Recent Decline in Consistency Among Gridded Precipitation Datasets for the Greater Horn of Africa

Reliable observation networks are an essential component in predicting and understanding the mechanisms underlying seasonal rainfall in the Greater Horn of Africa (GHA). The importance of these networks is only increasing as GHA seasonal rainfall continues to decline. However, between 1990 and 2010, the number of station observations of three commonly used gridded precipitation datasets (Global Precipitation Climatology Centre V6, GPCC; Climate Research Unit, CRU; NOAA's Precipitation Reconstruction over Land, PREC) fell by 91.4%, 94.4% and 72.0% respectively in the GHA. The number of station observations for the decade 2000-2010 is now comparable to the same period one century ago. Here, we examine the impact this decrease of observations has had on the recorded variability of the observed rainfall during the two major GHA wet seasons: the “long rains”, March-May and the “short rains” October-December. We also aim to address how confidence in climate projections and climate variability studies are affected by this recent data sparsity. Both spatial and temporal patterns are examined, in 10 and 20 year periods, with the null hypothesis that no significant difference in rainfall variation is observed due to the decrease in station number. Standard metrics such as skill score, standard deviation and z-score are used to quantify the spatial variance. The standard deviation of rainfall in the GHA region is shown to decrease markedly from 1950-1970 to 1990-2010, indicating a decrease in the ability of the gridded datasets to accurately represent the spatial pattern of GHA rainfall. As well as the internal variation of each gridded dataset, the differences between the gridded datasets are also examined, and are shown to be significantly larger between 1990-2010 than between 1950-1970 (figure included). Both the long rains and the short rains have frequently been examined in terms of large scale teleconnections and interactions with global climate indices, such as the El Niño Southern Oscillation and Atlantic Multidecadal Oscillation. This study performs some basic analysis using climate index data to examine if the major mechanisms are represented differently in the gridded datasets in the more recent decades. Multiple satellite datasets are used as a measure of “ground truth”, for the period 1979-2010, to differentiate between physical and data quality based variations. Given the uncertain nature of climate projections for the East African region in general, it is important to clarify which variations in observed long term rainfall are physically interpretable. By examining the currently available rainfall data, we hope to emphasize these features while also quantifying regions of the GHA where changes in observed rainfall are coincident with a decrease in data availability and should therefore be treated with caution.