Conductance Model for Extreme Events : Impact of Auroral Conductance on Space Weather Forecasts
Agnit Mukhopadhyay, Daniel T Welling, Michael W Liemohn, Aaron J Ridley, Shibaji Chakrabarty, Brian J Anderson
mmanuscript submitted to
Space Weather
Conductance Model for Extreme Events : Impact ofAuroral Conductance on Space Weather Forecasts
Agnit Mukhopadhyay , Daniel T. Welling , Michael W. Liemohn , Aaron J.Ridley , Shibaji Chakraborty , and Brian J. Anderson Climate and Space Sciences and Engineering Department, University of Michigan, Ann Arbor, MI, USA Department of Physics, University of Texas at Arlington, Arlington, TX, USA Department of Electrical and Computer Engineering, Virginia Polytechnic Institute and State University,Blacksburg, VA, USA Applied Physics Laboratory, Johns Hopkins University, Baltimore, MD, USA
Key Points: • An updated auroral conductance module is built for global models using nonlin-ear regression & empirical adjustments spanning extreme events. • Expanded dataset raises the ceiling of conductance values, impacting the polar cappotential, dB/dt & ∆ B predictions during extreme events. • Application of expanded model with empirical oval adjustments refines the con-ductance pattern, and drastically improves dB/dt predictions.
Corresponding author: Agnit Mukhopadhyay, [email protected] –1– a r X i v : . [ phy s i c s . s p ace - ph ] A ug anuscript submitted to Space Weather
Abstract
Ionospheric conductance is a crucial factor in regulating the closure of magnetosphericfield-aligned currents through the ionosphere as Hall and Pedersen currents. Despite itsimportance in predictive investigations of the magnetosphere - ionosphere coupling, theestimation of ionospheric conductance in the auroral region is precarious in most globalfirst-principles based models. This impreciseness in estimating the auroral conductanceimpedes both our understanding and predictive capabilities of the magnetosphere-ionospheresystem during extreme space weather events. In this article, we address this concern, withthe development of an advanced Conductance Model for Extreme Events (CMEE) thatestimates the auroral conductance from field aligned current values. CMEE has been de-veloped using nonlinear regression over a year’s worth of one-minute resolution outputfrom assimilative maps, specifically including times of extreme driving of the solar wind-magnetosphere-ionosphere system. The model also includes provisions to enhance theconductance in the aurora using additional adjustments to refine the auroral oval. CMEEhas been incorporated within the Ridley Ionosphere Model (RIM) of the Space WeatherModeling Framework (SWMF) for usage in space weather simulations. This paper com-pares performance of CMEE against the existing conductance model in RIM, througha validation process for six space weather events. The performance analysis indicates over-all improvement in the ionospheric feedback to ground-based space weather forecasts.Specifically, the model is able to improve the prediction of ionospheric currents whichimpact the simulated dB/dt and ∆ B , resulting in substantial improvements in dB/dt predictive skill. Plain Language Summary
Electric currents generated in the Earth’s space environment due to its magneticinteraction with the Sun leads to charged particle deposition and closure of these cur-rents in the terrestrial upper atmosphere, especially in the high latitude auroral region.The enhancement in the electrical charge carrying capacity as a result of this process inthe Earth’s upper atmosphere, also known as the ionosphere, is challenging to estimatein most numerical simulations attempting to study the interactive dynamic and chem-ical processes in the near-Earth region. The inability to accurately estimate this quan-tity leads to underprediction of severe space weather events that can have adverse im-pacts on man-made technology like electrical power grids, railway and oil pipelines. Inthis study, we present a novel modeling approach to address this problem, and provideglobal simulations with a more accurate estimate on the electrical conductivity of theionosphere. Through this investigation, we show that the accurate measurement of thecharge carriers in the ionosphere using the new model causes substantial improvementsin the prediction of space weather on the ground, and significantly advances our under-standing of global dynamics causing ground-based space weather.
The interaction of the solar wind and the terrestrial magnetic field produces mag-netospheric current systems such as field aligned currents (FACs) which close throughthe conductive ionosphere, thereby allowing magnetospheric convection to eventuate (e.g.Axford & Hines, 1961; Dungey, 1963; Iijima & Potemra, 1976). For precise investigationsof the magnetospheric feedback on the ionosphere and vice versa, an accurate estimateof the ionospheric conductance is critical for realistic global modeling of the magneto-sphere, especially during space weather events (e.g. Merkine et al., 2003, Ridley et al.,2004, Merkin, Sharma, et al., 2005; Merkin, Milikh, et al., 2005, Liemohn et al., 2005).Two dominant sources contribute to the ionosphere’s enhanced but finite conductivity- solar extreme ultra-violet (EUV) flux on the dayside, and auroral precipitation in thepolar region predominantly on the nightside (Schunk & Nagy, 2009; Newell et al., 2009; –2–anuscript submitted to
Space Weather
Fuller-Rowell & Evans, 1987). Conductance due to solar EUV radiation is relatively wellunderstood through the use of radiative transfer (e.g. Chapman, 1931). The EUV fluxis accounted for in most modern modeling tools as a physics-based empirical functionof the solar zenith angle (e.g., Brekke & Moen 1993). Auroral electron and ion precip-itation, largely driven by magnetospheric processes, further ionizes neutrals and ions inthe ionosphere (e.g., Frahm et al., 1997; Ahn et al., 1998), and enhances the electricalconductivity in the high-latitude auroral regions (Robinson et al., 1987). Since auroralprecipitation of charged particles is directly related to variations in the intrinsic mag-netic field (e.g., Roederer, 1970), auroral conductance is an important quantity to pre-dict when investigating the ionosphere’s impact on the magnetosphere, and vice versa,during strong driving when the global magnetic field changes rapidly (e.g., Welling, 2019).Although several studies have examined the influence of the ionospheric conduc-tance on the global state of the magnetosphere, ionospheric dynamics and their couplednon-linear feedback system (e.g., Raeder et al., 2001; Ridley et al., 2001, 2004; Liemohnet al., 2005; Wiltberger et al., 2001, 2004; Zhang et al., 2015; Connor et al., 2016; Oz-turk et al., 2017), few studies have actually explored the contribution of conductance onspace weather forecasts (e.g. Hartinger et al., 2017), especially during extreme space weatherevents. This is very difficult to do with data, since measurements of the ionospheric con-ductance are notoriously inaccurate (Ohtani et al., 2014). Investigations using global mod-els such as Ridley et al. (2004) have indulged in the broad quantification of the conduc-tance due to EUV illumination and auroral precipitation. Studies such as Wiltberger etal. (2001), Zhang et al. (2015), Yu et al. (2016) and Wiltberger et al. (2017) addressedthis further by identifying the source and impact of various contributors to the auroralconductance. Additional evaluations by Perlongo et al. (2017) included the effect of au-roral precipitation due to the ring current using a kinetic ring current model coupled toan ionosphere-thermosphere model. Modeling efforts by Ahn et al. (1998), Newell et al.