Monsoon rainfall is central to the climate of West Africa, and understanding its variability is a challenge for which satellite rainfall products could be well suited to contribute to. Their quality in this region has received less attention than elsewhere. The focus is set on the scales associated with atmospheric variability, and a meteorological benchmark is set up with ground-based observations from the African Monsoon Multidisciplinary Analysis (AMMA) program. The investigation is performed at various scales of accumulation using four gauge networks. The seasonal cycle is analyzed using 10-day-averaged products, the synoptic-scale variability is analyzed using daily means, and the diurnal cycle of rainfall is analyzed at the seasonal scale using a composite and at the diurnal scale using 3-hourly accumulations. A novel methodology is introduced that accounts for the errors associated with the areal–time rainfall averages. The errors from both satellite and ground rainfall data are computed using dedicated techniques that come down to an estimation of the sampling errors associated to these measurements. The results show that the new generation of combined infrared–microwave (IR–MW) satellite products is describing the rain variability similarly to ground measurements. At the 10-day scale, all products reveal high regional and seasonal skills. The day-to-day comparison indicates that some products perform better than others, whereas all of them exhibit high skills when the spectral band of African easterly waves is considered. The seasonal variability of the diurnal scale as well as its relative daily importance is only captured by some products. Plans for future extensive intercomparison exercises are briefly discussed.

International audienceIn the frame of the African Monsoon Multidisciplinary Analyses (AMMA) programme, a specific rainfall algorithm (EPSAT-SG; Estimation of Precipitation by SATellites ? Second Generation) was developed for the requirements of the scientific community and an intercomparison exercise was undertaken to assess the performance of various rainfall analyses to help users of satellite precipitation estimates to take into consideration the limitations of these products. The intercomparison exercise presented in this article includes three regional precipitation products as well as seven operational global products that are publicly available and easily accessible on websites. This study has been performed using validation data from rain gauge observations analyses on the Sahelian region provided by the AGRHYMET centre, for three rainy seasons from 2004 to 2006. The 10 different satellite-based precipitation products are verified against the same reference ground-based dataset of 10-day rainfall accumulations at the 0.5° × 0.5° latitude-longitude resolution. The performance of the different precipitation algorithms is assessed according to various indicators such as the behaviour of the precipitation distributions, several statistical parameters and spatial distribution of the errors. All the statistical results indicate that the three 'near-real-time' products (3B42-RT, CPC MORPHing technique (CMORPH) and Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks (PERSIANN)) have a poorer performance than the other products considered for intercomparison. In fact these algorithms cannot make use of useful inputs such as rain gauge observations that are not available at near real time. It is noted that the simple basic Geostationary Operational Environmental Satellite (GOES) Precipitation Index (GPI) product performs better, with a higher skill score index. The three products Global Precipitation Climatology Project (GPCP)-1dd, Global Satellite Mapping of Precipitation (GSMaP)-MVK and Tropical RainfallMeasuring Mission (TRMM)- 3B42 obtain better statistical results but the best results are obtained by the precipitation products created specifically for thisAfrican region.TheEPSAT-SGproduct has the best performance according to several statistical criteria including skill score, coefficient of determination and rootmean square (RMS) errorwhereas theRainFall Estimation (RFE)-2.0 estimates offer the bestmatchwith validation estimates in term of distribution and bias. The Tropical Applications ofMeteorology using SATellite and other data (TAMSAT) estimates have also similar statistically good results