Assessing the skill of a numerical weather prediction model over the central Mediterranean region

Abstract

Numerical Weather Prediction (NWP) models have become the most vital tools in weather forecasting, all over the world. NWP models have a vital role in providing both short- and long-range weather forecasts. Humans has always been captivated in the weather; it has always played a vital role in their everyday activities. With accurate weather forecasts, people can know and organise better for what would come or for what to expect. For example, in such weather events, warning processes are crucial to the public, most specifically to the frontlines (civil protection, aviation, hospital department) which need to be immediately warned and prepared for what can occur. Moreover, such areas like the central Mediterranean, can be more susceptible to the growing vulnerability of natural hazards due to the increase of densely population. Hence, the importance of good performance in NWP models is vital for best accuracy in weather forecasts. Although, by time NWP models have improved, there are still significant challenges in certain models, which are causing uncertainties in weather forecasting. Thus, the importance of model verification and validation from time to time is essential to discover any inaccuracies in the model. The main aim of this study was to assess the credibility of a short-range weather prediction model over the central Mediterranean. This was achieved by assessing the credibility of the High-Resolution Limited Area Model (HIRLAM), where model data was analysed and compared with Meteosat Second-Generation (MSG) Imagery data (EUMETSAT) at the 10:00 UTC point forecasts. Furthermore, a statistical non-parametric Mann-Kendall test and a parametric Pearson Correlation test were used to analyse any trends in the model parameters (Hit, False Alarm, Miss, Correct Negative) over a period of 37 days, covering the three seasons (Summer, Autumn, Winter). Additionally, any improvements or poor model performance were remarked. The study exhibited that with respect to cloud cover at the 10.00 UTC point forecasts, a good HIRLAM model performance was seen in 75% of the cases. This was observed in all of the seasons during the analysis of the hit and correct negative parameters. In which the hit scores indicated good presence of cloud forecasting whilst the correct negative indicated a good model performance for the absence of clouds. Furthermore, regarding the false alarm scores, a good improvement in model performance was shown during the winter season, in which the model does not increasingly overforecast cloud cover. Regarding the miss scores, the model showed an overall good improvement to accurate forecast cloud cover, during the summer and winter seasons. Lastly the statistical test analysis also illustrated graphically any trends shown during every day/ every season.