Statistical Postprocessing of Numerical Weather Predictions Using a Stochastic Advection-Diffusion Model
Numerical weather prediction (NWP) models are capable of producing predictive fields at spatially and temporally high frequencies. They provide however only point forecasts, or in the case of ensemble forecasts,they are typically underdispersed. Statistical postprocessing serves to overcome these shortcomings. We present here a method which is able to take the space-time correlations of prediction errors from the NWP into account. For this we use a stochastic advection-diffusion partial differential equation (SPDE) whose parameters have a physical interpretation. We derive computationally efficient spectral methods to fit such a model on a dense grid, based on possibly censored and noisy observations of the solution at some grid points. The proposed model is applied to precipitation forecasts for northern Switzerland. Our postprocessed forecasts outperform the raw NWP predictions and they quantify prediction uncertainty. Show more
External linksSearch via SFX
SubjectSpace-time models; Stochastic partial differential equations; Numerical weather prediction; Hierarchical Bayes methods; Statistical postprocessing
Organisational unit03217 - Künsch, Hans Rudolf
NotesLecture hold on August 7, 2013 from 2:00 P.M. to 3:50 P.M..
MoreShow all metadata