Flood Prediction in Hilly Terrains: A Review of Models, Challenges, and Advancements

Flood forecast in hilly/mountain terrains poses unique challenges because of steep gradients, sparse hydro-meteorological networks, short-duration localized cloudbursts, and glacial lake outburst floods (GLOFs). To serve this purpose, the paper offers a terrain-specific overview of flood forecasting models, structured in methodo-geographic terms; it does so by reviewing more than 24 studies over the past years by five methodological traditions: physically-based hydrological models, machine learning (ML) models, deep learning (DL) models, hybrid ML-hydrology frameworks, and remote sensing in GIS-based models. We demonstrate that classical models like SWAT or GR4J have difficulties dealing with intense topography and have relatively little calibrating data, while models like LSTM–SWAT, XGBoost–MIKE FLOOD and Transformer–LSTM by integrating elevation-aware inputs and high-resolution satellite data outperform the others in terms of predictive accuracy. Comparative study showed that other terrain attributes (slope, drainage density, snowmelt profiles) and remote sensing data (DEM, NDVI, SAR) are indispensable for accurate prediction in ungauged mountainous catchments. To the best of our knowledge, this study is the first to systematically assess the ability, generalizability, and domain-responsiveness of models for hilly terrain flood prediction—connecting hydrology, machine learning, and remote sensing together by an integrated benchmark suite. The review also identifies main gaps in research and suggests some terrain-adapted trajectories for future model development for complex mountainous terrain.