Rainfall from ground-based rain gauges vs satellite products at different time steps in the southeast of the Republic of Djibouti

The wadi Ambouli is at the origin of rare and severe flash floods with increasing frequency in recent years. These floods are responsible for a lot of damage including human lives and good losses. Moreover, the aquifer located in the south-eastern part of the wadi Ambouli watershed (794 km²) is essential as the main source of drinking water for the city of Djibouti. Therefore, knowledge and prevention of flows events are of primary relevance for the development and flood prevention of the city of Djibouti neasted downstream to the wadi Ambouli.

Within the framework of the management of the urban expansion of the city, it is necessary, in the long term, to set up a hydrological model in order to model the flow and flood processes and to better understand the groundwater/surface water dynamics. A special focus is needed toward the extreme events.

To overcome the few rainfall available data, remote sensing techniques have been used to provide more representative information on rainfall spatio-temporal distributions. We present here a comparative study of rainfall data from 15 satellite products with those from rain gauge stations located in the wadi Ambouli watershed.

The analysis is conducted on various time steps, annual, monthly and daily, at the spatial scale from the Ambouli watershed (794 km²) to the point scale (comparison of point rainfall station and satellite grid). The comparisons of time series are based on the correlation coefficient r with the corresponding p-value and the Kling Gupta Efficiency (KGE) criterion.

The first results at annual and monthly time steps show that 5 satellite products present a high correlation coefficient and KGE with ground rainfall data (EWEMBI, GPCC, JRA_55_Adj, MSWEP and WFDEI_GPCC). For the Djibouti Aerodrome station, which is the one with the longest temporal data, the satellite products EWEMBI, JRA_55_Adj, and WFDEI_GPCC provide the best estimates of rainfall depth (annual scale). The MSWEP product (with r = 0.76) underestimates the annual rainfall over almost the entire study period.