Publications

Numerical weather prediction (NWP) models generally exhibit systematic errors in the forecast of near-surface weather parameters due to a wide number of factors, including poor resolution of model topography, or deficient physical parameterizations. In this work, deviations between 2 m-temperature observations and forecasts provided by the European Centre for Medium-Range Weather Forecasts (ECMWF) are analysed at 12 synoptic stations located in Portugal. Systematic errors vary considerably with geographical location and time of day as well as throughout the year. The Kalman Filter theory provides a suitable tool to correct systematic errors of this type and therefore improve model forecasts. Accordingly, a Kalman Filter is applied to 2 m-temperature forecasts issued in 2003, a year marked by one of the most severe heat waves in Europe. It is shown that the developed methodology is versatile in adapting its coefficients to different seasons and weather conditions. The proposed Kalman Filter allows an objective forecast correction for 2 m-temperature, reducing the bias of the forecasts at each station to values close to zero, and improving the root mean square error from 10% up to over 70%, with respect to the raw ECMWF forecasts.