Kalman Filter and Analog Schemes to Postprocess Numerical Weather Predictions

• DOI: 10.1175/2011mwr3653.1
• Publication Date: 2011-03-18T20:32:24Z
• Authors: Luca Delle Monache; Thomas Nipen; Yubao Liu; Gregory Roux; Roland Stull
• Keywords: Model output statistics; Quantitative precipitation forecast

Abstract Two new postprocessing methods are proposed to reduce numerical weather prediction’s systematic and random errors. The first method consists of running a algorithm inspired by the Kalman filter (KF) through an ordered set analog forecasts rather than sequence in time (ANKF). forecast for given location is defined as past prediction that matches selected features current forecast. second weighted average observations verified when 10 best analogs were valid (AN). ANKF AN tested 10-m wind speed predictions from Weather Research Forecasting (WRF) model, with 400 surface stations over western United States 6-month period. Both predict drastic changes error (e.g., associated rapid regime changes), feature lacking KF 7-day running-mean correction (7-Day). almost eliminates bias raw (Raw), while drastically reduces it values slightly worse KF. analog-based also able errors, therefore improving predictive skill Raw. consistently best, improvements 10%, 20%, 25%, 35% respect ANKF, KF, 7-Day, Raw, measured centered root-mean-square error, 5%, 40%, rank correlation. Moreover, being based solely on observations, results efficient downscaling procedure representativeness discrepancies between predictions.