A0783
Title: Enhancing multivariate post-processed visibility predictions utilizing CAMS forecasts
Authors: Maria Nagy-Lakatos - University of Debrecen (Hungary) [presenting]
Sandor Baran - University of Debrecen (Hungary)
Abstract: Modern weather forecasts increasingly use ensemble predictions for visibility, which is crucial for aviation, maritime navigation, and air quality and impacts public health. However, visibility predictions often lack the accuracy seen in other weather variables. To address this, statistical post-processing is recommended, using historical data to align predictive distributions with actual observations. Visibility is reported in discrete values by world meteorological organization standards, leading to discrete predictive distributions. Although classification algorithms can improve forecast accuracy, they might lose spatial and temporal dependencies. Whether including Copernicus Atmosphere Monitoring Service (CAMS) predictions as an additional covariate is investigated in visibility ensemble forecasts from the European Centre for Medium-Range Weather Forecasts enhances classification-based post-processing methods and preserves spatial dependence. Joint multivariate post-processing of forecasts for all 30 investigated SYNOP observation stations is performed using the two-step ensemble copula coupling and Schaake shuffle approaches, which utilize the dependence structure of the raw vector ensemble forecasts and historical observations, respectively. It is confirmed that post-processed forecasts significantly outperform raw and climatological predictions, and incorporating CAMS forecasts further improves both univariate and multivariate predictions.
