Local Wind Speed Estimation by Physical Downscaling of Weather model forecasts

Wind speed forecasts by numerical weather prediction (NWP) models in heterogeneous terrain lack local representativity as they are derived using grid-box averaged roughness lengths. In this paper a downscaling method is presented to increase the local accuracy of NWP-wind speed forecasts.

The method includes a simple two-layer model of the atmospheric boundary layer, used in combination with a high-resolution roughness map. The two-layer is used to post-process direct NWP-model output. The model comprises a surface layer and an Ekman-layer. In the surface layer vertical wind speed transformations are done using the logarithmic wind speed profile. In the Ekman-layer geostrophic resistance laws are applied.

The roughness map is derived from a land-use map and a simple footprint model. The roughness lengths are wind direction dependent and the footprint area of the Ekman-layer extends farther upstream than that of the surface layer. The roughness lengths compare well to those derived from gustiness analysis for station locations. The adjustment of the surface wind after a roughness transition as modeled by the two-layer model is similar to that of internal boundary layer models.

The NWP-model wind and the downscaled wind are evaluated in coastal zone areas, in estuaries, and at an airport in the Netherlands. Verification against in-situ observations shows that the downscaling method reduces the NWP surface wind speed error significantly, largely in terms of bias. The quality of the downscaled wind, however, depends highly on the quality of the high-resolution roughness map: inaccuracies of the land-use map may lead to local errors in the downscaled wind.