Diurnal Corrections of Short-Term Surface Temperature Forecasts Using the Kalman Filter

Abstract

Surface temperature forecasts from numerical weather prediction models are known to have systematic errors, partly due to poor resolution of the topography and deficiencies in the physical formulation. The Kalman filter theory provides an excellent tool for combining observations and forecasts to correct such systematic errors. As detailed in this paper, deviations between temperature forecasts and observations were studied at several locations in Norway. It was found that the sign and magnitude of the deviations varied considerably in both time and space. Moreover, the diurnal variation of the systematic errors increased as the diurnal variation between daytime and nighttime increased. A Kalman filter model has been constructed that allows for diurnally varying corrections. Eight corrections, valid at 0000, 0300, 0600, 0900, 1200, 1500, 1800, and 2100 UTC, are calculated simultaneously. These corrections, which are weighted means of previous differences between observations and forecasts, are applied to correct +3-, +6-, +9-,. ? , +48-h temperature forecasts. One set of corrections is calculated and updated for each of about 240 observing stations throughout Norway. The results of the correction procedure have been verified for about one-half of them. It was found that the correction procedure reduced the monthly bias of the forecasts at each observing station to values close to zero. The monthly standard deviations of the differences between forecast and observations remained essentially unchanged. They were slightly reduced in summertime but could be slightly increased in wintertime for the long-term forecasts.