Understanding Spatiotemporal Dynamics of Urban Water Bodies via Multi-Classifier Machine Learning Approaches in Semi-Arid Ahmedabad City

Ankit Deshmukh , Shreekant Varshney

The research examines the spatiotemporal change of water bodies in Ahmedabad city in western part of India which is a rapidly urbanizing area. It integrates high-resolution remote sensing data with modern machine learning approaches. Understanding the long-term pattern is crucial to understanding and managing the water resources in the changing climate, also city experiences the low precipitation with long summer months that makes managing water resources more challenging. The study utilizes the Landsat-8 imagery from 2013 to 2024 to indices that highlights the water signatures. Normalized Difference Water Index (NDWI) and Modified Normalized Difference Water Index (MNDWI) are computed using green, near infrared and mid-infrared Landsat bands. Machine learning classifier algorithms are used to identify the water area and non-water area pixel. The selected algorithms are classification and regression trees (CART), minimum distance (MD), naïve bias (NB) and random forest (RF) for that training data points are collected with visual inspection of available satellite images and manual in-situ inspection. Train and testing data is split in 80-20 to train and validate the classification of water area. A time series are generated for 12 years in both monthly and yearly scale. The primary objective is to assess the water area changes across the past decade, changing water area due to the anthropogenic impacts and climatic factors effect on it due to abrupt changing climate. Furthermore, the study attempted to provide a systematic approach to analysis the water area changes and patter identification that will be useful for policy making on water resource management for the city. The study demonstrates the essential requirement for data-driven governance of water resources, providing insights into sustainable urban development and environmental conservation strategies aligned with Sustainable Development Goals (SDGs) 6 and 11, which focus on clean water and urban sustainability. Based on the analysis of the water area time series a slight upward trend is evident with both yearly and monthly tread with Mann-Kendall test across all the classifiers. Temperature has significantly driver of water change and precipitation has low impact in the long-term water area change. CART and Naïve Bayes showed better performance, with CART being linearly sensitive to temperature and others exhibiting non-linear precipitation dependencies. The deployed machine learning models underscoring the effectiveness of combining remote sensing with machine learning. The research enhances hydrological monitoring and ongoing environmental evaluations by adeptly combining traditional remote sensing methods with modern computational methodologies.