Gradient‐Boosted Mixture Regression Models for Postprocessing Ensemble Weather Forecasts

Nowadays, weather forecasts are based on ensemble forecasts which originate from multiple runs of numerical weather prediction models. However, such forecasts are usually miscalibrated and/or biased, thus requiring statistical postprocessing. Non‐homogeneous regression models, such as ensemble model output statistics, are frequently applied to correct these forecasts. Most of these models are based on a unimodal parametric distribution and provide improved, but not fully calibrated forecasts. To address this issue, a mixture regression model is presented, where the ensemble forecasts of each exchangeable group are linked to only one mixture component and mixture weight, called mixture of model output statistics (MIXMOS). In order to remove location‐specific effects and to use longer training data, standardized anomalies of the response and ensemble forecasts are used for the mixture of standardized anomaly model output statistics (MIXSAMOS). As carefully selected covariates, for example, from different weather variables, can enhance model performance, the non‐cyclic gradient‐boosting algorithm for mixture regression models is introduced. Furthermore, MIXSAMOS is extended by this gradient‐boosting algorithm (MIXSAMOS‐GB) that provides automatic variable selection. The novel mixture regression models substantially outperform the state‐of‐the‐art postprocessing models in a case study for 2 m$$ \mathrm{m} $$ surface temperature forecasts in Germany.