(2009), Korth et al. (2014) have estimated ionospheric auroral conductance through em-pirical relations, using global quantities like solar wind input, ground-based magnetic per-turbations and field aligned currents as inputs. The Robinson conductance model (Robin-son et al., 1987; Kaeppler et al., 2015) relating downward precipitating fluxes to auro-ral conductance is yet another prominent example of empirically-derived conductancefrom global magnetospheric quantities. Recently, Robinson et al. (2018) developed anempirical model using incoherent scatter radar measurements against AMPERE FACestimations, which spanned the St. Patrick’s Day Storm of 2015, an event studied ex-tensively for ionospheric disturbances (e.g., Le et al., 2016). In spite of its importance,the impact of auroral conductance during extreme events in global simulations has beenhard to determine, due to inaccuracies in conductance estimations within global mod-els, leading to possible underprediction of global quantities like cross polar cap poten-tial (e.g., Honkonen et al., 2013; Mukhopadhyay, 2017), field aligned currents (Ander-son et al., 2017), storm indices (Liemohn, McCollough, et al., 2018) and transient ground-based magnetic perturbations (Welling et al., 2018).With rising operational usage of first-principles-based geospace models in space weatherprediction, the need for accurate conductance models is even more necessary. Operationalforecasts of the near-Earth space environment using first-principles based global numer-ical frameworks (e.g., T´oth et al., 2005), combining global magnetohydrodynamic (MHD)models (e.g., Powell et al., 1999; Raeder et al., 2001) with suitable inner magnetosphericmodels (e.g., De Zeeuw et al., 2004) and ionospheric models (e.g., Ridley & Liemohn,2002; Wiltberger et al., 2004), have been in use for space weather prediction (Liemohn,Ganushkina, et al., 2018) since the end of the GEM Challenge of 2008-09 (Pulkkinen etal., 2011, 2013, Rastaetter et al. 2013). The procedural assessment specifically presentedin Pulkkinen et al. (2013) (hereinafter referred to as
Pulkkinen2013 ) to investigate pre-dictive skill of global first-principles-based models in predicting ground-based magneticperturbations dB/dt , initiated the transition of model usage toward operational predic-tion at the NOAA Space Weather Prediction Center (SWPC). Several investigations, since –3–anuscript submitted to
Space Weather then, have further reviewed and systematically addressed the results from this effort, andhave suggested rectifications to improve predictive skill (e.g., Honkonen et al., 2013; Glo-cer et al., 2016; Anderson et al., 2017; Mukhopadhyay, 2017; Liemohn, Ganushkina, etal., 2018; Liemohn, McCollough, et al., 2018; Welling et al., 2018). In particular, the studyby Welling et al. (2017) indicated inherent deficiencies in auroral conductance modelsused in global models that inhibited them from estimating conductance accurately dur-ing extreme space weather events. The study concluded that the inability of global mod-els to estimate the ionospheric conductance accurately during extreme events led to un-derprediction of dB/dt .A key conclusion in the study by Welling et al. (2017) (hereinafter referred to as
Welling2017 ) questions the dataset used in estimating a geospace model’s auroral con-ductance during extreme weather, and hypothesizes that the inclusion of information froma larger dataset, including sufficient coverage of extreme events, may lead to improve-ments in a model’s space weather predictive metrics during extreme events. The studyfalls short of addressing supplementary effects due to the auroral oval’s pattern estima-tion in aforementioned models, and the acute effect such a pattern may have on predic-tive skill. In this paper, we describe the development and validation of an updated em-pirical auroral conductance model, specifically including data that spans several extremeevents, which addresses the concerns raised in
Welling2017 . We use this conductance modelwithin the geospace variant of the Space Weather Modeling Framework (SWMF; T´othet al., 2005, 2012), identical to the version used operationally at the NOAA Space WeatherPrediction Center for space weather forecasting, to investigate the effect of this enhancedconductance model on space weather predictions, and compare these results to the already-existing conductance model within the SWMF. We additionally study the effect of ad-justing the pattern of the auroral oval using empirical enhancements based on field alignedcurrent strength, to alter the model’s space weather predictions. As a result, in this ar-ticle, we investigate three major science questions:1. Addressing
Welling2017 : Does expanding the dataset used to create the initialconductance model help improve space weather predictions?2. How significant is the improvement in the space weather predictions due to theenhanced auroral oval adjustment parameters?3. Can the combination of the expanded dataset and an auroral oval enhancementcause significant improvement in the global model’s space weather prediction?In order to address the aforementioned questions, a new C onductance M odel for E xtreme E vents (CMEE) has been developed. CMEE is based on the SWMF’s empirical auro-ral conductance model, which uses an inverse-exponential relation to estimate the con-ductance, and employs an empirically-driven auroral oval adjustment to enhance con-ductance in regions of strong FACs. A key difference in CMEE, however, is in the datasetit was developed from: CMEE uses one whole year of AMIE data to estimate its con-ductance. Compared to the old model which was derived from the relatively quiet monthof January 1997, minute-data from the whole year of 2003 was utilized to develop CMEE.This included some of the most extreme geospace events ever observed (Cid et al., 2015).In addition to an enlarged training dataset, the value of the empirical coefficients in CMEEare deduced using a non-linear fitting algorithm with suitable extreme boundary con-ditions that minimizes the absolute error and maximizes the prediction efficiency. Theglobal model configurations used and the science questions addressed in this study, andthe subsequent results from this study are described in Sections 2 and 3 respectively, whilethe algorithm used to develop the advance conductance model and the auroral oval ad-justment module have been described in Section 2.2. –4–anuscript submitted to Space Weather
Figure 1.
