Support Vector Machine Quantile Regression based ensemble postprocessing

Current practice in predicting future weather is the use of numerical weather prediction (NWP) models to produce ensemble forecasts. Despite of enormous improvements over the last few decades, they still tend to exhibit bias and dispersion errors and consequently lack calibration. Therefore, these forecasts need to be statistically postprocessed.

Support vector machines are often used for classification and regression tasks in a wide range of applications, as e.g. energy, ecology, hydrology and economics. In this study, ensemble forecasts of 2m surface temperature are post-processed using a quantile regression approach based on support vector machines (SVMQR). This approach will be compared to the benchmark postprocessing methods ensemble model output statistics (EMOS), boosted EMOS and quantile regression forests (QRF). Instead of only regarding temperature variables as predictors, other weather variables including time dependence are taken into account as independent variables. The considered dataset consists of observations and forecasts for five years which cover Germany including three different forecast horizons. Despite of a shorter training period for SVMQR in contrast to e.g. boosted EMOS or QRF, SVMQR yields more calibrated quantile ensemble forecasts than the other approaches. Additionally, a comparable performance in terms of CRPS to the benchmark methods and a great improvement in comparison to the raw ensemble forecasts could be detected.