Spatiotemporal Land Use and Land Cover Change Analysis (2005–2025) and Future Projections to 2040 in Prayagraj, India, Using Machine Learning and Google Earth Engine

Abstract

Rapid urbanization has significantly transformed land use and land cover (LULC) patterns in Indian cities, creating a need for reliable spatial information to support urban planning and sustainable land management. This study investigates LULC dynamics in Prayagraj, Uttar Pradesh, India, from 2005 to 2025 and projects potential changes for 2030, 2035, and 2040 using multi-temporal Landsat imagery and Google Earth Engine (GEE). Four LULC classes: water bodies, vegetation, built-up land, and barren land were mapped using three supervised machine learning algorithms: Random Forest (RF), Support Vector Machine (SVM), and Classification and Regression Trees (CART). Classification performance was evaluated using overall accuracy (OA) and the Kappa coefficient. Among the three classifiers, SVM demonstrated the most consistent performance, achieving a mean OA of approximately 0.93 across the five reference years. The historical analysis revealed substantial landscape transformation, particularly the expansion of built-up areas and the overall decline in vegetation cover. The RF-based prediction model was evaluated by comparing a simulated LULC map for 2025 with the corresponding classified map derived from satellite imagery. Although the model reproduced several broad spatial patterns, it overestimated built-up and barren land in some areas, indicating uncertainty in the projected results. The future projections suggest increasing built-up land and declining vegetation between 2030 and 2040, while water bodies and barren land exhibit more variable patterns. These findings demonstrate the potential of integrating satellite remote sensing, machine learning, and cloud-based geospatial processing for long-term LULC assessment. However, the projected changes should be interpreted cautiously because of classification uncertainty, spatial-resolution constraints, and limitations in representing the complex processes governing urban expansion. Further validation and comparison with alternative spatial prediction models are necessary to improve the reliability of future LULC scenarios.