Some of the atmospheric processes leading to the formation of precipitation are highly nonlinear. This is reflected in the spatially and temporally discontinuous and highly variable nature of precipitation. Discontinuity and nonlinearity have their challenges not only in numerical modeling in terms of difficulties in tangent linear and adjoint model formulation and forecast verification, but also in statistical post-processing of numerical forecasts, including ensemble forecasts. While forecasts of continuous variables lend themselves to straightforward statistical post-processing techniques such as Model Output Statistics, precipitation is usually dealt with in a 2-step process. Considering the processing of ensemble data, in the first step, a sample of single or ensemble precipitation forecasts are compared with observations considering the occurrence of precipitation to derive a Probability Of (>= .01 inch) Precipitation (POP) forecast. Then, in a second round of analysis, forecasts with non-zero precipitation are compared with observations to derive the Probability Distribution of Precipitation (PDP), conditioned on precipitation being above zero. In this presentation, a continuous variable related to precipitation is introduced to overcome some limitations in statistical post-processing associated with the discontinuous nature of precipitation.
This publication was presented at the following: