B0544
Title: Statistical post-processing of visibility ensemble forecasts
Authors: Maria Nagy-Lakatos - University of Debrecen (Hungary) [presenting]
Sandor Baran - University of Debrecen (Hungary)
Abstract: Accurate and reliable forecasting of visibility is vital in aviation meteorology and has great importance in shipping and road transportation as well. Recently, major meteorological services issue ensemble forecasts of visibility; however, these forecasts are often uncalibrated and have far lower forecast skill than ensemble forecasts of other weather quantities. Hence, some form of statistical post-processing is required to improve predictive performance. According to the suggestions of the World Meteorological Organization, visibility observations are reported in discrete values. Thus, calibration can be considered as a classification problem. Based on visibility ensemble forecasts of the European Centre for Medium-Range Weather Forecasts for Central-Europe for calendar years 2020-2021 and corresponding observations, we investigate the predictive performance of proportional odds logistic regression and multilayer perceptron neural network, which approaches provide the best forecast skill in a similar problem of calibrating total cloud cover ensemble forecasts. We show that compared with the raw ensemble forecasts, post-processing results in more than 20\% improvement in forecast skill, and clustering the observation stations based on station climatology is superior to regional modelling.
