The Disturbance storm time (Dst) index has been widely used as a proxy for the ring current intensity, and therefore as a measure of geomagnetic activity. It is derived by measurements from four ground magnetometers in the geomagnetic equatorial regions. We present a new model for predicting $Dst$ with a lead time between 1 and 6 hours. The model is first developed using a Gated Recurrent Unit (GRU) network that is trained using solar wind parameters. The uncertainty of the $Dst$ model is then estimated by using the ACCRUE method [Camporeale et al. 2021]. Finally, a multi-fidelity boosting method is developed in order to enhance the accuracy of the model and reduce its associated uncertainty. It is shown that the developed model can predict $Dst$ 6 hours ahead with a root-mean-square-error (RMSE) of 13.54 $\mathrm{nT}$. This is significantly better than the persistence model and a simple GRU model.

Abstract Suprathermal electrons in the near‐Earth plasma sheet are important for inner magnetosphere considerations. They are the source population for outer radiation belt electrons and they pose risks to geosynchronous satellites through their contribution to surface charging. We use empirical modeling to address relationships between solar driving parameters and plasma sheet electron flux. Using Time History of Events and Macroscale Interactions during Substorms, OMNI, and Flare Irradiance Spectral Model Version 2 data, we develop a neural network model to predict differential electron flux from 0.08 to 93 keV in the plasma sheet, at distances from 6 to 12 RE. Driving parameters include solar wind (SW) density and speed, interplanetary magnetic field (IMF) BZ and BY, solar extreme ultraviolet flux, IMF BZ ultra‐low frequency (ULF) wave power, SW‐magnetosphere coupling functions Pα1 and NXCF, and the 4‐hr time history of these parameters. Our model predicts overall plasma sheet electron flux variations with correlation coefficients of between 0.59 and 0.77, and median symmetric accuracy in the 41%–140% range (depending on energy). We find that short time‐scale electron flux variations are not reproduced using short time‐scale inputs. Using a recently published technique to extract information from neural networks, we determine the most important drivers impacting model prediction are VSW, VBS, and IMF BZ. SW‐magnetosphere coupling functions that include IMF clock angle, IMF BZ ULF wave power, and IMF BY have little impact in our model of electron flux in the near‐Earth plasma sheet. The new model, built directly on differential flux, outperforms an existing model that derives fluxes from plasma moments, with the performance improvement increasing with increasing energy.

The magnetospheric physics research community uses a broad array of quantitative data-model comparison methods (metrics) when conducting their research investigations. It is often the case, though, that any particular study will only use one or two metrics, with the two most common being Pearson correlation coefficient and root mean square error (RMSE). Because metrics are designed to test a specific aspect of the data-model relationship, limiting the comparison to only one or two metrics reduces the physical insights that can be gleaned from the analysis, restricting the possible findings from modeling studies. Additional physical insights can be obtained when many types of metrics are applied. We organize metrics into two primary groups: 1) fit performance metrics, often based on the data-model value difference; and 2) event detection metrics, which use a discrete event classification of data and model values determined by a specified threshold. In addition to these groups, there are several major categories of metrics based on the aspect of the data-model relationship that the metric assesses: 1) accuracy; 2) bias; 3) precision; 4) association; 5) and extremes. Another category is skill, which is a measure of any of these metrics against the performance of a reference model. These can be applied to a subset of either the data or the model values, known as reliability and discrimination assessments. In the context of magnetospheric physics examples, we discuss best practices for choosing metrics for particular studies.

The 400 worst‐case severe environments for surface charging detected at Los Alamos National Laboratory satellites during the years of 1990–2005 as binned by the definitions of four criteria developed by Matéo‐Vélez et al. (2018, https://doi.org/10.1002/2017sw001689) and the solar wind and Interplanetary Magnetic Field (IMF) parameters and geomagnetic activity indices are analyzed. The conducted analysis shows that only Auroral Electrojet/Auroral Lower index determines the highest risk for severe environments for surface charging to happen. The presence of a substorm with the southward turning pattern in IMF Bz indicates that the environment can be severe for surface charging to occur but this environment will not depend on whether a substorm was moderate or intense. No clear dependence on IMF Bz is found for risk to a severe environment to occur. Appearances of severe environments for surface charging do not necessarily require high values of Kp (Planetarische Kennziffer) and no storm is needed for such an event to be detected. Among solar wind parameters, solar wind velocity Vsw is directly related to the highest risk of severe environments, dependent on the Vsw value; and number density Nsw is of no importance. Two criteria for severe environment events based on the enhancements of low energy particle fluxes exhibit clearer dependencies on the solar wind and IMF parameters and geomagnetic activity indices with more distinct patterns in their time history.

The near-Earth plasma sheet is the source for electrons in the inner magnetosphere. The coupling between the solar wind and the near-Earth plasma sheet is dominated by non-linear processes, making any relationship difficult to infer. We report on the development of a neural network to capture the non-linear behavior between solar wind variations and the response of energetic electron flux in the plasma sheet. To train the neural network algorithm, we developed a data set with inputs from solar wind monitoring spacecraft. The targets come from three probes of the Time History of Events and Macroscale Interactions during Substorms mission as the spacecraft traversed the plasma sheet from years 2008–2019. Preliminary findings during the development of the neural network model show that tuning input parameters based on previously known physical properties is conducive to improving model performance.