Integrative deep learning framework for analysing population density shifts and flood inundation patterns – a case study of the Sundarbans

ABSTRACT This study investigates the interplay between climate-induced environmental changes and human migration in the Sundarbans, a globally significant mangrove ecosystem spanning India and Bangladesh. Utilizing high-resolution satellite imagery from Sentinel-1 and Sentinel-2 missions, combined with geospatial datasets, we employ advanced deep learning models to map and analyze population density shifts and flood inundation patterns from 2016 to 2023. A dual-encoder UNet architecture is validated for precise flood mapping, while a multi-encoder deep learning model is developed for accurate population density estimation. Our results reveal a strong correlation between increased flood frequency and population displacement, with areas experiencing severe flooding showing a 15-20% decline in population density, while safer zones saw a 10-15% increase. The proposed approach outperforms traditional methods by integrating SAR and multispectral data, achieving a flood detection accuracy of 92% (IoU) and population density classification accuracy of 88% (F1-score). These findings underscore the need for integrated policy frameworks to enhance resilience and sustainable development in flood-prone regions. The study's methodological advancements and quantitative insights provide a robust foundation for future research and policy formulation in similar vulnerable ecosystems worldwide.