Combined Kalman Filter and Universal Kriging to Improve Storm Wind Speed Predictions for the Northeastern United States

Abstract

Abstract The scope of this study is to assess a combination of well-known techniques for bias reduction and spatial interpolation in an attempt to improve wind speed prediction for storms on a gridded domain. This is accomplished by implementing Kalman filter (KF) for bias reduction and universal kriging (UK) for spatial interpolation as postprocessing steps for the Weather Research and Forecasting (WRF) Model. It is shown that for surface wind speed, a linear KF is adequate for eliminating systematic model errors with the available storm history. KF-estimated wind speed biases at station locations are then interpolated across the model domain using UK. The combined KF–UK approach improves the wind speed forecast median bias by 55% and RMSE by 15% (bulk statistics), while benefits obtained at station-specific locations can reach maximum improvements of 72% for RMSE and 100% for bias. Contingency statistics that inform on model performance over four categories of wind speed magnitude reveal that calm/moderate winds are successfully corrected but strong/gale winds cannot be adequately corrected by the combination of KF and UK, which is a disadvantage for improving prediction of severe storm conditions.