Summary: Volterra network modeling of the nonlinear finite-impulse
reponse of the radiation belt flux
, I.A. Daglis
, A. Anastasiadis
and D. Vassiliadis
Institute for Space Applications and Remote Sensing(ISARS), National Observatory of Athens(NOA), Metaxa
and Vasillis Pavlou Street, Penteli, Athens 15236, Greece.
Department of Physics, Hodges Hall, PO Box 6315, West Virginia University, Morgantown, WV 26506-6315,
Abstract. We show how a general class of spatio-temporal nonlinear impulse-response forecast networks (Volterra networks)
can be constructed from a taxonomy of nonlinear autoregressive integrated moving average with exogenous inputs (NAR-
MAX) input-output equations, and used to model the evolution of energetic particle f uxes in the Van Allen radiation belts.
We present initial results for the nonlinear response of the radiation belts to conditions a month earlier. The essential fea-
tures of spatio-temporal observations are recovered with the model echoing the results of state space models and linear f nite
impulse-response models whereby the strongest coupling peak occurs in the preceding 1-2 days. It appears that such networks
hold promise for the development of accurate and fully data-driven space weather modelling, monitoring and forecast tools.
Keywords: Radiation belts, input-output models, nonlinear neural networks