In this project we will use Bayesian parameter estimation to enhance physics-based simulations of high-energy electron flux in the radiation belt. The goal is to be able to predict the fluxes of killer electron along a given satellite orbit.
We are studying advanced neural network architectures to forecast geomagnetic indexes, such as Dst or Kp. A particular emphasis is posed on the computational efficiency of the training phase, and on the robustness of the forecast, with respect to hiddent, long-term seasonality effects.
We are employing information theory tools to understand the physical coupling between solar wind and magnetosphere. This will help establish the relative importance of exogenous parameters, such as solar wind velocity, flux and interplanetary magnetic field, in predicting local flux enhancement or geomagnetic storm commencement in the radiation belts.
We are investigating how uncertainties on the definition of diffusion coefficients and boundary conditions affect the results of numerical simulations, in the standard quasi-linear framework. In particular, we intend to use a Bayesian approach, where the outcome of a simulation will be interpreted within a given confidence interval, as a posterior probability.
We are building an open platform where Space Weather modelers can share and compare their models.Our intent is to help the Space Weather community by making this website wiki-styled, non-commercial, and academic oriented. If you are interested in testing, comparing, and ultimately sharing your prediction model, that is the right venue for you!