Abstract
This study presents an integrated geospatial–InSAR–machine learning framework for analyzing land subsidence, assessing flood vulnerability, and projecting future relative sea-level rise (RSLR) in the Indian Sundarbans. By combining multi-temporal Sentinel-1 InSAR deformation data, GRACE/GRACE-FO hydrological anomalies, and geomorphic–hydroclimatic predictors with supervised learning models—including Artificial Neural Networks (ANN), Support Vector Machines (SVM), Linear Regression (LR), and Ensemble methods—we evaluated subsidence dynamics and predicted future trends. Model validation applied standard accuracy metrics (R2, RMSE, MAE, MSE), with LR and Ensemble models showcasing exceptional predictive performance (R2 = 0.86–0.99). Utilizing a correlation matrix and multicollinearity screening for variable selection demonstrated strong relationships among subsidence, tidal patterns, bathymetry, rainfall, land-use changes, and sea-level fluctuations. The autoregressive integrated moving average (ARIMA) model forecast predicts that current deformation rates exceeding 2 mm yr−1 in multiple areas are expected to escalate to around 8.33 mm yr−1 by 2100. This acceleration, coupled with anticipated sea-level rise (SLR), significantly exacerbates RSLR, highlighting a robust statistical relationship between subsidence and flood depth (R2 = 0.77–0.85). Comparative flood mapping from 2015 to 2024 reveals an ongoing expansion of inundation due to ground subsidence and tidal effects. Findings emphasize the urgency for coordinated adaptive strategies, such as sustainable groundwater management, mangrove restoration, and the creation of ecosystem-based flood defences. By merging relevant predictors with cutting-edge ML techniques, this study presents a transferable framework for investigating subsidence–flood interactions in deltaic environments and reinforces the alignment of policies with SDGs 11, 13, and 15.