Recommendations for Next‐Generation Ground Magnetic Perturbation Validation

Abstract

Data‐model validation of ground magnetic perturbation forecasts, specifically of the time rate of change of surface magnetic field, dB/dt, is a critical task for model development and for mitigation of geomagnetically induced current effects. While a current, community‐accepted standard for dB/dt validation exists (Pulkkinen et al., 2013), it has several limitations that prevent more complete understanding of model capability. This work presents recommendations from the International Forum for Space Weather Capabilities Assessment Ground Magnetic Perturbation Working Team for creating a next‐generation validation suite. Four recommendations are made to address the existing suite: greatly expand the number of ground observatories used, expand the number of events included in the suite from six to eight, generate metrics as a function of magnetic local time, and generate metrics as a function of activity type. For each of these, implementation details are explored. Limitations and future considerations are also discussed.

Comparison of Deep Learning Techniques to Model Connections Between Solar Wind and Ground Magnetic Perturbations

“…The two types of models we trained are used to predict B N and B E during storms that occurred on 5 August 2011 and 17 March 2015. The selection was based on the recommendations from the Pulkkinen-Welling validation set for ground magnetic perturbations (Pulkkinen et al, 2013;Welling et al, 2018). These two storms were selected because they are outside of the time range used to train and validate the models and because they correspond to two very different years in terms of geomagnetic activity, 2011 being on the minimum-ascending part of the solar cycle, and 2015 corresponding to the solar cycle maximum.…”

“…For this study, we use solar wind and interplanetary magnetic field (IMF) data obtained from the OMNIWeb dataset available through NASA's Space Physics Data Facility from 1995 through 2010 for the purpose of training and validation of the models, and from 2011 and 2015 for testing. These 2 years were selected for testing because they include storms from the Pulkkinen-Welling validation set for ground magnetic perturbations (Pulkkinen et al, 2013;Welling et al, 2018). Baseline-removed ground magnetometer data from OTT has been obtained from SuperMag (Gjerloev, 2012).…”

“…In this study we present a comparison of models using a feed-forward artificial neural network (ANN) with a builtin time dependence and a LSTM neural network to predict the ground magnetic field north and east components (and therefore the perturbations dB H /dt) at the mid-latitude ground magnetometer station located in Ottawa (OTT). We then discuss the performance of the models by using two of the benchmark geomagnetic storms suggested by Welling et al (2018) and Pulkkinen et al (2013). Finally, we discuss several model variations that were implemented during the course of this study to determine possible improvements to the performance of the models.…”

Challenges and Opportunities in Magnetospheric Space Weather Prediction

“…For the electric power industry, for example, appropriate metrics may correspond to either geoelectric field (Pulkkinen et al, ) or geomagnetically induced current (Campanyà et al, ; Marsal & Torta, ), rather than the Kp index. Recent work toward this goal has been published by working teams of the International Forum on Space Weather Capabilities Assessment (M. W. Liemohn et al, ; Welling et al, ; Scherliess et al, ; R. Robinson et al, ).…”

Quantifying the Performance of Geomagnetically Induced Current Models

“…When it comes to evaluate the performance of a given model, one is faced with the question of what is the most suitable method to compare it with real observations (in the case of GIC in power grids, the currents measured directly in the transformer neutrals by using Hall effect transducers; e.g., Torta et al, ; or those indirectly obtained by differential magnetometry under power lines; e.g., Matandirotya et al, ). The broader community studying SW impacts is actively discussing how to evaluate model performance across a variety of prediction domains (e.g., Bruinsma et al, ; Liemohn, Ganushkina, et al, , Liemohn, McCollough, et al, ; Morley, Brito, et al, , Morley, Welling, et al, ; Shim et al, ; Welling et al, ; Wintoft & Wik, ). This includes predictions of the ground magnetic perturbations leading to GIC (e.g., Pulkkinen et al, ; Welling et al, ).…”