Several months of convection-allowing summertime WRF simulations in eastern and western North Dakota across multiple seasons were evaluated using object-based methods. The analysis of these simulations revealed that forecasts generated too many convective cells, convection that was too large, and overly-intense convection. Additionally, simulation performance was shown to have high sensitivity to the concentration of cloud droplets, affecting the number and area biases by means of the autoconversion process.
To further investigate the sensitivity and quantify model biases in the region, four months of regional convection-allowing forecasts using different versions of WSM6 and Thompson microphysics (including the “aerosol-aware” scheme) are analyzed. Forecasts are evaluated by comparing the simulated and observed reflectivity fields at multiple heights. Using the reflectivity fields allows for direct analysis of the structure of convective systems and provides information on the vertical extent of convection that can be missing in fields traditionally verified, such as accumulated precipitation. The evaluation looks at the number and area of objects created using different reflectivity magnitudes. To compare with an operational baseline, results from the operational HRRR model relative to the locally-run simulations in the study region are also discussed.