A Simplified Bivariate Meta-Gaussian Model of Forecast-Observation Dependence Based on the Pseudo Precipitation

Mixed-type meta Gaussian distribution (MMGD) is a statistical mechanism employed by National Weather Service’s Meteorological Ensemble Forecast Preprocessor (MEFP) for creating postprocessed distribution of forecast variables. Sampling from MTMGD conditioned on specific forecast values at given location and lead time normally produces a set of modified ensemble forecasts with higher reliability and lower bias.. MMGD has a clear limitation that it requires a large of parameters to characterize the dichotomous-continuous nature of precipitation, and the parameters may not be reliably estimated from a limited sample of historical rainfall. To address this limitation, we propose a new, simplified framework for MEFP that relies on the concept of pseudo precipitation(PP), i.e. the measure of column atmospheric water vapor deficit. This results in a symmetric distribution which is more amenable to optimization. We conduct a case study over the State of Texas to examine the robustness of the new framework, which we term MEFP-PP, wherein we apply both the operational and the simplified versions of MEFP to model the bivariate distribution of 6-h precipitation forecast and observations. The analyses suggest that the MEFP-PP has advantages for basins with short historical records where estimation of MEFP parameters is hampered by the lack of a sufficient number of forecast-observation pairs with positive precipitation.