Contribution to the forecast skill of meteorological and air pollution numerical predictions at meso and urban scale with post-processing algorithms

Thumbnail Image




Παππά, Αρετή

Journal Title

Journal ISSN

Volume Title



Significant scientific and technοlogical breakthroughs in the last century have enabled the quantification of uncertainty in unstable nonlinear dynamic systems, such as the atmosphere. Complex atmospheric processes, including atmospheric dynamics, energy transfers, and chemical reactions have been successfully simulated by Numerical Weather Prediction (NWP) models. The ability of these models to accurately represent phenomena across a diverse range of scales, from the microscale to the global scale, has established them as a fundamental component in atmospheric studies. Their pivotal role has, in turn, spurred an increased focus in scientific research on enhancing the accuracy of numerical forecasts. This entails refining NWP models for a more precise representation of atmospheric processes and applying advanced statistical methods for improving model outputs. This dual approach - advancing model sophistication while also developing robust statistical techniques - has been instrumental in elevating the precision and reliability of weather forecasting. The subject of this dissertation focuses on developing and assessing advanced statistical techniques aimed at enhancing the accuracy of weather and air quality numerical predictions. These approaches aim to tackle the inherent uncertainties in weather and air quality modeling by correcting errors in NWP outputs, showcasing a comprehensive methodology for improving forecasting in these vital domains. As a preliminary step, a comprehensive assessment is carried out to evaluate the accuracy and effectiveness of the numerical weather and air quality predictions. This is followed by an analysis of how inaccuracies in meteorological forecasts impact air quality forecasting. The study culminates in the employment of state-of-the-art statistical methods. These include post-processing filters, analytically optimized multi-model ensemble techniques, and the generation of neural networks, all aimed at enhancing the accuracy of numerical predictions. The versatility of these techniques extends to a variety of applications, ranging from improving air quality forecasts to supporting renewable energy generation



Numerical weather prediction, Long short-term memory networks, Post-processing algorithms, Air pollution