Component layout of the geospace version of the SWMF, same as the layout in
Pulkkinen2013 , used in this study to investigate the role of auroral conductance in space weatherprediction.
The SWMF is a flexible framework that executes, synchronizes and couples manyotherwise independent models together as one. It has performed favorably in predictivemetric challenges and investigations (e.g.,
Pulkkinen2013 ; Honkonen et al., 2013; Mukhopad-hyay, 2017; Welling et al., 2017; Liemohn, McCollough, et al., 2018), contains an easily-modifiable empirical conductance model in the ionospheric electrodynamics module (Ri-dley et al., 2004), and is capable of calculating perturbations to the magnetic field (∆ B )by applying Biot-Savart integrals across its domain to estimate magnetometer values vir-tually (Yu et al., 2010). For this study, we have used the SWMF with three physical mod-ules activated (Figure 1; details below). Identical to the study conducted by Pulkkinen2013 ,the SWMF’s geospace version was configured to use three components: Global Magne-tosphere (GM), Inner Magnetosphere (IM), and Ionospheric Electrodynamics (IE).The GM module uses the Block Adaptive Tree Solar-Wind Roe Upwind Scheme(BATS-R-US, Powell et al., 1999; Gombosi et al., 2003) model which solves for the idealnon-relativistic magnetohydrodynamic (MHD) equations in the magnetosphere with aninner boundary at ∼ . R E ).The computational domain for geospace sim-ulations of BATS-R-US extends from 32 R E upstream to 224 R E downstream in the x di-rection and 128 R E in the y and z coordinates (GSM). The key feature of BATS-R-USis its flexible, block-adaptive Cartesian grid that reserves the highest resolution to re-gions of interest, ensuring the best combination of performance and accuracy.The IM region is characterized by closed magnetic field lines and particles of keVenergies. This module uses Rice Convection Model (RCM; Wolf et al., 1982). RCM solvesfor the bounce averaged and isotropic but energy resolved particle distribution of elec-trons and various ions. RCM receives flux tube volumes from BATS-R-US and returnsthe pressure and density values to correct those calculated within GM (De Zeeuw et al.,2004). It receives the ionospheric electric potential from the 2-dimensional IE module. –5–anuscript submitted to Space Weather
The density and temperature initial and boundary values are computed from the GMsolution.The IE component calculates height integrated ionospheric quantities at an alti-tude of about 110 km. To do so, it receives field aligned currents (FACs) from GM anduses the Ridley Ionosphere Model (RIM, Ridley et al. 2001; Ridley & Liemohn 2002; Ri-dley et al. 2004), a finite-difference Poisson solver, to calculate the electric potential andhorizontal currents using a prescribed but dynamic conductance pattern. The modulemaps FACs at 3.5 Earth radii ( R E ) over a two dimensional ionospheric domain, solvesfor the resulting potential using Ohm’s Law (Goodman, 1995), and returns this valueto GM and IM. The functioning of and developments to the ionospheric conductance modelof RIM are the key features of this article, and are discussed in detail in Section 2.2, alongwith the development of a more advanced empirical conductance model, CMEE, as a re-placement to the aforementioned model. Figure 2.
X-Z cuts showing cell sizes in the two MHD grids (reproduced from Haiducek et al.,2017). (Left) The grid used for the
SWPC configuration (minimum cell size of 0.25 R E ). (Right)The higher-resolution grid used for the Hi-Res SWPC configuration (minimum cell size of 0.125 R E ) In order to simulate a given event, we drive the model using solar wind velocity,magnetic field, density, and temperature, which are used to specify the upstream bound-ary condition of BATS-R-US. The only other input parameter is F10.7 flux, which is usedby IE in computing the dayside EUV-driven ionospheric conductivity (Moen & Brekke,1993; Ridley et al., 2004). Simulation parameters have been kept similar to
Pulkkinen2013 ,throughout the study; the model input conditions and parameters are not tailored to in-dividual events. The same solar wind values derived in
Pulkkinen2013 from instrumentsonboard the Advanced Composition Explorer (ACE) satellite were used to drive simu-lations in the present study. For this study, we have simulated the events using two dif-ferent resolutions of BATS-R-US :
SWPC and
Hi-Res SWPC (see Figure 2). The
SWPC configuration is nearly identical to the
Pulkkinen2013 study, and is used operationallyby the Space Weather Prediction Center (SWPC). This grid (Figure 2, left) has cell sizesranging from 8 R E in the distant tail to 0.25 R E at the inner boundary, a 16 R E diam-eter cube surrounding the Earth, and contains around 1 million cells. The other config-uration, Hi-Res SWPC , is similar to the previous configuration but uses a higher-resolutiongrid (among other modifications), to help resolve field aligned currents at the spatial in-ner boundary. The cell size of this grid (Figure 2, right) varies from 8 R E in the tail to –6–anuscript submitted to Space Weather R E near the Earth, and contains ∼ ×
181 cell configuration in the IE domain, with a 2 degree cadence in both latitude andlongitude. For a detailed description of the above configurations, please refer to Welling& Ridley (2010) and Haiducek et al. (2017).
