Data scarcity has been a huge problem in modelling various catchments especially in the tropical region. Satellite data and different statistical methods are being used to improve the quality of conventional meteorological data. However, the potential of using these data needs to be further investigated. This paper evaluates the performance of three reanalysis datasets in hydrological modelling of the Manafwa Catchment, Uganda. Two reanalysis datasets were selected for studying both rainfall and temperature in Manafwa, whereas a satellite algorithm was selected for studying rainfall alone. The Modern-Era Retrospective Analysis for Research and Applications (MERRA-2), which has a horizontal resolution of 50 km and is suitable for investigating rainfall in small precipitation systems, and the Climate Forecast System Reanalysis and Reforecast (CFSR), which has a horizontal resolution of 38 km, were chosen as the reanalysis datasets. The Tropical Rainfall Measuring Mission (TRMM3B42) data were used as the satellite product in this study. The SWAT was set up to compare the performance of reanalysis datasets with that of the observed data. The model performance indicators indicated that, at daily time steps, all the three datasets produced the values of Nash–Sutcliff Efficiency (NSE≥0.4), coefficient of determination (R2≥0.4) and Percent Bias ±25%. Despite a general underperformance compared to MERRA-2, CFSR performed better than TRMM. On applying generated bias corrections for precipitation and temperature climate data, overall results showed that the bias-corrected data outperformed the original data. We conclude that, in the absence of gauged hydro-meteorological data, bias-corrected MERRA-2, CFSR and TRMM data could be used for simulating river discharge in data-scarce areas such as the tropical catchments in Uganda.
Comparing CFSR, TRMM, MERRA-2 and gauged precipitation datasets.
Investigating the potential of using the above reanalysis datasets to run the SWAT model.
Computing bias correction factors for the datasets using Linear Scaling and Local Intensity Scaling Methods.
Using the bias-corrected data to drive the SWAT model.
Comparing the performance of the SWAT models before and after application of bias correction factors.