Statistical Correction of the COSMO Model Weather Forecasts Based on Neural Networks

Different methods for the statistical correction of the forecasts of surface parameters using the COSMO-Ru13-ENA model with the lead time up to 117 hours are considered. The methods include the systematic correction using the data from recent observations at a weather station, the correction based on special neural networks as well as different combinations of these two techniques. The study presents the estimates of the results of applying the analyzed correction methods to the forecasts of surface air temperature, dew point, and wind speed modulus based on the independent sample for 2018 with the total volume of 2.34 x 10(7) forecasts. The correction method based on neural networks reduces forecast errors even at the points where meteorological observations have not been carried out.