Prediction of MeV electron fluxes throughout the outer radiation belt using multivariate autoregressive models

Abstract

The Van Allen radiation belts surrounding the Earth are filled with MeV-energy electrons. This region poses ionizing radiation risks for spacecraft that operate within it, including those in geostationary orbit (GEO) and medium Earth orbit. To provide alerts of electron flux enhancements, 16 prediction models of the electron log-flux variation throughout the equatorial outer radiation belt as a function of the McIlwain L parameter were developed using the multivariate autoregressive model and Kalman filter. Measurements of omnidirectional 2.3 MeV electron flux from the Van Allen Probes mission as well as >2 MeV electrons from the GOES 15 spacecraft were used as the predictors. Model explanatory parameters were selected from solar wind parameters, the electron log-flux at GEO, and geomagnetic indices. For the innermost region of the outer radiation belt, the electron flux is best predicted by using the Dst index as the sole input parameter. For the central to outermost regions, at L ≧ 4.8 and L ≧ 5.6, the electron flux is predicted most accurately by including also the solar wind velocity and then the dynamic pressure, respectively. The Dst index is the best overall single parameter for predicting at 3 ≦ L ≦ 6, while for the GEO flux prediction, the K P index is better than Dst. A test calculation demonstrates that the model successfully predicts the timing and location of the flux maximum as much as 2 days in advance and that the electron flux decreases faster with time at higher L values, both model features consistent with the actually observed behavior.

The Challenge of Machine Learning in Space Weather: Nowcasting and Forecasting

“…Sakaguchi et al () and Sakaguchi et al () have proposed multivariate autoregressive models based on Kalman filters to forecast GEO and Medium Earth Orbit energetic electrons (see also Rigler et al, ). A cross‐correlation analysis was carried out to identify physical drivers, for a range of time lags and different L shells.…”

Relativistic electrons of the outer radiation belt and methods of their forecast (review)

“…Упомянем также экзотические модели, привле-кающие к объяснению динамики внешнего радиа-ционного пояса релятивистские электроны юпите-рианского происхождения ] и элек-троны солнечного ветра, проникающие через касп [Sheldon et al, 1998]. …”

Relativistic electrons of the outer radiation belt and methods of their forecast (review)

“…Unlike other studies relying on the multivariate statistical analysis [Simms et al, 2014[Simms et al, , 2015[Simms et al, , 2016Sakaguchi et al, 2015], in this case, there is an additional element in the forecast procedure, which allows us to revise regression coefficients obtained from the preliminary analysis of regression relationships . [Potapov et al, 2014] As predictors we take daily averages of the following parameters: 1) solar wind velocity (V sw ); 2) amplitude of Pc4-5 ULF oscillations as inferred from ground-based (A gr ) and solar wind (A sw ) observations; 3) horizontal component of the magnetic field at geosynchronous orbit (H p ); 4) density of solar wind protons (N p ); 5) solar wind dynamic pressure (p dyn ); 6) interplanetary electric field (E y ); 7) maximum daytime flux J 600 of seed electrons with energy of about 600 keV, measured at geosynchronous orbit.…”

“…To date, literature has offered dozens of methods for forecasting electron fluxes in the outer radiation belt [Baker et al, 1990;Ling et al, 2010;Weigel et al, 2003;Ukhorskiy et al, 2004;Li et al, 2001;Perry et al, 2010;Sakaguchi et al, 2015;Efitorov et al, 2016;. They differ in: 1) selection of predictors; 2) forecasting methods; 3) lead time.…”

“…As forecasting A.S. Potapov 62 methods researchers most often employ linear filters [Baker et al, 1990], formed on the basis of the multivariate analysis [Simms et al, 2014[Simms et al, , 2016, in particular using the Kalman filter [Sakaguchi et al, 2015], as well as nonlinear methods such as neural networks Wide et al, 2016] and nonlinear autoregressive moving average modeling (NARMAX) [Balikhin et al, 2011]. By the lead time, forecasts are divided into short-, medium-, and long-term.…”