighted the importance of combining GIS and remote
sensing in the study of water surfaces. Through GIS, spatial manipulations can be
carried out, for example, the calculation of the area of water bodies can be done easily,
the analysis of overlay between land and water for different time periods, and
detecting the changes of water bodies between different time periods. A GIS, based
study is good for spatial interpretation and visualization, however, it is still very much
dependent on the reliability of the water masks that have been extracted. In most cases,
a human has to do some corrections, thus the work cannot be fully automated. The
main advantage: Huge capacity for spatial analysis and very good for visualization.
The main disadvantage: Only semi, automated, dependent on the quality of the input
data.

2.5 Unsupervised Classification Techniques for Water Detection (Zhang et al.,

2022)

One of the works that Zhang et al. (2022) investigates are unsupervised methods like
KMeans clustering to delineate water bodies. These methods classify image pixels
exclusively according to their spectral features without needing any pre, labeled
training samples and hence can be applied particularly in regions where data is scarce.

On the other hand, in complex environments, unsupervised techniques are prone to

splitting a single object into several segments and to produce inaccurate classification,
notably when dealing with mixed landwater pixels.

Strengths: No labeled data required, computationally efficient.

Weaknesses: Over, segmentation, parameter sensitivity.
2.6 Deep Learning for Surface Water Segmentation (Ronneberger et al., 2021)

Ronneberger et al. (2021) experimented with deep learning architectures to segment

water bodies at high precision, from remote sensing images, the authors U, Net model
to accurately delineate the water boundaries. These models, by learning the spatial
patterns deeply, are able to delineate the accurate boundaries even in such difficult
environments where there is cpnfilct between the water and the other land features.

Deep learning methods have shown great accuracy but still require well, annotated
large datasets and computing power, however, for small academic projects, these
requirements may not be met.

Pros: Accurate detection of water boundaries.

2.7 Climate Change and Surface Water Dynamics (IPCC, 2023)

According to the Intergovernmental Panel on Climate Change (IPCC, 2023), climate

change in its different expressions affects the surface water systems of the whole
world drastically. Various changes to the water cycle such as the shifting rain patterns,
the rise in evaporation, and the occurrence of severe weather have all resulted in the
regular surface water changes.

The panel recommended that water surface monitoring should be done regularly
through satellites to help in managing water resources efficiently and implementation
of the climate adaptation measures.

Pros: Good understanding of climate impacts, globally applicable.

Cons: Needs to be combined with local, scale data for more detailed analysis.

2.8 Research Gap and Motivation

The literature review shows that even though deep learning and other advanced
techniques can give very accurate results, they need huge labeled datasets and
powerful computers. In contrast, basic spectral indexbased methods are quick but
need very precise setting of the thresholds.

This work covers the gap mentioned above by using a straightforward and
inexpensive NDWI, based method coupled with automated processing, hence it can
be used for monitoring changes in surface water under the influence of climate by
means of medium, resolution satellite data. The method is fit for both, the higher
precision without compromising the level of computation and thus it allows to look at
changes of water bodies over time.

[Link]-BASED SURFACE WATER DYNAMICS MONITORING METHOD

3.1 Architecture of the System

3.1 Architecture of the System

The automatic surface water dynamics monitoring system under climate variation
through satellite imagery is the main idea of the proposed system. Remote sensing,
based water indices, image pre-processing techniques, and machine learning
preparation steps are the main components of the system that together give the
accurate and efficient extraction of surface water bodies. The system architecture is
explained below step, by, step, each part reflecting a stage shown in the architectural
flow.

