Evaluating Water-Driven Ecosystem Sustainability Using Machine Learning Assisted GeoAI and Remote Sensing: Evidence from Rajbari, Magura, and Kushtia Districts

Author: Shahina Akter, Khan Mohammad Ibtehal, K M Azam Chowdhury, Md Yeamun Hasan Soumick, Mohammad Azharul Islam, Muntasir Faisal
DOI: doi.org/10.70279/bmj-v1-1100

Bangladesh, where floodplain landscapes integrate growing urbanization and environ mental vulnerability, monitoring Land Use and Land Cover (LULC) trends is crucial. This study uses multi-temporal Landsat 8 images from 2015, 2020, and 2024 to examine spatio temporal LULC variations in the southwestern districts of Kushtia, Rajbari, and Magura, also using Geospatial AI (GeoAI). Land alteration patterns within the Ganges floodplain were quantified using supervised classification based on a Random Forest algorithm in Google Earth Engine and ArcGIS. Four LULC classes waterbodies, vegetation, built-up areas, and barren land were analyzed to determine urban expansion, land degradation, and ecological change. The outcomes reveal significant differences between districts. Kushtia had substantial urban expansion, with built-up areas expanding from 12.48% in 2015 to 29.73% in 2024, mostly displacing vegetated and barren land. Increased imperme able surfaces and possible hydrological disruptions through modified runoff and decreased infiltration are suggested by the expansion's coincidence with decreasing plant cover and slight decreases in surface water extent. With implications for hydrological instability, Rajbari showed a persistent decrease in vegetation and a nearly threefold increase in barren land, indicating ecological degradation probably caused by riverbank erosion, land abandonment, and climate stress.Whereas Magura displayed a more sustainable trend, highlighted by increased plant cover and a large reduction in bare terrain alongside signif icant urban growth. Although localized changes show the susceptibility of peripheral aquatic systems, waterbodies in all districts remained largely constant. Classification accuracy was high, with total accuracy surpassing 87% and Kappa coefficients above 0.85. To reduce hydrological disruption and improve landscape resilience, the results high light the necessity of district-specific land-use and water-sensitive design techniques.