Quantification of Predictive Skill for Mesoscale and Synoptic-Scale Meteorological Features as a Function of Horizontal Grid Resolution

Abstract

To quantitatively assess numerical predictive skill for synoptic and mesoscale features as a function of horizontal grid resolution, a series of experiments is conducted using the Pennsylvania State University-National Center for Atmospheric Research Mesoscale Model. For eight cases of continental cyclogenesis, 72-h integrations are examined using grids of 160, 80, and 26.7 km. First, we briefly examine error statistics for synoptic-scale cyclones and anticyclones. Next, a detailed analysis of model errors for mesoscale features is presented. A bandpass filtering technique, based on the Barnes objective analysis scheme, is used to separate perturbation quantities associated with the mesoscale features from the synoptic-scale fields. Error statistics are then compiled for various mesoscale features, including the intensity of mesolows, damming ridges, and postfrontal troughs, and the thermal gradients, propagation speed, and vertical velocity maxima associated with surface cold fronts. Finally, the accuracy of the predicted precipitation fields, produced using the Anthes-Kuo cumulus parameterization, is examined. Objective verification reveals that forecast skill does not improve uniformly for all types of mesoscale features as horizontal grid resolution is increased, although the general trend is for reduced errors as expected. Improvements do occur on both the 80- and 27-km grids for all geographically related mesoscale features (such as orographic lee troughs). A similar improvement is seen for propagating mesoscale features (such as postfrontal troughs) and synoptic-scale cyclones as the grid length is reduced from 160 to 80 km. However, when the grid length is further reduced to 27 km, mean absolute errors and mean position errors actually increase for both classes of features. This greater variability in model performance suggests that as grid resolution is enhanced, other factors such as the accuracy of model physics and initial conditions become increasingly important. The effect on precipitation bias and threat scores in these experiments is positive (reduced errors) when resolution is improved from 160 to 80 km but is generally insignificant or negative for the 27-km grid. Based on these results, the Anthes-Kuo convective parameterization used in these experiments is not recommended for application on grids of about 30 km or less.