1. Input Layer (Acquisition of Satellite Images)

Satellite images are first obtained of the area of interest, for instance, using Sentinel, 2,
after which the process satellite image retrieval those images furnish multispectral
bands, with a particular focus on the Green and Near, Infrared (NIR) bands that are
instrumental in mapping surface water bodies such as lakes, reservoirs and rivers. ...
The input layer is the raw image data, from which the quality of the images is set up
for the next analysis phases. Pre-processing stage is a short series of steps:

NDWI Computation:
Once the raw satellite images have been retrieved, the Normalized Difference Water
Index (NDWI) is calculated from them in order to identify water and non, water areas.
Since NDWI operates to emphasize water areas and to de-emphasize vegetation and
soil areas, thus it can be used for the identification of water bodies very effectively

Image Resizing and Normalization:

Since satellite images come in different sizes and resolutions, the images are resized
to a predetermined size, e.g., 256 256 or 512 512 pixels. The pixel values are also
normalized to a fixed range (i.e. 0 to 1), which is a prerequisite for machine learning
applications and helps in standardizing the data representation.
The satellite image is compiled so that all the necessary bands (e.g., Green and NIR)
become available for the next phase of processing.

Input: Raw satellite images from Sentinel, 2.

Output: Unprocessed image data for use in the following step.

2. Preprocessing Layer Data Preparation

Different preprocessing operations, like image clipping to the study area,

normalization, and water mask generation, are done in order to strengthen the system
and increase its robustness. These operations lead to more consistent data and better
surface water detection quality even under varying climatic conditions.

Input: Raw satellite images.

Output: Normalized and preprocessed images, prepared for surface water analysis.

3.2 Modules Description

The system is made up of the following modules:
3.2.1 Data Acquisition Module

This module handles the retrieval of Sentinel, 2 satellite images for various time
periods. The photos are taken in a way that the necessary spectral bands for water
detection are present.

Input: Raw satellite images

Output: Satellite images ready for preprocessing

3.2.2 Image Preprocessing Module

This module is mainly concerned with preparing the acquired satellite images visually
and technically for the next step of the experiment. The preprocessing comprises
several steps such as trimming the image to the area of study (Upper Lake), selecting
the bands, resizing, and normalization. These are essential for the extraction of the
surface water to be correct.

Input: Raw satellite images

Output: Preprocessed and normalized images

3.2.3 NDWI Computation Module

This unit calculates the Normalized Difference Water Index (NDWI) from the Green
and Near, Infrared bands of the satellite imagery. NDWI helps to highlight water
bodies on the surface by separating them from vegetation and bare land.

Input: Satellite images after preprocessing

Output: NDWI image

3.2.4 Water Mask Generation Module

A binary water mask is created here from the NDWI image by using thresholding
technique. Pixels are grouped into water and non, water classes which lead to a clear
isolation of surface water bodies.

Input: NDWI image

Output: Binary water mask

3.2.5 Tile Creation Module

In order to make the data more compatible with machine learning, the NDWI image
along with the water mask are split into smaller fixed, size tiles of 512 512 or 128 128
pixels, for example.

Input: NDWI image and water mask

Output: Image tiles and mask tiles

3.2.6 Dataset Preparation and Model Training Module

Here, image tiles along with the corresponding mask tiles are combined to create a
dataset that is ready for machine learning. Subsequently, this dataset serves as model
training data for the automated surface water detection and monitoring task.

Input: Imagemask tile pairs

Output: Trained surface water monitoring model

4. Implementation

4.1 Datasets

· The satellite images utilized for this research are Sentinel, 2 multispectral images
downloaded from the European Space Agency (ESA) Copernicus Open Access
Hub. The dataset consists of several years in order to observe surface water
changes due to climate variations and the geographical area covered is Upper
Lake. The spectral bands used for the purpose of water detection on surface are
· Band 3 (Green) 560 nm
· Band 8 (Near Infrared NIR) 842 nm

4.2 Tools & Technologies Used

· Index Calculation: NDWI (Normalized Difference Water Index)

· Programming Language: Python

· Libraries: NumPy, OpenCV, Rasterio, Matplotlib

· Platform: Jupyter Notebook / Google Colab

· Image Format: JP2 and TIFF formats from Sentinel-2 data

· Machine Learning (Optional): Deep learning models for water segmentation

4.3 Algorithm 1: NDWI ,based Surface Water Detection and Dynamics
Monitoring

This algorithm works on satellite images by calculating first NDWI, then water masks,
and finally creating datasets for monitoring surface water changes under climate
variation.

Step 1: NDWI Computation

Objective:

Derive the Normalized Difference Water Index (NDWI) for every pixel of the
image to distinguish water bodies on the surface.

