Analysis, Characterization, Prediction and Attribution of Extreme Atmospheric Events with Machine Learning: a Review

Here, we provide a comprehensive review and systematic comparison of eight statistical and machine learning methods for probabilistic wind gust forecasting via ensemble postprocessing, that can be divided in three groups: State of the art postprocessing techniques from statistics (ensemble model output statistics (EMOS), member-by-member postprocessing, isotonic distributional regression), established machine learning methods (gradient-boosting extended EMOS, quantile regression forests) and neural network-based approaches (distributional regression network, Bernstein quantile network, histogram estimation network). The methods are systematically compared using six years of data from a high-resolution, convection-permitting ensemble prediction system that was run operationally at the German weather service, and hourly observations at 175 surface weather stations in Germany.

Abstract We present an ensemble prediction system using a Deep Learning Weather Prediction (DLWP) model that recursively predicts six key atmospheric variables with six‐hour time resolution. This computationally efficient model uses convolutional neural networks (CNNs) on a cubed sphere grid to produce global forecasts. The trained model requires just three minutes on a single GPU to produce a 320‐member set of six‐week forecasts at 1.4° resolution. Ensemble spread is primarily produced by randomizing the CNN training process to create a set of 32 DLWP models with slightly different learned weights. Although our DLWP model does not forecast precipitation, it does forecast total column water vapor and gives a reasonable 4.5‐day deterministic forecast of Hurricane Irma. In addition to simulating mid‐latitude weather systems, it spontaneously generates tropical cyclones in a one‐year free‐running simulation. Averaged globally and over a two‐year test set, the ensemble mean RMSE retains sk …

Abstract Probabilistic forecasts in the form of ensembles of scenarios are required for complex decision-making processes. Ensemble forecasting systems provide such products but the spatiotemporal structures of the forecast uncertainty is lost when statistical calibration of the ensemble forecasts is applied for each lead time and location independently. Nonparametric approaches allow the reconstruction of spatiotemporal joint probability distributions at a small computational cost. For example, the ensemble copula coupling (ECC) method rebuilds the multivariate aspect of the forecast from the original ensemble forecasts. Based on the assumption of error stationarity, parametric methods aim to fully describe the forecast dependence structures. In this study, the concept of ECC is combined with past data statistics in order to account for the autocorrelation of the forecast error. The new approach, called d-ECC, is applied to wind forecasts from the high-resolution Consortium for Small- …

While probabilistic forecast verification for categorical forecasts is well established, some of the existing concepts and methods have not found their equivalent for the case of continuous variables. New tools dedicated to the assessment of forecast discrimination ability and forecast value are introduced here, based on quantile forecasts being the base product for the continuous case. The relative user characteristic (RUC) curve and the quantile value plot allow us to analyse the performance of a forecast for a specific user in a decision‐making framework. The RUC curve is designed as a user‐based discrimination tool and the quantile value plot translates forecast discrimination ability in terms of economic value. The relationship between the overall value of a quantile forecast and the respective quantile skill score is also discussed. The application of these new verification approaches and tools is illustrated based on synthetic datasets, as well as for the case of global radiatio …

Here, we provide a comprehensive review and systematic comparison of eight statistical and machine learning methods for probabilistic wind gust forecasting via ensemble postprocessing, that can be divided in three groups: State of the art postprocessing techniques from statistics (ensemble model output statistics (EMOS), member-by-member postprocessing, isotonic distributional regression), established machine learning methods (gradient-boosting extended EMOS, quantile regression forests) and neural network-based approaches (distributional regression network, Bernstein quantile network, histogram estimation network). The methods are systematically compared using six years of data from a high-resolution, convection-permitting ensemble prediction system that was run operationally at the German weather service, and hourly observations at 175 surface weather stations in Germany.