Assessment of Numerical Weather Prediction Model Storm Forecasts using an Object-Based Approach

Traditional pixel-by-pixel verification scores, such as the Critical Success Index (CSI) based on contingency table statistics, are widely used to evaluate the performance of forecasts of discontinuous fields such are precipitation. However, assessments based on simple lump scores offer little information of what aspect of a particular forecast may be good or bad and, therefore, they cannot provide diagnostic feedback that can be used to possibly improve a forecasting system.

In this study, we contrast traditional verification scores with results obtained utilizing an object-based forecast evaluation tool called MODE (Method for Object-based Diagnostic Evaluation) to provide more insightful performance statistics about convective storms predicted by the High-Resolution Rapid Refresh (HRRR) model that is under development by the NOAA Earth System Research Laboratory (ESRL) Global System Division (GSD). The object-based analysis yields valuable information about the HRRRs ability to predict the timing and location of new storm initiation and to reproduce the observed distribution of storm size and total number of storms over the diurnal cycle. Preliminary results based on three weeks of HRRR forecast data suggest that the HRRR is slow in initiating new storms compared to the observations.