Geostatistics to integrate gauge measurements with downscaled satellite estimates suitable for the local scale

Abstract

In countries such as Mexico, there is a lack of rain measurement stations. Additionally, in the Bajo Grijalva Basin, data of only three or fewer stations are integrated into satellite products of missions such as Tropical Rainfall Monitoring Mission (TRMM) and Global Precipitation Mission (GPM). Although Satellite missions enable obtaining rainfall at constant spacing (e.g., 11 km for GPM), this resolution is not suitable for local management. Integrating a larger quantity of gauge data with downscaled satellite values allows for obtaining local-scale precipitation data. In this work, Ordinary kriging (OK) was applied to downscale yearly aggregated precipitation satellite data (GPM-IMERG and TRMM: TMPA/3B43) and regression kriging (RK) to integrate them with the gauge measurements available in the basin of study. The resulting data were compared with the interpolation results of gauge measurements using OK and universal kriging (UK). Leave-one-out cross-validation (Lou-CV), principal components analysis, a correlation matrix, and a heat map with cluster analysis helped to evaluate the performance and to define similarity. An Inverse Distance Weighting (IDW) interpolation was included as a low-performance criterion in the comparison. OK performed well to downscale GPM satellite estimates. The RK integration of gauge data with downscaled GPM data got the best validation values compared to the interpolation of gauge measurements. Geostatistical methods are promising for downscaling satellite estimates and integrating them with all the available gauge data. The results indicate that the evaluation using performance metrics should be complemented with methods to define similarity among the values of the obtained spatial layers. This approach allows obtaining precipitation data useful for modeling and water management at the local level.