Spatial and temporal variability of rainfall over the Republic of Djibouti from 1946 to 2017

For the first time, the spatial and temporal variability of rainfall in the Republic of Djibouti is investigated using data from 14 weather stations over the period 1946-2017. Due to limited data availability, high-resolution long-term satellite rainfall products (CHIRPS, PERSIANN-CDR, TAMSATv3, ARC2) and ERA5 reanalysis also contribute to document time-space rainfall variability at monthly, seasonal and annual scales. Principal component analysis identifies two spatially coherent regions of rainfall variability in the east (coastal zone) and the west (inland zone) of the country. Annual rainfall amounts are everywhere very low (60-300 mm), but with contrasted regimes. At seasonal scale, the highest rainfall amounts in the eastern part of the country are found between October-December (OND) and March-May (MAM), while July-September is the wettest season in the western part. The monthly rainfall regimes are relatively well reproduced by most products. ERA5 displays the highest monthly correlations with observations, followed by PERSIANN-CDR and CHIRPS. Trend analysis since 1983 shows a significant decrease of rainfall during MAM which is in agreement with other parts of East Africa. On the other hand, nonsignificant decreasing trends are observed in January-February (JF) and OND. Only June-September (JJAS) revealed a nonsignificant increasing trend, but it follows a prior drying trend since the 1950s. The impact of large-scale background climate on rainfall variability is assessed with focus given on El Nino Southern Oscillation (ENSO) and the Indian Ocean Dipole (IOD). Rainfall variability during OND shows a significant correlation with IOD, while in JJAS it is significantly negatively correlated with ENSO. In general, ERA5, CHIRPS and PERSIANN datasets are best able to reproduce rainfall patterns in Djibouti and suitable for further analysis. The fact that the interannual and decadal-scale rainfall variations in Djibouti show large-scale teleconnections with global sea-surface temperature fields, as demonstrated in this study, provides good prospects for the prediction of rainfall variations at a range of different temporal scales.