Formula:

NDWI = (Green, NIR) / (Green + NIR)

Band Information:

Green: Band B03

NIR: Band B08

Process: Open the Green and NIR bands from Sentinel2 satellite images.

Calculate the NDWI for each pixel using the given equation.

Output:

An array of NDWI values between, 1 (non, water) and +1 (water).

Step 2: Water Mask Generation

• Objective:To separate water and non-water regions using NDWI values.

• Process:
o Apply thresholding to the NDWI image.

o Pixels with NDWI values above the threshold are classified as water.

o Remaining pixels are classified as non-water.

• Output:A binary water mask representing surface water regions.

Step 3: Tile Creation

Objective: Preparing data for a machine learning algorithm.

Process:

o Split the NDWI image and corresponding water mask into fixed-size tiles

(e.g., 512 × 512 or 128 × 128 pixels).

o Each tile represents a small portion of the study area.

Output: Tiles from both the image and the corresponding water mask.

Step 4: Dataset Preparation

• Objective:To create a machine-learning-ready dataset for surface water analysis.

• Process:
o Pair each NDWI tile with its corresponding water-mask tile.

o Store the paired data in efficient formats such as .npy.

• Output:A structured dataset suitable for training water-segmentation models.

Step 5: Visualization and Temporal Analysis Objective

To visually represent the extent of surface water and study its variation over different
time periods.

Process:

o Show NDWI images as well as water masks.

o Overlay water masks on the satellite images.

o Check surface water extent for different years.

o Create graphs to show changes in water area.

Output:

Visual outputs of surface water behavior under climate variation, including maps and
comparative graphs.
5. Analysis of Surface Water Dynamics

5.1 Performance Measures

Accuracy:Due to the lack of ground truth data, the assessment was done by a visual
comparison of NDWI images and the production of water masks with the known
surface water extents.

NDWI Validity:It was confirmed that water bodies have higher positive NDWI
values, and land and vegetation areas have lower or negative values, which confirms
the suitability of NDWI for detecting surface water.

Water Mask Quality:The water masks created were able to perfectly differentiate
between water and non, water areas with only a few errors.

Surface Water Area Estimation:The fraction of surface water was determined by

enumerating the number of water pixels in the binary mask.

5.2 Results for Upper Lake (Year-wise Analysis)

Surface Water Distribution (%):

Observation:

Significant changes in the extent of surface water from one year to another were
observed.

Changes in water area showed a decrease during years with less rainfall.

The surface water extent was found to increase in the years when the precipitation
was higher.
Year Water Non-Water
5.4 Visual Analysis based on NDWI

NDWI Distribution:

Observation:

Positive NDWI values mostly match areas of surface water.

Boundary areas have mixed NDWI values as there is vegetation growing along the
water edges.

The NDWI, based method is very successful in delineating the water surfaces that are
not covered by any object.

5.5 Comparative Analysis

Graphical analysis:

Graphs depicting changes in surface water over different years illustrate the situation
clearly.

Declining rainfall phases are shown with the shrinking of water area.

Water areas in the years with more precipitation generally recover or grow.

Insights:

Surface water changes depend a lot on climate changes.

It is very important to keep monitoring to get a hold of the main changes over a long
time.

Using satellite technology to carry out NDWI studies is an effective way of

monitoring water on a large scale.
6. Conclusions and Future Enhancements
Conclusions

The proposed project aimed at analyzing the dynamics of surface water influenced by
climate variability using satellite images and NDWI analysis. The project has been
able to successfully detect surface water features and analyze their changes using the
system developed.

The above results clearly indicate that remote sensing methods can be efficiently used
to monitor surface water resources. The paper emphasizes the effect of climate
variability on surface water resources and the need for continuous monitoring of
surface water resources using satellite images.

Future Improvements

To further enhance the system and make it more applicable, the following are
proposed:

• Integration of Climate Data:

The integration of rainfall, temperature, and evaporation data to improve the

correlation of climate variables with changes in surface water.

• Deep Learning-Based Segmentation:

The use of CNN-based models like U-Net for improved water boundary segmentation.

• Multi-Season Analysis:

The extension of the system to perform analysis based on seasonal changes that occur
in a year.

