An enhanced meteorological ensemble postprocessing scheme for improving the prediction of extreme rainfall events

Suppression of large precipitation amounts in the forecasts of  heavy-to-extreme events has been a critical limitation of contemporary postprocessing schemes. To tackle this limitation, we investigate a statistical postprocessing scheme that explicitly accounts for the Type-II conditional bias in establishing the predictand-predictor relationship.  This scheme is a variant of Mixed-type Meta-Gaussian Distribution (MMGD) that relies on Conditional Bias Penalizing Regression (CBPR), rather than simple linear regression to relate precipitation forecast and observations.  To assess the effectiveness of this scheme, we perform a set of hindcast experiments wherein both CBPR and MMGD are applied to the Global Ensemble Forecast System (GEFS) to create probabilistic quantitative precipitation forecasts (PQPFs) over subbasins of three major river basins in California, namely the American, Russian, and Elk River Basins. These experiments broadly confirm that CBPR scheme is capable of producing PQPFs that are both better calibrated and less biased (in Type-II sense) than the baseline from MMGD.  Potential regionalization approach for determining the weight parameter a priori is discussed.