NSWP: Space Weather Prediction Using Hybrid Physics/Black-Box Methods

Utah State University, Logan UT

Investigators

Abstract

This project will combine a relatively low-order (N=8), physics-based dynamical model of the Earth's magnetosphere (WINDMI) with a nonlinear black box model based on multiresolution wavelet decomposition. The goal is to have the hybrid model take into account physics that is missing from the low-dimension analytic model. Three different approaches will be investigated. The first suggested configuration involves replacing significant portions of the low-order model that are already dedicated to calculating plasma pressure with a single black box model. The second proposed configuration would replace only a relatively small nonlinear unloading function embedded in the previously mentioned pressure calculation with a black box model. The last configuration will assume the physics in WINDMI are mostly correct, and use a nonlinear black box model to correct for residual errors. This hybrid model will be trained using a genetic nonlinear optimization algorithm to predict the AL and Dst indices using ACE solar wind data as input. The result will be a model of the magnetosphere that can predict the Dst and AL magnetic indices with high accuracy.