For Ohm’s Law to be solved within IE, knowledge of the ionospheric conductancetensor must be known a priori (e.g., Goodman, 1995). Within RIM, the legacy code es-timating the ionospheric conductance (Ridley et al., 2004) distinguishes two dominantsources of ionospheric conductance: solar EUV conductance on the dayside, and the au-roral precipitative conductance in the polar regions. Supplementary sources of conduc-tance, like nightside ”starlight” conductance, seasonal dependencies and polar rain, areadded as either functions of the dominant sources of conductance, solar zenith angle orscalar constants. The solar EUV component to the conductance is dependent on the ab-sorption and ion production function of the atmosphere as a function of the solar zenithangle, and is therefore straightforward to estimate using radiometry; the model describedin Moen & Brekke (1993) is used to estimate this component of the conductance in mostglobal models (e.g. Raeder et al., 2001 Wiltberger et al., 2004), including RIM. The con-ductance due to ion and electron precipitation in the auroral region is harder to predict,as this would require the precise knowledge of the charged particle distribution in themagnetosphere. While a physics-based approach to precipitation has been applied in sev-eral global models (e.g. Raeder et al., 2001, Zhang et al., 2015, Yu et al., 2016, Perlongoet al., 2017) using kinetic theory (e.g. Knight, 1973), RIM uses a different and simplerapproach to estimate the auroral conductance.
The auroral conductance module in RIM (briefly described in Ridley et al., 2004),hereinafter referred to as the Ridley Legacy Model (RLM), uses the magnitude and di-rection of modelled FACs to empirically determine the auroral conductance. This is sim-ilar to existing statistical models constructed using FACs to predict and examine pre-cipitation in the auroral ionosphere (e.g. Ahn et al., 1998, Korth et al., 2014, Carter etal., 2016, Robinson et al., 2018). While the numerical domain of RIM spans the entireionosphere, the RLM domain is considerably limited, spanning from the magnetic poleto magnetic latitude of 60 ◦ for all magnetic local times (MLT). The auroral conductanceat a given magnetic latitude and MLT is assumed to have the form:Σ HorP = A − A e − A | J || | (1)where Σ HorP denotes the auroral Hall or Pedersen Conductance in the ionosphere (insiemens), J || denotes the field aligned current density (in µA/m ), and A , A (in siemens)and A (in m/µA − / ) are fitting coefficients dependent on location. Note that this in-verse exponential relation is different from the one mentioned in Ridley et al. (2004); thiswas a typographical error and the actual relation is given by Equation 1.The empirical coefficients are the result of fitting based off of conductance and field-aligned current maps derived from assimilative maps of ionospheric electrodynamics (AMIE;Richmond & Kamide, 1988; Kihn & Ridley, 2005) for the month of January 1997 (Boon-siriseth et al., 2001), using ground magnetic perturbations from ∼
150 ground-based mag-netometers. AMIE derives the auroral conductance using the formulation in Ahn et al.(1998) and Lu et al. (1997), which relate ground-based magnetic perturbations to theHall and Pedersen conductance, and FACs. The exact parameters and version of AMIEused in the development of RLM, with further information about the datasets used havebeen described in detail in Kihn & Ridley (2005). The month of runs encompasses ∼ , –7–anuscript submitted to Space Weather ral oval adjustments were applied to constrain and enhance the conductance in regionsof strong FAC driving.
The conductance pattern in RLM tends to produce broad regions of high conduc-tance that are discontinuous between regions of strong FACs. To improve upon this, anadjustment to the conductance pattern is applied to the estimated pattern described above.The purpose of this is to create a channel for electrojets to form in the model and to im-prove on the overall electrodynamic result. Though this feature has been implementedin RLM for over a decade, this work is the first to formally describe it and evaluate itsimpact.The algorithm for this adjustment starts by estimating the location of the auro-ral oval. The location of the oval is updated at each simulation timestep of the ionosphere.Across all local time values ( φ ) in the model’s grid, the geomagnetic co-latitude of themaximum upward FAC at that local time slice ( J max ( φ )) is obtained. The result is θ ( φ ),or co-latitude as a function of local time. The mean co-latitude, θ mean , weighted by J max ( φ ),is then obtained as follows: θ mean = (cid:80) θ ( φ ) J max ( φ ) (cid:80) J max ( φ ) (2)A day-night shift in the center of the oval is calculated using the co-latitudes of J max ( φ )at noon and midnight:∆ θ = J noon × ( θ noon − θ mean ) − J midnight × ( θ midnight − θ mean ) J noon + J midnight (3)Using these values, the location of the auroral oval is modeled as follows: θ ( φ ) aurora = θ mean + ∆ θ cos( φ ) (4)With the oval location set, an adjustment is applied to the conductance values aboutthe oval by adjusting the fitting coefficients, A and A : A ,adj = A e − d W (5) A ,adj = A − ( A − A ) e − d W (6)...where, for each line of constant local time, d is the co-latitude distance from the oval’slocus and W is the width of the oval (default is 2.5 ◦ ). A baseline conductance about theoval is also applied to avoid nonphysical solutions in regions of low FACs:Σ baseline = 1 . × (Σ HorP + ke − d W ) (7)where 1.7 is a multiplier meant to amplify the value of the conductance, and k is a con-stant derived from the aggregate value of the AMIE-derived auroral conductance in re-gions of high precipitation (magnetic latitude ∈ [65 ◦ , ◦ ]). The 1.7 multiplier is a legacyvalue and was chosen for robustness and stability of dB/dt results. In this study, the valueof k was found to be 7.5 siemens for Hall conductance, and 5 siemens for Pedersen con-ductance from the AMIE dataset. The net result of this adjustment is that at each timestep,about the oval, the range of possible conductance values is dynamically narrowed andenhanced, and a coherent, sharper auroral conductance pattern arises. Based on the same formulation as RLM, CMEE was developed using a larger datasetin order to include information during intense space weather events (
Dst < − nT ). –8–anuscript submitted to Space Weather
For this model, minute-resolution data from AMIE for the whole year of 2003 were uti-lized to estimate the new fitting coefficients. For consistency, the same version of AMIE(Kihn & Ridley, 2005) used in the development of RLM has been used for the develop-ment of CMEE. The use of a year’s worth of minute-data significantly increased the model’sbase dataset from ∼ ,
000 2D maps used in RLM, to over ∼ ,
000 2D maps usedin the present study. In addition, the year of 2003 included several intense space weatherevents. Specifically, the latter half of the year saw some of the largest geomagnetic stormsever recorded by mankind (e.g. Cid et al., 2015; Doherty et al., 2004 ), while January1997 (the month off of which RLM is based) hardly saw any event with a
Dst ≤ -100nT. In addition to this, the value of the empirical coefficients in CMEE are deduced as-suming the same empirical relationship between upward or downward FACs with Halland Pedersen Conductance, as given by Equation 1. However, unlike RLM which esti-mates the fitting using equal weighting, the new fitting has been designed using a novelnonlinear regression algorithm which imposes sufficient boundary conditions to ensurethat the fitted curve extends to these extreme values and is not just limited to the ag-gregate value of conductance. This was done by basing the max endpoints of the fittingson the 90% percentile of the FAC values.Figure 3 (a) presents a representative line plot of Equation 1, and demarcates theconductance vs FAC space into bounded regions designed to estimate fitting coefficients.The regression algorithm of CMEE classifies FAC data into low and high magnitude bins,separately for upward and downward FACs. The bin boundary for low magnitude FACs,including zero FACs was based on the approximate order of low magnitude FAC den-sity, where asymptotic behavior of conductance values is prevalent and a median valuecould be found. The median value of the conductance populations in this FAC bin is theminima of the curve ( A − A ). For the low FAC case, setting the bin boundary at ± − µ A/m for both upward and downward field aligned currents at all locations led to optimum re-sults. To deduce the conductance maxima as a constant asymptotic value, the FAC datasetwas binned into 10 discrete bins with respect to the absolute value of FAC, and the me-dian value of conductance in the bin with the highest FAC values (10th bin) was definedas A . A Levenburg-Marquadt (e.g. Pujol, 2007) type bounded least-squares method wasused to estimate the non-linear fitting coefficient A . The fitting error was defined as thearithmetic mean of the median absolute percentage error (MAPE) and the median sym-metric efficiency ( ξ ) ratio of the data, as defined in Morley et al. (2018). In order to avoidnonphysical solutions from the ionospheric solver due to large gradients (spikiness) inthe conductance values, a smoothing filter was applied on the coefficients. The filter wasbased on a Laplacian mesh smoothing algorithm (e.g. Herrmann, 1976), commonly usedin image processing (Yagou et al., 2002) and mesh refinement (Sorkine et al., 2004). Thefilter is applied such that at each node i , x i = x i if x i − XX ≤ λX if x i − XX > λ (8)where X = 1 N N (cid:88) j =1 x j (9)Here, λ is the prescribed difference, N is the number of adjacent vertices to node i , x j is the position of the j -th adjacent vertex and x i is the new position for node i . The pre-scribed difference, similarly defined as the relative difference, is kept at 10%.Figures 3(b) shows an example of the fitting using the regression algorithm men-tioned above over a map of Hall conductance and FAC distribution from AMIE, at thegeomagnetic latitude of 62 o and MLT 23 for upward FACs. Figure 3(c) compares the fit-ting function using CMEE’s regression with coefficients from RLM for the same geomag-netic location, but for both upward and downward FACs. The usage of a regression al- –9–anuscript submitted to Space Weather gorithm over a larger span of data shows visible differences in Figure 3(c), where CMEE,denoted in red, is able to push the max value of the conductance to better estimate thequantity during extreme driving. In addition, because of the usage of low FAC bins, themodel is also able to provide uniformity in conductance values when field aligned cur-rents are low and/or switch directions. This was previously not included in RLM, de-noted in blue in Figure 3(c), as the coefficient values were estimated using uniform weight-ing on a case-by-case basis separately for upward and downward FACs.
