Evaluating remotely sensed rainfall estimates using nonlinear mixed models and geographically weighted regression

This article evaluates an infrared-based satellite algorithm for rainfall estimation, the Convective Stratiform technique, over Mediterranean. Unlike a large number of works that evaluate remotely sensed estimates concentrating on global measures of accuracy, this work examines the relationship between ground truth and satellit0e derived data in a local scale. Hence, we examine the fit of ground truth and remotely sensed data on a widely adopted probability distribution for rainfall totals – the mixed lognormal distribution – per measurement location. Moreover, we test for spatial nonstationarity in the relationship between in situ observed and satellite-estimated rainfall totals. The former investigation takes place via using recent algorithms that estimate nonlinear mixed models whereas the latter uses geographically weighted regression.