Assessment of 21 gridded precipitation datasets quality in estimating rainfall in Madagascar

In November 2021, the United Nations has declared that Madagascar current humanitarian crisis was the world first famine due to drought induced by climate change. Understand the factors triggering these crises is fundamental to anticipate and reduce the effects of future crises. In the ongoing climate change context, rainfall deficit is a key variable in setting up these crises. However, the Malagasy weather station network is far too sparse to describe the spatial variability of rainfall. Thus, this study compares 21 gridded precipitation products (see Fig. 1) to 19 gauges data to assess their reliability to monitor Madagascar precipitation pattern. Based on four statistical indices (Kling-Gupta Efficiency, Root Mean Square Error, Correlation coefficient and Biais), WFDEI-GPCC, CMORPH-BLD and, MSWEP appeared the most reliable product at national scale (Fig. 1). However, the products performances varied throughout Madagascar because the island counts four contrasted bioclimatic zones due to topographic and meteorological influence. To circumvent this variability issue, we developed a new product based on ensemble approach (“EM” in the Fig. 1). EM is calculated from a weighted average of the products according to their KGE score. The weights are specifically defined by bioclimatic zone. EM shows better performances at both regional and national scale. Finally, we evaluated the quality of our results by comparing them with five independent gauge dataset and on a different period. These satisfactory results allow us to continue studying the impact of rainfall related droughts on humanitarian crisis.

 

Figure 1: Scores for 21 gridded products + the new product created based on ensemble approach (« EM »). The monthly precipitation estimated by these products is compared to the monthly precipitation measured by 19 gauges data between 2011 and 2016. Four statistical indices (Kling-Gupta Efficiency, Root Mean Square Error, Correlation coefficient and Biais) are shown. Their optimum is indicated by the dashed green line. The red point is the median value obtained when considering only dry season, and the blue dot is the median value when considering only wet season. The products are sorted by the KGE median value.