Assessment of Gridded Precipitation Products for Hydrological Modelling in Semi-Arid Catchments of the City of Mbombela, South Africa

Introduction: Reliable precipitation information is critical in hydrological modelling, flood forecasting and water resources management. However, rainfall observation networks are sparsely distributed, especially in developing countries, limiting the spatial representativeness of rain gauge networks. Gridded precipitation products offer an attractive alternative to traditional ground observations, but their performance should be evaluated before application in hydrological studies. The study evaluated the performance of CHIRPS Version 2.0, ERA5-Land and IMERG version 07 in representing the Mean Annual Precipitation (MAP) of selected semi-arid quaternary catchments in the City of Mbombela, South Africa.

Methods: Long-term MAP was extracted for eight South African Weather Service (SAWS) rainfall stations using Google Earth Engine. Performance of the satellite and reanalysis datasets were evaluated against the observed precipitation values of the SAWS stations using coefficient of determination (R²), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Bias Error (MBE), Nash–Sutcliffe Efficiency (NSE) and Percent Bias (PBIAS).

Results: The three datasets captured the spatial rainfall gradient of the study area with higher precipitation amounts recorded in the western escarpment and lower values in the eastern Lowveld. ERA5-Land recorded the highest agreement with the observed values (R² = 0.90), the lowest RMSE (111.35 mm) and the lowest MBE (83.47 mm) and PBIAS (12.02%) of the three datasets. CHIRPS represented the spatial distribution of rainfall well but produced the lowest statistical performance (R² = 0.65). The dataset however has the longest historical period of record and the finest spatial resolution making it valuable for long-term hydrological applications.

Conclusion: ERA5 land had the best agreement with the observed values while IMERG had the least bias. CHIRPS is a good rainfall dataset due to its temporal extent and spatial resolution. The three datasets are recommended for hydrological modelling in the City of Mbombela after local validation and correction of any systematic errors.