Abstract: Systematic errors in the forecast of weather parameters appear in models for Numerical Weather Prediction (NWP), especially near ground level. The Kalman filter has been used for temperature prediction, satisfactorily reducing the systematic errors. We propose the use of a cascaded series of Kalman filters: the first Kalman filter uses real (actual) and forecast temperatures in order to compute prediction temperature, and each subsequent Kalman filter uses real (actual) and prediction temperature of the previous Kalman filter in order to compute prediction temperature. Temperature prediction improves by increasing the number of Kalman filters. Using three Kalman filters leads to negligible improvement in temperature prediction. The systematic errors almost disappear, since the percent successful prediction, which entails an absolute error less than 2 °C, is of the order of 97.5%.

Keywords: Forecasting, Prediction, Temperature, Kalman Filter, Steady-State, Finite Impulse Response

Pages: 195-200