In order to evaluate CMEE’s predictive capabilities and address the science ques-tions mentioned in Section 1, we have simulated a range of space weather events listedin Table 1(a) using variations of the auroral conductance model within the SWMF forcomparisons against observations. Since it is a de-facto standard in the space weathercommunity, the present investigation chose to simulate the same events listed in Table1 of the
Pulkkinen2013 study. Simulation of these events was administered for the tworesolutions described in Section 2.1, and using four different variations of the conduc-tance model :-1. Using only the empirical coefficients of RLM to specify the aurora,2. Using only the empirical coefficients of CMEE to specify the aurora,3. Adjusting RLM estimates with the additional enhancements in the auroral oval,and4. Adjusting CMEE estimates with the additional enhancements in the auroral oval.Table 1(b) lists the 8 sets of simulations resulting from the above combination.The study uses data from satellite in-situ measurements and ground-based obser-vations for comparisons against model results. Cross polar cap potential (CPCP) fromthe model variants was compared against values obtained via the AMIE model and ob-servations from the Super Dual Auroral Radar Network (SuperDARN; e.g. Khachikjanet al., 2008). Since AMIE has a tendency to overpredict CPCP (e.g. Gao, 2012), obser-vations from the SuperDARN were also used to provide a range to the CPCP estimates.Integrated field aligned currents derived from observations by the Active Magnetosphereand Planetary Electrodynamics Response Experiment (AMPERE) mission (Andersonet al., 2014; Waters et al., 2020), estimated using the methodology in Anderson et al.(2017), were used to compare modeled values of FACs. In addition, magnetometer ob-servations from the 12 magnetometer stations listed in Table 2 of the
Pulkkinen2013 studywere used to evaluate the predicted ground-based magnetic perturbation ∆ B and its tem-poral variant dB/dt .Using a similar approach as Pulkkinen2013 , a binary event analysis (e.g. Jolliffe& Stephenson, 2012; Wilks, 2011) was used to construct a set of relevant performancemetrics. An event is defined as the absolute value of a parameter-in-question (any phys-ical quantity like dB/dt ) exceeding a predetermined event threshold at any time withina comparison window t f . For each such window, four outcomes are possible: ”Hit” orTrue Positive (TP; event is observed, and also predicted), ”False Alarm” or False Pos-itive (FP; event is not observed, but predicted by model), ”Miss” or False Negative (FN;event is observed, but not predicted), and ”Correct No Events” or True Negative (TN;event is not observed, and not predicted). Similar to Pulkkinen2013 , the analysis fore-cast window t f was selected to be 20 minutes. The combined results from all events listedin Table 1(a) for a given simulation set are divided into discrete events by the forecastwindow, creating a contingency table accounting for TPs, FPs, FNs and TNs for a spe-cific threshold. Unlike the Pulkkinen2013 study, this study chose to discretize the dB/dt into thresholds ranging from 0.1 nT/s to 1.7 nT/s at intervals of 0.1 nT/s, including thethresholds 0.3 nT/s, 0.7 nT/s, 1.1 nT/s and 1.5 nT/s which were used in the former study. –10–anuscript submitted to
Space Weather
In addition to dB/dt , the ∆ B values have been discretized using thresholds obtained fromT´oth et al. (2014) and Welling2017 , ranging from 75 to 400 nT at intervals of 25 nT wereused. Once the contingency tables were prepared for each simulation variation, a com-bination of performance metrics were applied to study improvements. The metrics usedin this study and their respective definitions are listed in Table 2. Amongst these met-rics, the top four are accuracy measures that help describe the improvement of individ-ual outcomes in a contingency table, while the bottom four metrics quantify the accu-racy of a prediction. The Probability of Detection (POD), also called the Positive Pre-diction Value, is the ratio of positive and negative results, and ranges from 0 to 1, with1 being a perfect score. The Probability of False Detection (POFD) is the ratio of missesagainst total negative results. POFD ranges from 0 to 1, with 0 being a perfect score.Along with the POD, these two ratios are accuracy measures of model discrimination.The False Alarm Ratio (FAR), also called False Positive Rate is the ratio between thenumber of negative events wrongly categorized as positive and the total number of ac-tual negative events (false negatives + true negatives). The Miss Ratio (MR) is definedas the ratio between the number of misses and the sum of hits & misses, describing theconditional probability of a negative test result given that the condition being looked foris present. Both FAR and MR range from 0 to 1, with 0 being a perfect score. These twometrics are a measure of model reliability. The Threat Score (TS), also known as Crit-ical Success Index is the ratio of all true positives against the sum of total number of oc-currences and false alarms. Due to its neglect of non-occurrences, this score is well suitedfor scoring predictions of rare events like extreme driving during space weather events.The F score, another measure of a test’s accuracy, is defined as the harmonic mean ofthe POD and the hit rate, given by (1 − M R ). Similar to the Threat Score, the F scorereaches its best value at 1 and worst at 0. The True Skill Score (TSS) or Hanssen-KuiperSkill Score (Hanssen & Kuipers, 1965) is a performance metric with values ranging from-1 to +1, with 0 representing no skill. The TSS is defined as the difference between thehit rate (given by 1 − M R ) and false alarm rate. Lastly, the Heidke Skill Score (HSS;Heidke, 1926) is a performance metric that measures the improvements in a model’s re-sults against random chance. Similarly to the TSS, the value of HSS ranges from -1 to+1, with 0 representing no skill. The HSS is popular in space weather forecasting, andhas been established as a suitable comparative metric in several space weather studies(Welling & Ridley, 2010,
Pulkkinen2013 , T´oth et al., 2014, Welling et al., 2018).
Figure 4 exhibits the variations in the pattern and magnitude of Hall conductancefor simulations using the low-res
SWPC configuration. Each dial-plot column displaysthe high latitude Hall conductance at different time instances from the simulation setsA, B, C and D respectively. The first row shows results from 04:33 UT on October 29,2003 : toward the beginning of Event 1, before the sudden commencement with the stormindex Kp less than 4. The second and third rows, titled Epoch 2 and Epoch 3, comparethe four sets at 06:20 UT and 06:46 UT on the same day during the sudden commence-ment and main phase of Event 1, when 4 ≤ Kp < Kp ≥ Kp throughout the event, along with the pre-dicted Kp from the four simulation variants with the background coloured by the mag-nitude of Kp - green for Kp <
4, yellow for 4 ≤ Kp <
8, and red for Kp ≥ –11–anuscript submitted to Space Weather (a) List of EventsEvent
Date and Time (b) List of SWMF Simulations
RLM Coeffs CMEE Coeffs RLM w OA CMEE w OA
SWPC
Set A Set B Set C Set D
Hi-Res SWPC
Set E Set F Set G Set H
RLM Coeffs - Empirical Coefficients of the Ridley Legacy Model
CMEE Coeffs - Empirical Coefficients of the Conductance Model for Extreme Events
RLM w OA - Ridley Legacy Model, with Auroral Oval Adjustments
CMEE w OA - Conductance Model for Extreme Events, with Auroral Oval Adjustments
Table 1. (a) List of space weather events used in this study to test and validate the differentconductance models. This is the same set of events used in
Pulkkinen2013 . (b) A tabular descrip-tion of all the simulations conducted for this study, binned by SWMF domain variations used:Each set of runs (denoted as ’SET × ’, where × is the alphabetic value designated) is a simulationof all space weather events listed in (a), using a particular variation of the auroral conductancemodel (columns) within a given configuration of the SWMF (rows). Performance Metric Acronym Mathematical Definition
Probability of Detection POD
T P ( T P + F P ) Probability of False Detection POFD
F N ( F N + T N ) False Alarm Ratio FAR
F P ( F P + T N ) Miss Ratio MR
F N ( T P + F N ) Threat Score TS
T P ( T P + F N + F P ) F Score F
T P (2 T P + F P + F N ) True Skill Score TSS
T PT P + F N − F PF P + T N = (1 − M R ) − F AR
Heidke Skill Score HSS T P × T N − F P × F N )((
T P + F P )( F P + T N )+(
T P + F N )( F N + T N )) Table 2.
