On the generation of probabilistic forecasts from deterministic models

Abstract:

Space Weather forecasts need to be probabilistic, in order to be used in a decision-making scenario. However, most of the model used by the space weather community are completely deterministic, and provide a single-point estimate of the quantity of interest. Generating a probabilistic forecast from single-point predictions is a non-trivial task. Often, an ensemble of models or predictions is used for deriving confidence errors, but this approach is expensive and error-prone. In this work we propose a simple method to derive probabilistic forecast from single-point predictions without the need of generating an ensemble, using machine learning. Practical example applied to radiation belt plasma parameters are shown.