Mangrove ecosystems are among the most productive coastal environments and play a significant role in carbon sequestration, shoreline protection, biodiversity conservation, and climate change mitigation. The Sundarbans, the world’s largest mangrove forest, are ecologically sensitive and highly vulnerable to environmental degradation, sea-level rise, and anthropogenic disturbances. Accurate mangrove mapping and biomass assessment are therefore essential for sustainable coastal management and blue carbon estimation. This study developed an integrated geospatial and deep learning-based framework for mangrove mapping, Above Ground Biomass (AGB) estimation, carbon stock assessment, and carbon sequestration analysis within the Indian Sundarbans mangrove ecosystem.

The research utilized Sentinel-1 SAR imagery, Sentinel-2 multispectral imagery, vegetation and moisture indices including NDVI, NDWI, MNDWI, NDMI, SAVI, and Mangrove Vegetation Index (MVI), along with Random Forest supervised classification and U-Net deep learning segmentation techniques. The Random Forest classification generated an initial mangrove distribution map, while the U-Net model provided refined mangrove segmentation with improved spatial continuity and boundary delineation. Field-based survey observations and regression modelling approaches were integrated for biomass estimation and model validation.

The Above Ground Biomass map revealed substantial spatial variability in biomass distribution across the study area, with dense mature mangrove forests concentrated mainly in the southern and southeastern tidal regions exhibiting very high biomass values. The estimated total biomass of the study area was approximately 2,416,438 Megagrams (Mg). Carbon stock estimation demonstrated that the Sundarbans function as a major blue carbon reservoir with an estimated total carbon storage of approximately 1,135,726 Mg Carbon, equivalent to nearly 1.14 million tonnes of stored carbon. The Random Forest regression model achieved strong predictive performance with an R² value of 0.890, indicating high agreement between observed and predicted biomass values.

The findings highlight the effectiveness of integrating remote sensing, machine learning, deep learning, and field survey data for mangrove ecosystem assessment within complex tidal environments. The study provides valuable insights for blue carbon monitoring, climate change mitigation, mangrove conservation, and sustainable coastal ecosystem management in the Sundarbans region.

 

Keywords: Sundarbans, Mangrove Mapping, U-Net, Random Forest, Biomass Estimation, Carbon Stock, Blue Carbon, Sentinel-1, Sentinel-2, Remote Sensing, Deep Learning, Carbon Sequestration.