Air temperature forecasts' accuracy of selected short-term and long-term numerical weather prediction models over Poland

Abstract

The article discusses the results of air temperature forecasts from four short-term and two long-term forecasts of numerical weather prediction models. The analysis covered the results of model simulations from January 2015 to January 2016 and compared them at 14 meteorological stations in Poland. The comparison was made based on the most commonly used measures for continuous parameters i.e., ME (mean error), MAE (mean absolute error), RMSE (root mean square error), MSE (mean square error), BIAS and Pearson correlation. In the short time horizon, the best results in the context of the MAE, RMSE, MSE and correlation values were obtained by the Unified Model, although the diagnosed differences between the models are small. All models in the 0-72 h projection horizon reached a correlation of 0.95-0.97 and an MAE in the range of 1.5 °C to 2.1 °C. In the case of long-term forecasts, the HIRLAM model was slightly better than the GFS model. Clearly, in both cases, there is a marked decrease in quality after the fourth and in the following forecast lead days.