Nonlinear rainfall-runoff modelling of semi-arid regions using ERA5 data

Semi-arid regions are very challenging environments for rainfall-runoff modelling due to the high spatial variation of rainfall alongside extreme wet and dry periods resulting in distinctly different hydrological conditions within the same catchment. In order to account for the extreme wet and dry periods, a new non-linear rainfall-runoff model has been developed. The non-linear model is capable of capturing the quick recessions observed during the wet periods alongside accounting for very little flows during the dry periods. The ability of the new model was assessed by comparing the linear version with the non-linear version on two nested catchments located within South Africa. The catchments areas are 183 and 328 km² and include 15% and 10% urbanisation respectively. The average rainfall during the wet period (May-Sep) is approximately 130mm per month with the dry period (Jan- Apr and Oct-Dec) averaging less than 35mm per month for the years 2000-2017. To challenge the problem of high spatial variation of rainfall, the fifth generation of ECMWF atmospheric reanalysis of the global climate ERA5 data is used. Both locally collected gauge and ERA5 reanalysis data were compared to show that the ERA5 data set was more capable than the local gauge in rainfall-runoff simulations with performance increases of up to 30\%. When comparing the default linear model and the non-linear model results based on ERA5 data showed the same level of performance for each model. However when flow duration curves and hydrographs were examined results showed that the linear model was not capable of adequately capturing the low flows of the catchment, whilst at the same time overestimating the high flows. Conversely, the non-linear model was capable of capturing the low flows recession and whilst it did also overestimate peak flows it was to a lesser extent than the linear model.