| Sumario: | River salinity has been a critical factor influencing agricultural productivity in the polder systems of coastal Bangladesh, where sluice gates are used to regulate water inflows. Timely sluice gate operations during the early and late monsoon periods are essential for the successful rice cultivation and subsequent dry season crops to get high yields. Farmers, agricultural extension agents, and water management groups (WMGs) require timely and location specific salinity information to make informed decisions on water management. Existing hydrological forecasts in the region primarily focus on climate extremes, floods and droughts, with limited operational tools for river salinity forecasting. To address this gap, an AI-based river salinity advisory system has been developed and piloted in two polders in the Khulna region. The system forecasts temporal windows during which river salinity remains within acceptable thresholds for agricultural water use, particularly at the start and end of the monsoon seasons. Given the limited availability of long-term salinity observations, the forecasting framework uses river discharge in the Rupsa River as a proxy. A multivariate Long Short-Term Memory (LSTM) and Random Forest models were trained using historical salinity and discharge data from the Rupsa River, CHIRPS rainfall estimates, and upstream discharge measurements at the Hardinge Bridge on the Ganges River to estimate salinity windows at key polder intake points. The salinity advisory for farmers issued for July 2025 forecasted the lowering of the salinity window between the 9th and 14th of July 2025. Validation of forecasts issued in July 2025 for Polders 34/2P and 30 showed strong performance, with Nash–Sutcliffe Efficiency (NSE) values exceeding 0.80 and per cent Bias (PBIAS) under 10% for the forecasted salinity windows. The system has been developed in collaboration with national research institutions, local government agencies, and WMGs, and is accessible through a user-friendly dashboard and mobile interface. This initiative demonstrates the potential of AI to enhance irrigation advisory services in climate vulnerable coastal regions. The approach offers a scalable model for integrating machine learning into water resource management in other data-scarce deltaic settings.
|
|---|