Modeling Ensemble Prediction of Solar Flares
Abstract
The advantage of using ensemble techniques in forecasting has been proven in many fields, especially in terrestrial weather. In space weather the use of ensembles is becoming increasingly important. In particular, ensemble forecasting of solar flares has shown the potential to considerably improve the quality and performance of predictions and, as we will show, allows the estimation of uncertainties even in the cases when the uncertainties for input predictions are unknown. Using predictions for major flares (M- or X- class in the next 24 hours) from six (automated and human-influenced) methods we construct ensemble forecasts by linearly combining the input predictions. Each resulting ensemble (or combination) is characterized by a set of weights obtained by optimizing a particular performance metric, the Brier score, for example. Approximately a dozen performance metrics (probabilistic and categorical) were studied. We found that mathematical metrics such as non-linear correlation coefficients produced ensemble predictions performing at the top of our results. Uncertainties in the ensemble predictions are obtained through the combination weights, which are obtained using a MCMC-like algorithm and therefore their uncertainties can be assessed and propagated. Our results will provide operational forecasters with guidelines for constructing, training, and implementing ensemble forecasting of solar flares tailored for particular needs.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2018
- Bibcode:
- 2018AGUFMSM53A..04G
- Keywords:
-
- 1990 Uncertainty;
- INFORMATICSDE: 7924 Forecasting;
- SPACE WEATHERDE: 7959 Models;
- SPACE WEATHERDE: 7999 General or miscellaneous;
- SPACE WEATHER