Assessing uncertainty in space weather forecasting using quantile regression and complex nonlinear systems identification techniques

Space weather forecasting is of global interest, and its importance is well established in research community and recognized by government, industries and stockholders. Over the past years, many types of predictive models have been developed in the literature. There is a general agreement that forecasting models should not only provide point prediction but also inform the uncertainty associated with the prediction. This study presents a novel method bases on quantile regression and complex dynamic modelling for measuring uncertainties in space weather forecasting. The approach is implemented using Quantile regression and Nonlinear AutoRegressive Moving Average with Exogenous inputs (NARMAX) methods (for short the approach is called Q-NARMAX). The method is applied to Disturbance storm index (Dst) observations to examine its interpretability and capability for uncertainty analysis. Results show that the proposed Q-NARX model can produce excellent predictions of the Dst index, and meanwhile provides a measure for assessing the uncertainty in the forecast. The innovative integration of quantile regression, complex dynamic modelling and nonlinear system identification techniques enables the proposed work to have following attractive advantages and properties: 1) it can produce excellent prediction accuracy for space weather forecasting, 2) it uses transparent models to approximate (represent) black-box systems, enabling to interpret the dependent relationship between space weather indices (system outputs) and their drivers (system inputs), and 3) more importantly, it allows for uncertainty assessment and analysis of models and forecasts.