List of performance metrics used in this study.–12–anuscript submitted to
Space Weather
Sets C & D illustrate how the adjustments intensify the conductance in regions of highfield aligned currents, mimicking discrete arcs. The difference in Sets C & D, while notso apparent in Epochs 1 and 2, are substantially distinct in Epoch 3, when Kp ≥ Kp , CMEE increases nightsideconductance and lowers dayside conductance. This is because CMEE coefficients, a byprod-uct of an increased dataset spanning seasonal changes in addition to being estimated us-ing a nonlinear regression algorithm, computes lower dayside conductance and highernightside conductance in comparison to the RLM coefficients. An unusual feature of us-ing FAC-directed empirical models is the emergence of islands of conductance during thepeak of the storm (Epoch 3). These discontinuities are reduced by the inital usage of thesmoothing function on the coefficients, and addition of a baseline value in the auroraloval region.Figure 5 compares integrated field aligned currents (iFACs) observations during Event5 by AMPERE, against estimates from SWMF. Events 5 and 6 were observed by AM-PERE, and compared to models in Anderson et al. (2017). The iFACs were estimatedsimilarly to Anderson et al. (2017) and were used to compare the effect of dataset ex-pansion in the top panel (a), the impact of oval adjustments in the middle panel (b), andthe combined influence both in the bottom panel (c). In each of these panels, we com-pare the low resolution SWPC configuration of the SWMF simulations (Sets A, B, Cand D) with the
Hi-Res SWPC configuration simulations (Sets E, F, G and H) to vi-sualize the impact of conductance on the input conditions to IE. While minor variationsare caused by the usage of different conductance models, no significant changes are ob-served either by using the CMEE coefficients or by adjusting the auroral oval. Instead,the results show the
Hi-Res SWPC simulations being able to better capture the mag-nitude and dynamics of the iFACs than the
SWPC configurations. This is in agreementwith results from the study of Ridley et al. (2010) who investigated the impact of res-olution on ionospheric quantities like FACs, especially with respect to variation in val-ues as we change numerical resolution. While there are definite changes in the FACs andiFAC values due to the different auroral models, the increased resolution helps to cap-ture more of the FACs, dramatically improving the data-model comparison.Figure 6 compares simulated cross polar cap potential (CPCP) for all simulationsets against values obtained from AMIE and SuperDARN, for Event 3, which was theonly event in this study for which high quality AMIE and SuperDARN data were avail-able. Figure 6 is divided into three groups: in each group, the low res and high res sim-ulations are compared in separate subplots with the topmost group in part (a) illustrat-ing the impact of updated conductance coefficients on CPCP, middle group in part (b)investigating the impact of oval adjustments, and the bottom group in part (c) compar-ing the combined influence of dataset expansion and oval adjustments The difference be-tween the AMIE CPCP, denoted by the solid black line, and SuperDARN CPCP, de-noted by the dot-dashed line, has been demarcated using a thick dark grey region in eachsubplot to give an envelope of expected values based on the observations-based estimates.As shown in Figures 4 and 5, the introduction of CMEE and oval adjustments in-creases the value of the auroral conductance but does not dramatically impact the strengthof FACs, for a given domain resolution. Since the electrostatic potential is the direct out-put of Ohm’s Law, an increment in conductance with no substantial change in FACs leadsto a lower value of CPCP. This is explicitly observed in part (a), where RLM-driven sim-ulations overestimates the CPCP in both the
SWPC and
Hi-Res SWPC cases, in com-parison to CMEE-driven simulations. The
Hi-Res
RLM case, denoted in yellow (Frame6a-ii), particularly stands out because the FAC-driven conductance reaches the ceilingset by the coefficient A , i.e. as the magnitude of FACs increases, the value of conduc-tance attains the asymptotic maximum value ( A ) in the given model. Since the median –13–anuscript submitted to Space Weather A value is higher in CMEE it is able to give a reasonable CPCP estimate, while RLM’sreduced conductance peaks during the strongest driving resulting in the CPCP being anorder of magnitude greater. In part (b), conductance increments driven by oval adjust-ments largely reduces the CPCP, except during the main phase of the event when Kp >
4. This is because, during peak driving, the conductance from both models is so largethat the oval adjustments do not affect results substantially. In part (c), CMEE-drivenCPCP is lower than RLM-driven CPCP, as is expected. The CPCP values from Set D(Frame 6c-i) are too low, indicating that the model is overestimating the conductancewhich resulted in a lower CPCP. For the
Hi-Res case in Frame 6c-ii, the higher conduc-tance estimation coupled with better resolved FACs acts in favour of CMEE-driven sim-ulations in Set H, and leads to a more realistic CPCP as shown by the comparison againstAMIE and SuperDARN. In all events, simulations driven with RLM tend to have a higherCPCP compared to CMEE, as the conductance ceiling is higher in CMEE than RLM.Figure 7 illustrates the impact of conductance on dB/dt predictions during Event2, at two magnetometer stations - the high-latitude magnetometer station at Yellowknife(YKC) located at magnetic latitude (MLat) 68 . ◦ N and magnetic longitude (MLon)299 . ◦ , and the mid-latitude magnetometer station at Newport (NEW) located at MLat54 . ◦ N and MLon 304 . ◦ . While YKC and NEW are far apart latitudinally, longi-tudinally they are separated by less than 5 ◦ , making them a good candidate to studythe expansion of the auroral oval under strong driving conditions. The background ineach subplot, in addition to being coloured by Kp similar to Figures 5 and 6, are dark-ened to indicate times when the magnetometer was on the nightside. Additionally, dash-dot lines in all subplots indicate the four thresholds chosen in the Pulkkinen2013 study.Between 14:08 UT and 18:17 UT on December 14, 2006, as activity increases, mas-sive dB/dt spikes were observed at YKC with values crossing the four
