Forecast Verification Analysis of the CombiPrecip Ensemble

CombiPrecip is a real-time application developed by MeteoSwiss since 2012, which combines point raingauge measurements with radar-derived spatial estimations of precipitation over a vast 710x640km2 domain, extending beyond the Swiss borders. It relies on the geostatistics-based kriging with external drift as an interpolation technique. This method is probabilistic by nature, yielding both a mean value and an associated variance for every estimation. The purpose of our study is two-fold: (i) validate that the variance provided by the underlying geostatistical method of CombiPrecip does properly represent the uncertainty of the CombiPrecip product and (ii) devise a numerical method to build an ensemble of realistic-looking members based on this geostatistical variance. For this, we employ widely used probabilistic verification measures (reliability diagrams, rank histograms, ROC curves) for a large set of cross – validation results over the period 2016 – 2022. In addition, based on established methods developed within the nowcasting community, we produce ensembles of N realistic precipitation members that not only mimic the spatial autocorrelation of the mean-value CombiPrecip but also replicate its pixel-scale variance. Overall, our results indicate that observations fall reasonably well in the uncertainty range provided by the CombiPrecip ensemble.