Downscaling to high-resolution and correcting air temperature from the ERA5-Land over Ethiopia

Climate information from in-situ observation networks can be used to significantly improve the accuracy of gridded climate datasets, even in data-scarce regions. We applied a bias correction and spatial disaggregation method on daily maximum and minimum ERA5-Land (ERA5L) 2-m air temperature dataset covering Ethiopia. Due to large gaps in the observed temperature data, the bias correction is based on the statistics rather than the complete time series. First, long-term daily, monthly and annual temperature statistics (mean and variance) were summarized for the time series obtained from 155 stations covering the period 1981-2010. Second, the temperature statistics were interpolated onto a 0.05° x 0.05° grid using an inverse non-Euclidean distance weighting approach. This method accounts for the effects of elevation, thus enabling downscaling of the temperature to a higher spatial resolution. Next, the ERA5L maximum and minimum temperature were bias-corrected using quantile mapping assuming a Gaussian distribution transfer function. The quantile mapping was performed at daily, monthly and annual time steps to reproduce the climatology, seasonality, and interannual variability of the data. The performance of the bias correction was evaluated using the leave-out-one cross-validation method. The cross-validation shows that the bias-corrected maximum (minimum) daily temperature has an improved mean absolute error value of 68% (52%) in comparison to the original ERA5L reanalysis air temperature bias. The bias-corrected dataset is therefore suggested as an alternative for the ERA5L and can be used in a wide range of applications in Ethiopia.