A gridded maximum and minimum (Tx and Tn) daily temperature dataset derived by spatial downscaling and bias correction of the ERA5-Land (ERA5L) for the period 1981-2010 is presented. Observed daily Tx and Tn at 154 stations in Ethiopia covering record lengths of 5-30 years were used as a reference. The statistics that define the Gaussian distribution (mean and standard deviation) of Tx and Tn from the station observations were interpolated in space to create a monthly climatology and interannual statistics at 0.05° × 0.05° resolution using a hybrid interpolation approach that combines linear regression with topographic and location attributes, and non-Euclidean inverse distance weighting interpolation. The interpolated monthly and interannual statistics were then used to debias the ERA5L Tx and Tn using a quantile mapping approach. Leave-one-out cross-validation showed that the mean absolute errors in the corrected and downscaled daily temperatures are about 0.7 °C for Tx and 1.1 °C for Tn, reducing the statistical biases in the ERA5L Tx and Tn by 68% and 25% respectively. For monthly climatology, 40-64% of the biases were removed for Tx while for Tn the reductions range from 19% to 32%. The correction also improved commonly used indices for extremes like the probability of warm days, cold days, and warm nights, but overestimated the probability of cold nights. The presented open-access Tx and Tn dataset is a substantial improvement over existing gridded temperature datasets for Ethiopia, such as ERA5L and the Climate Hazards Infrared Temperature with Station (CHIRTS), and we suggest it is suitable for a wide range of environmental applications, e.g. in the fields of hydrology, agriculture, and ecology.
Keywords: Bias correction; Climate data; Inverse distance weighting; Linear regression; Quantile mapping; Spatial interpolation.
© 2022 The Author(s).