State-of-the-art hydrological impact studies are based on weather data from regional climate model ensembles and require a certain degree of accuracy. Despite regional downscaling to finer resolution, RCM simulations often show considerable biases when compared to observed data and bias-correction is an attempt to improve the quality and reliability. However, correction methods are based on the stationarity assumption which presumes that future physical processes in the atmosphere are comparable to the period used to correct the simulations. This dataset provides bias-corrected daily precipitation, min/mean/max air temperature of ten CORDEX RCM runs covering the country of Ethiopia for historical (1970-1999) and over the 21st century for RCP 4.5 and RCP 8.5.

Precipitation biases in most CORDEX RCMs show a high seasonality for grid boxes within the evaluation domain of Ethiopia. This limits a bias correction based on seasonal or annual means. However, as some of these grid boxes do show almost no precipitation events for single months, a harmonic-based bias correction method analogously to the one applied to temperature is not feasible for precipitation. Furthermore, this results in a large uncertainty in the estimation of the corresponding monthly biases. Thus, a bias correction is only applied on months and grid boxes with more than 100 rainy days (rainfall above 1 mm/day) within the calibration period (1951--2001). The method applied is based on a local rainy day intensity scaling, correcting the frequency of rainy days and the mean precipitation on rainy days to fit the observed values in a specific calibration period. The underlying idea is the assumption of a smooth seasonal cycle for the variables simulated by the RCM and the observational reference (WATCH Forcing Data, WFD). These cycles are modelled with a series of harmonic functions using vector generalised linear models and the difference in cycles between an RCM reference simulation and the observational product is used for bias correction of the RCM projection.