High-Resolution Monthly Rainfall Database for Ethiopia: Homogenization, Reconstruction, and Gridding

Recent heightened concern regarding possible consequences of anthropogenically induced global warming has spurred analyses of data aimed at detection of climate change and more thorough characterization of the natural climate variability. However, there is greater concern regarding the extent and especially quality of the historical climate data. In this paper, rainfall records of 233 gauge stations over Ethiopia for the 1978-2007 period are employed in an analysis that involves homogenization, reconstruction, and gridding onto a regular 0.5 degrees x 0.5 degrees resolution grid. Inhomogeneity is detected and adjusted based on quantile matching. The regularized expectation-maximization and multichannel singular spectrum analysis algorithms are then utilized for imputation of missing values, and the latter has been determined to have a marginal advantage. Ordinary kriging is used to create a gridded monthly rainfall dataset. The spatial and temporal coherence of this dataset are assessed using harmonic analysis, self-organizing maps, and intercomparison with global datasets. The self-organizing map delineates Ethiopia into nine homogeneous rainfall regimes, which is consistent with seasonal and interannual rainfall variations. The harmonic analysis of the dataset reveals that the annual mode accounts for 55%-85% of the seasonal rainfall variability over western Ethiopia while the semiannual mode accounts for up to 40% over southern Ethiopia. The dataset is also intercompared with Global Precipitation Climatology Project (GPCP), Climate Prediction Center (CPC) Merged Analysis of Precipitation (CMAP), Climatic Research Unit time series version 3 (CRUTS3.0), Tropical Rainfall Measuring Mission (TRMM), and interim ECMWF Re-Analysis (ERA-Interim) rainfall. The correlation of the dataset with global datasets ranges from 0.52 to 0.95 over sparse to dense rain gauge regions. The GPCP rainfall has a small bias and good correlation with the new dataset whereas TRMM and ERA-Interim have relatively large dry and wet biases, respectively.