Forecasting of solar particle event integral proton fluences using Bayesian inference

Previous work by the authors has demonstrated the ability to predict solar particle event dose and dose rate temporal profiles using Bayesian inference models as implemented by Markov chain Monte Carlo sampling techniques. Dose and dose rate time profiles have a nonlinear temporal dependence, usually modeled as a nonlinear sigmoidal growth curve. These models lend themselves to the prediction of future doses given data from early in the event. The operational implementation of this methodology would utilize onboard instruments to mark the beginning of an event and onboard dosimeters to provide real-time dose and dose rate values as input to the empirical model. Similarly, solar particle event integral proton fluences demonstrate a nonlinear temporal dependence and may be modeled as a nonlinear sigmoidal growth curve. Predicting fluences rather than doses allows the forecaster to then calculate and predict the response function of choice. A larger sample of historical solar particle events was used for proton fluence prediction model development than previously for the dose and dose rate prediction efforts. In addition to quantitative forecasts, our models provide almost immediate qualitative classification of new events as significant versus insignificant, thus providing a tool to operators for making decisions concerning the commitment of forecasting resources. The justification for modeling fluence using non-linear sigmoidal growth curves and hierarchical models is examined, and the hypothesis that significant events (in terms of fluence) can be identified by fluence alone, early in the evolution of the event, is examined