• High-Resolution Satellite Data:

The use of high-resolution satellite images to improve the detection of small and
narrow water bodies.

• Automated End-to-End Workflow:

The development of an automated system from image acquisition to visualization of

results.

• User-Friendly Interface:

The development of a web-based or GIS-based interface for easy access and

interpretation of results by water resource planners.
REFERENCES

1. McFeeters, S. K., “The Use of the Normalized Difference Water Index (NDWI)
in the Delineation of Open Water Features,” International Journal of Remote
Sensing, Vol. 17, pp. 1425–1432, 1996.
2. European Space Agency (ESA), Sentinel-2 Overview,
[Link]
3. Gonzalez, R. C. and Woods, R. E., Digital Image Processing, 4th Edition,
Pearson Education, 2018.
4. Jensen, J. R., Introductory Digital Image Processing: A Remote Sensing
Perspective, 4th Edition, Pearson Education, 2015.
5. Pekel, J.-F., et al., “High-Resolution Mapping of Global Surface Water and Its
Long-Term Changes,” Nature, Vol. 540, pp. 418–422, 2016.
6. Goodchild, M. F., et al., “GIS and Remote Sensing for Surface Water
Monitoring,” International Journal of Geographical Information Science, Vol. 35,
pp. 1–15, 2021.
7. Zhang, Y., Li, X., and Chen, Y., “Unsupervised Surface Water Extraction from
Sentinel-2 Imagery Using Spectral Indices,” ISPRS Journal of Photogrammetry
and Remote Sensing, Vol. 178, pp. 23–35, 2021.
8. Jain, A. K., Fundamentals of Digital Image Processing, 1st Edition, Prentice Hall,
1989.
9. Ronneberger, O., Fischer, P., and Brox, T., “U-Net: Convolutional Networks for
Biomedical Image Segmentation,” IEEE Conference on Computer Vision and
Pattern Recognition (CVPR), pp. 234–241, 2015.
10. IPCC, Climate Change 2023: Impacts, Adaptation, and Vulnerability,
Intergovernmental Panel on Climate Change, Cambridge University Press, 2023.
GLOSSARY

C
Climate Variation – Long-term changes in climatic factors such as rainfall,
temperature, and evaporation that influence surface water availability.

D
Dataset – A structured collection of satellite images, NDWI outputs, and water masks
used for analysis and model training.

F
Flood Event – A condition where surface water spreads beyond its normal boundaries
due to excessive rainfall or runoff.

I
Image Patch – A small cropped section of a satellite image used for visualization and
machine-learning processing.
Index (NDWI) – A water index used to highlight surface water bodies using satellite
data.

M
Mask (Water Mask) – A binary image where pixels represent either water or non-
water regions.
Model – A trained machine learning or deep learning algorithm used to detect surface
water patterns.

N
NDWI – Normalized Difference Water Index, used to enhance and extract surface
water features from satellite imagery.

P
Processing – Image operations such as NDWI computation, thresholding, tiling, and
normalization applied to satellite data.

R
Remote Sensing – The science of collecting information about the Earth’s surface
using satellite sensors.
Render – To generate and display processed images or analytical outputs.

S
Satellite Image – An image of the Earth captured by a satellite sensor.
Sentinel-2 – A multispectral Earth observation satellite mission by ESA used for
surface water monitoring.

T
Temporal Analysis – Analysis of changes in surface water extent over different time
periods.
TIFF (.tif) – A geospatial image file format used for storing satellite data.
U
Upload Folder – Location where satellite images or datasets are stored before
processing

APPENDIX
Programming Language:
Python ,Python Libraries and Frameworks:
NumPy, Pandas (data handling)
OpenCV, scikit-image (image processing)
Rasterio, GDAL (geospatial data processing)
TensorFlow/Keras or PyTorch (deep learning – optional)
scikit-learn (machine learning – optional)

GIS Tools:
QGIS for visualization and spatial analysis

Web Framework :Flask

Satellite Image Sources:Source: Sentinel-2

Image Format:GeoTIFF / JP2

Spatial Resolution:10 m

Timeframe: 2021–2025