ppt课件-introductiontonumericalweatherpredictionandensemble.pptVIP

Introduction to Numerical Weather Prediction and Ensemble Weather Forecasting Tom Hamill NOAA-CIRES Climate Diagnostics Center Boulder, Colorado USA NWP Process Other parameterizations Land surface Cloud microphysics Turbulent diffusion and interactions with surface Orographic drag Radiative transfer A lot happens inside a grid box Questions Can we accurately forecast the evolution of the pdf of the grid-box average weather? (focus mostly on this) How do we downscale from a grid-box average to a particular river basin or sub-area? Estimating the pdf of the weather: problem 1: chaos Problem 2: model error Problem 2: model error Probabilistic numerical weather prediction: theory Assume prior pdf of model state Step 1: assimilate new observations to sharpen pdf Step 2: forecast pdf forward in time Cycle short-range forecasts back to step 1. Probabilistic NWP: data assimilation Probabilistic NWP: forecasting pdfwith perfect model Probabilistic NWP: approximatingwith deterministic ensemble forecasts Probabilistic NWP: forecasting pdfwith imperfect model Probabilistic NWP: approximatingwith stochastic ensemble forecasts Ensemble forecasts: where are we today? Generating initial conditions: Each center has adopted their own approximate way of sampling from initial condition pdf. Breeding (NCEP) Singular vector (ECMWF) Perturbed observation (Canada) Stochastic-dynamic ensemble work just beginning (e.g., Buizza et al. 1999) Many attempts to post-process ensemble forecasts to provide reliable probability forecasts. Typical problems with current generation ensemble forecasts Would like to maximize pdf sharpness subject to calibration. But: Ensemble forecasts are biased Ensemble mean different (systematic model error; improve the model or post-process to correct errors) Ensemble spread less than it ought to be (better initial conditions, higher-res forecasts, incorporating stochastic effects). Downscaling Even if NWP centers produce calibrated ensemble forecasts a