Object-Based Verification of Precipitation Forecasts. Part I: Methodology and Application to Mesoscale Rain Areas

TLDR

The study introduces a new method for defining rain areas to verify precipitation forecasts from numerical weather prediction models. The method defines precipitation objects in forecasts and observations using convolution and thresholding, then matches them by centroid distance, and is applied to WRF 22‑km grid forecasts versus stage‑IV 4‑km observations over the continental United States in July–August 2001. WRF forecasts produce too many large rain areas, exhibit a positive size bias peaking in the late afternoon, underestimate the diurnal cycle of rain‑area occurrence, narrow the rainfall intensity range, and show matching performance that depends on object size and synoptic‑scale predictability.

Abstract

Abstract A recently developed method of defining rain areas for the purpose of verifying precipitation produced by numerical weather prediction models is described. Precipitation objects are defined in both forecasts and observations based on a convolution (smoothing) and thresholding procedure. In an application of the new verification approach, the forecasts produced by the Weather Research and Forecasting (WRF) model are evaluated on a 22-km grid covering the continental United States during July–August 2001. Observed rainfall is derived from the stage-IV product from NCEP on a 4-km grid (averaged to a 22-km grid). It is found that the WRF produces too many large rain areas, and the spatial and temporal distribution of the rain areas reveals regional underestimates of the diurnal cycle in rain-area occurrence frequency. Objects in the two datasets are then matched according to the separation distance of their centroids. Overall, WRF rain errors exhibit no large biases in location, but do suffer from a positive size bias that maximizes during the later afternoon. This coincides with an excessive narrowing of the rainfall intensity range, consistent with the dominance of parameterized convection. Finally, matching ability has a strong dependence on object size and is interpreted as the influence of relatively predictable synoptic-scale systems on the larger areas.