Near-real-time flood modelling in the upper Roshi River catchment in Nepal

Abstract

Annually, the Upper Roshi River Catchment (URRC) in Nepal experiences flooding during the monsoon season due to extreme precipitation events. Yet, data scarcity has hindered reliable flood modeling in the URRC. This study addresses this data gap through fieldwork installing Automatic Weather Stations (AWS), Soil moisture sensors, and a Hydrological Station (HS), enabling a quantitative analysis of precipitation, soil moisture, and water-level interactions. The Stations data, in combination with additional infiltration and cross-sectional measurements, were used to create basemaps to build a flood hazard model in FastFlood. Data analysis revealed lag times ranging from 55 to 255 minutes, depending on the AWS location and the river's morphology. Soil moisture analyses also indicated strong spatial variability in infiltration, with low-permeability zones near the upper Roshi River increasing runoff risk and underscoring the importance of infiltration input in the Flood hazard model. Model results accurately simulate the extreme October 2025 event, with the simulated water levels matching the observed water levels at the HS. The September 2024 event water level at the HS was underestimated by 1.7 meters, largely due to spatiotemporal gaps in the event's precipitation data. Also, the rainfall descriptors (Intensity/Duration/Shape) analyses indicated that precipitation intensity is the dominant driver for flood effects in the URRC. In the end, a flood hazard model was developed, which delivers reliable flood hazard maps for inundation, maximum velocities, and maximum discharges. These maps can be used for a full flood risk assessment of the URRC, which, in the future, could help form the basis of an early warning system for the catchment.