Abstract
Soil salinity is a growing concern for crop productivity and ecosystem resilience, particularly in coastal regions like the Sundarbans mangrove ecosystem, where salinization has worsened since the 1980s due to climate change and diminished freshwater flow. Nonetheless, previous studies have often relied on limited in situ data or remote sensing, lacking comprehensive integrated approaches. This study aims to fill that gap by utilizing the Harmonized World Soil Database combined with approximately 100 in situ salinity measurements from 2022 to create regional and global soil salinity maps. We introduce an innovative machine learning framework that combines satellite-derived time-series imagery, detailed soil characteristics maps, and various ensemble models, including Random Forest, Artificial Neural Network, and a hybrid LASSO–GA–BPNN model. This ensemble approach effectively captures seasonal variability and attains high prediction accuracy (90%–99%) across agricultural and industrial zones throughout all three seasons. The findings facilitate targeted land-use planning, salinity-resilient crop management, and the prioritization of restoration efforts in degraded coastal areas. By providing dynamic predictions and addressing the physiological stress that tropical mangroves endure under hypersaline conditions, our study not only showcases advanced modeling techniques for salinity prediction but also emphasizes the necessity for implementing nature-based restoration strategies to promote sustainable land use and biodiversity conservation in estuarine ecosystems.
