Combining Global and Local Grid-Based Bias Correction for Mesoscale Numerical Weather Prediction Models

Two methods for objective grid-based bias removal in mesoscale numerical weather prediction models are proposed, one global and one local. The global method is an elaboration of model output statistics (MOS), combining several modern methods for multiple regression: alternating conditional expectation (ACE), regression trees, and Bayesian model selection. This allows the representation of nonlinear aspects of the bias, and the selection of the important ones in an overall nonlinear but parsimonious statistical model. The local method is the method of neighbors, which estimates the bias as the average bias over the “neighbors” of the grid point and time point being forecast, consisting of recent observations at stations that are close geographically and have similar elevation and land use. A case study based on MM5 48-hour surface temperature forecasts over the US Pacific Northwest indicates a 26% to 40% reduction in the mean-squared error (MSE) for either method, or a combination thereof, as compared to unadjusted model output. The method of neighbors yields a greater reduction in MSE, but the regression method is more flexible in that it can be applied even to grid points that have no neighbors.