Temporal and spatial evaluation of long-term satellite-based precipitation products across the complex topographical and climatic gradients of Chile

Satellite-based rainfall estimates (SRE) have become a promising data source to overcome some limitations of ground-based rainfall measurements, in particular for hydrological and other environmental applications. This study evaluates the spatial and temporal performance of four long-term SRE products (TMPA 3B42v7, CHIRPSv2, MSWEPv1.1 and MSWEPv2.2) over the complex topography and climatic gradients of Chile. Time series of precipitation measured at 371 stations are compared against the corresponding grid cell of each SRE (in their original spatial resolution) at different temporal scales (daily, monthly, seasonal, annual). The modified Kling-Gupta efficiency along with its three individual components were used to assess the performance of each SRE, while two categorical indices (POD, and fBIAS) were used to evaluate the skill of each SRE to correctly capture different precipitation intensities. Results revealed that all SREs performed best in Central-Southern Chile (32.18-36.4 degrees S), in particular at low- and mid-elevation zones (0-1000 m a.s.l.). Seasonally, all products performed best in terms of KGE' during the wet autumn and winter seasons (MAM-JJA) compared to summer (DJF). In addition, all SREs were able to correctly identify no rain events, but during rainy days all SREs that did not use a local dataset of precipitation to recalibrate their estimates presented a low skill in providing an accurate classification of different precipitation intensities. Overall, MSWPEPv22 showed the best performance at all time scales and country-wide, due to the use of a Chilean dataset of daily data for calibrating its precipitation estimates, making it a good candidate for hydrological applications in Chile. Finally, we conclude that when the in situ precipitation dataset used in the evaluation of different SREs does not cover the headwaters of the catchments, the obtained performances should only be considered as first guess about how well a given SRE represent the real amount of water in an area.