RainForest: a random forest algorithm for quantitative precipitation estimation over Switzerland

Abstract. Quantitative precipitation estimation (QPE) is a difficult task, particularly in complex topography, and requires the adjustment of empirical relations between radar observables quantities, as well methods to transform observations aloft estimations at ground level. In this work, we tackle classical problem with new twist, by training random forest (RF) regression learn QPE model directly from large database comprising 4 years combined gauge polarimetric observations. This algorithm carefully fine-tuned optimizing its hyperparameters then compared MeteoSwiss' current operational non-polarimetric method. The evaluation shows that RF able significantly reduce error bias predicted intensities, especially for solid or mixed precipitation. weak precipitation, however, despite posteriori correction, method has tendency overestimate. trained adapted run quasi-operational setup providing 5 min estimates on Cartesian grid, using simple temporal disaggregation scheme. A series six case studies reveal creates realistic fields, no visible artifacts, appear less smooth than original offers an improved performance five out events.