Fixing WRF's High Speed Wind Bias: A New Subgrid Scale Drag Parameterization and the Role of Detailed Verification

A major problem for the Weather Research and Forecasting (WRF) mesoscale modeling system and many others is a tendency to overpredict wind speed at the surface. Associated with this problem is a robust wind direction error in which the flow tends to be too geostrophic at low levels in areas of complex terrain. It is hypothesized that part of this problem is that even at fairly high horizontal resolution (grid spacings of 5-15 km), the drag from hills and other smaller scale terrain features are not well represented. The new parameterization is based on making the surface roughness length, z0, proportional to the magnitude of sub-grid scale terrain variance. This parameterization is tested for a several-month period using the WRF model run at 12-km resolution over the Pacific Northwest. As will be described in this talk, the forecast wind speed biases are nearly removed, and the wind direction bias is reduced from approximately 10° to 2-5°. Verifying wind speed frequency distributions shows that the new parameterization produces a distribution nearly identical to that of observations, in stark contrast with the default WRF model. Verifying the wind speed bias versus forecast wind speed category shows large improvements for all wind categories. In contrast, verifying against observed wind categories shows improvements for low and moderate winds speeds, but degradation for high wind speeds. Using detailed spatial verifications and examining individual cases reveals that such errors are mainly along coastlines and in high terrain where inadequate resolution is undermining forecast accuracy. Timing errors also inevitably produce errors at high wind speeds. This and other evidence indicates that the model overprediction bias tends to compensate for other errors, and that the new drag parameterization produces large and consistent improvements in the low-level winds produced by the WRF system.