Salma Hoque Talukdar Koli

Salma Hoque Talukdar Koli is a researcher working at the intersection of machine learning and hydrology, with a focus on flood early warning in river basins where ground-based instrumentation is sparse or absent. Her work combines satellite remote sensing, particularly Sentinel-1 SAR backscatter, with ensemble and deep learning methods to build forecasting systems that can operate without dense gauge networks.

Her main project, HaorFloodAlert, is a 72-hour flash flood early warning system for the Sunamganj haor wetlands of northeast Bangladesh, a basin of roughly 8,000 square kilometres fed partly by an ungauged transboundary river. It combines SAR backscatter, satellite rainfall, soil moisture and modeled upstream discharge in a deseasonalized Random Forest and XGBoost ensemble, validated on 77 radar-verified events and tested in a live ten-day prospective trial. The system delivers Bengali-language warnings together with an estimate of Boro rice damage, so that alerts reach farmers in the terms they actually use.

She also works on explainable AI for humanitarian and agricultural applications, including multimodal trauma screening and crop disease classification. Her current interests are hybrid process-based and machine learning hydrology, cross-basin model transfer, and predicting how long floodwater persists on a floodplain rather than only when it arrives.