Probabilistic thunderstorm forecasting by blending multiple ensembles

Abstract

In numerical weather prediction models, point thunderstorm forecasts tend to have little predictive value beyond a few hours. Thunderstorms are difficult to predict due largely to their typically small size and correspondingly limited intrinsic predictability. We present an algorithm that predicts the probability of thunderstorm occurrence by blending multiple ensemble predictions. It combines several post-processing steps: spatial neighbourhood smoothing, dressing of probability density functions, adjusting sensitivity to model output, ensemble weighting, and calibration of the output probabilities. These operators are tuned using a machine learning technique that optimizes forecast value measured by event detection and false alarm rates. An evaluation during summer 2018 over western Europe demonstrates that the method can be deployed using about a month of historical data. Post-processed thunderstorm probabilities are substantially better than raw ensemble output. Forecast ranges from 9 hours to 4 days are studied using four ensembles: a three-member lagged ensemble, a 12-member non-lagged limited area ensemble, and two global ensembles including the recently implemented ECMWF thunderstorm diagnostic. The ensembles are combined in order to produce forecasts at all ranges. In most tested configurations, the combination of two ensembles outperforms single-ensemble output. The performance of the combination is degraded if one of the ensembles used is much worse than the other. These results provide measures of thunderstorm predictability in terms of effective resolution, diurnal variability and maximum forecast horizon.

Verification: Assessment of Calibration and Accuracy

“…In a way, RAFT can be regarded as in the same spirit as other methods that combine numerical weather forecasts and recent observations, such as nowcasting or short-range forecast blending . The latter involves the estimation of blending weights, which can be achieved based on a range of criteria and with a variety of different methods (e.g., Atencia et al, 2020 ;Bouttier and Marchal, 2020 ;Schaumann et al, 2020 ). However, RAFT is not a direct competition of these techniques, but rather complements them.…”

A deep‐learning model to predict thunderstorms within 400 km2 South Texas domains

“…Finally, the performance of the DL model was compared with a recent state‐of‐the‐art ensemble prediction system. Bouttier and Marchal (2020) assessed the performance of four ensemble systems (four different sets of ensembles) derived from three separate NWP models, and three ensemble blends (combinations of two ensemble systems), when predicting thunderstorm occurrence (convective initiation) in Western Europe. They defined an ensemble thunderstorm prediction as the probability that a thunderstorm activity diagnostic variable exceeded a specific threshold.…”

“…Kain et al . (2013) and Bouttier and Marchal (2020) demonstrated the utility of this state‐of‐the‐art ensemble approach to thunderstorm prediction.…”

Machine Learning for Climate Precipitation Prediction Modeling over South America

“…The Météo-France is using machine learning to predict the probability of thunderstorm occurrence by blending multiple ensemble predictions. Several post-processing steps are used: spatial neighborhood smoothing, dressing of probability density functions, adjusting sensitivity to model output, ensemble weighting, and calibration of the output probabilities [8]. Accordingly, it is important to recognize the impact of Artificial Intelligence (AI) on meteorology.…”