Impact of Horizontal Resolution on CAM-Derived Next-Day Probabilistic Severe Weather Forecasts

Convection-allowing Models (CAMs) allow for the explicit simulation of thunderstorms without a convective parameterization scheme and typically have horizontal grid-spacing of 4 km or less. CAMs provide significant improvements in forecast quality and value relative to non-CAMs, as CAMs offer guidance on storm mode and better resolve individual thunderstorms compared to models with coarser resolution. However, previous research offers conflicting results on the benefits of further increasing the horizontal resolution of a model already at convection-allowing resolution. Given the continued increase in computing power in recent years, an important question is whether additional computing resources would be better spent on increasing the horizontal resolution of CAMs or on adding members to form CAM ensembles.

The present study analyzes Advanced Research Weather Research and Forecasting (ARW-WRF) model data from 63 days of the 2010 and 2011 NOAA Hazardous Weather Testbed Spring Forecasting Experiments (HWT SFEs) to investigate the above question as it pertains to next-day probabilistic severe weather forecasts. Next-day probabilistic severe weather forecasts are produced from forecast updraft helicity fields from a deterministic 1-km horizontal grid spacing CAM, a deterministic 4-km CAM, and an 11-member 4-km CAM ensemble. Relative operating characteristic (ROC) curves, performance diagrams, and reliability diagrams are constructed to compare the three forecasts. The area under the ROC curve (AUC) is also used to evaluate forecast quality. A resampling methodology is employed to test for significant differences in the AUC between the forecasts.

Results indicate that, while no significant difference in the AUC exists between the 1-km and 4-km deterministic forecasts, a significant difference in the AUC does exist between the 4-km deterministic and the 4-km ensemble forecasts. Such a result suggests that, for purposes of improving CAM-derived next-day probabilistic severe weather forecasts, additional computational resources may be better spent on adding members to create CAM ensembles than on increasing the horizontal resolution of deterministic CAMs.