Simulating riverflow in South Africa’s Limpopo River Basin using TerM land surface model

Accurate riverflow simulation is essential for water resource management, flood and drought prediction, and understanding the hydrological response to climate change. This study presents the first application of the INM RAS-MSU (Terrestrial Model) land surface model in South Africa, assessing its performance in the Limpopo River Basin, which differs significantly from the cold-region environment for which it was originally developed. The study evaluated the capability of ERA5 reanalysis data to replicate observed station rainfall from 1994 to 2020 and its influence on model performance. Model simulations were conducted for the period 1991–2020 using ERA5 hourly and daily forcing data. Simulations were compared with observed riverflow from three-gauge stations and the GLOFAS database using PBIAS and NSE metrics. Results reveal that ERA5 overestimates rainfall, particularly during the austral summer, with the overestimation most pronounced in January. While the model captures seasonal variability well, manual calibration significantly improved its performance, yielding NSE values above 0.8 and PBIAS below 8.3% across all riverflow stations when forced with hourly data. Temporal resolution strongly influences model accuracy, with hourly forcing generally overestimating and daily forcing underestimating riverflow compared with observations. Furthermore, the spatial resolution and quality of the data for both forcing and observations affected model performance. These findings highlight the importance of model evaluation and data quality assessment to improve hydrological predictions and support early warning systems for climate-induced disasters.