Modeling and prediction of electron fluxes in GEO with NARMAX approach using data set with missing data points

NARMAX approach, originally developed for identifying models for complex nonlinear dynamic systems, has been used for space system and environment modelling and space weather forecasting for more than a decade. In this study, the NARMAX approach is employed to build predictive models for 1.5 MeV electron fluxes in geostationary Earth orbit (GEO). A challenge in modelling space weather is that there are many gaps due to missing data points in the available dataset for modelling, and we usually end up building the model with some imputed data to fill the gaps. The challenge is addressed in this paper. We propose to segment the data set with missing data point and reformulate the task as a modelling problem from multiple data sets and then introduce an effective model identification scheme that allows us to identify a single model directly from multiple datasets without resorting to any data interpolation or imputation techniques. The models are identified using two different model term selection methods and the prediction performance between models will then be compared to providing some guidelines for choosing modelling methods. We will also discuss best practices, limitations and caveats to consider when building prediction models from a given/or an available data set with missing data points.