Pulkkinen2013 thresh-olds. These spikes died down as activity increased, indicated by the increment in the Kp values. From ∼ dB/dt spikes at YKC barely cross the second and third threshold. During thistime period, the magnetometer was mostly on the nightside. Interestingly, all substan-tial perturbations observed at NEW occur during this same time interval, between 22:21UT and 07:54 UT. This is an indication that the auroral oval expanded equatorward dur-ing this given time interval as shown by the auroral radiance measurements by DefenceMeteorological Satellite Program (DMSP) F16 passes, with the storm intensifying. Thisexpansion of the oval resulted in latitudinally-high YKC no longer being in the auroralzone and instead being in the polar cap region, while the lower boundaries of the auro-ral oval reached latitudinally-lower NEW. Starting at 07:54 UT, spikes at NEW died downand were almost negligible throughout the rest of the event. Around the same time, mas-sive spikes crossing all four thresholds were observed again at YKC as the magnetome-ter station approaches the midnight-dawn sector. The spikes at YKC were observed un-til 16:33 UT as the magnetometer station rotated to the dawn-noon sector, through therecovery period of the event.In parts (b) and (c) of Figure 7, modeled dB/dt at YKC and NEW are comparedagainst observations. The topmost panel in part (b) compares modeled dB/dt from SetsE and F addressing the impact of dataset expansion. The middle panel in (b) comparesSets F and H to address the effect of auroral oval adjustments, while the bottom panelcompares Sets G and H to study the combined influence of both the expanded datasetand the oval adjustments. In part (c), modeled dB/dt from Sets G and H are comparedagainst observations at NEW. To simplify visualization, the minute-resolution data fromboth observed and modeled dB/dt values in parts (b) and (c) have been max-filtered forevery 10 minute interval. Additionally, the subplot background and threshold lines inparts (b) and (c) are plotted and coloured similarly to part (a).In the top panel of part (b), the magnitude of the CMEE-simulated dB/dt spikesare mostly at par with or moderately larger than the RLM-simulated spikes through most –14–anuscript submitted to Space Weather of the event. Both Sets E and F reasonably modeled the dB/dt during the time inter-val when the oval expanded and YKC was in the polar cap. However, they were unableto reproduce the heavy spikes that appeared both before and after the time interval, barelycrossing the fourth threshold of 1.5 nT /s at any given instance. In the middle panel, boththe frequency and magnitude of the dB/dt spikes increased significantly with the intro-duction of the oval adjustments. While this led to minor improvements in reproducingobservations at time intervals when YKC observed heavy spikes, a substantial changeoccured during the oval expansion when there were minimal dB/dt perturbations in boththe observations and the coefficient-driven simulation results but intense spikes at highfrequencies in the oval-adjusted simulation output. This increment in dB/dt spikes is dom-inant in the bottom panel of part (b) in both CMEE and RLM driven simulations. Theimpact of the dataset expansion combined with the oval adjustment in Set H simulationsled to a sharp increase in the magnitude of the spikes, in addition to the sharp rise infrequency. Part (c) indicate that the model does not reproduce the dB/dt spikes at NEW,regardless of the conductance model used. This is in direct contrast to the results fromthe last panel of part (b) which compares the same model variants but shows multipleintense dB/dt spikes at YKC during the same time interval. This indicates that whileusage of CMEE + oval adjustments improved the performance, there were still outstand-ing issues concerning the expansion and location of the oval that may require a more com-prehensive, physics-based approach.Figure 8 illustrates comparison magnetic perturbations ∆ B at the same magne-tometer stations during the same event to provide further clarity on the issue of auro-ral expansion. Part (a) compares the modeled and simulated ∆ B at YKC and NEW dur-ing the event. At YKC, heavy fluctuations were observed in the ∆ B values correspond-ing with the same time intervals when the massive spikes in dB/dt were observed in Fig-ure 7(a): between 14:21 UT and 18:19 UT, on December 14, and 06:42 UT and 17:07UT on December 15. The magnitude of ∆ B were ≥ nT during these time intervals.At NEW, while all variations in ∆ B were comparatively lower ( ≤ nT ), heavy fluc-tuations were seen during the same time interval when the auroral oval expands and sig-nificant dB/dt perturbations in Figure 7(a) occur, between 23:37 UT and 12:07 UT. Dur-ing the oval expansion phase, YKC-observed ∆ B increases steadily with time produc-ing minimal fluctuations during this period, retroactively indicating why the dB/dt islow. In parts (b) and (c) of Figure 8, the simulated ∆ B from Sets G and H reasonablyreproduce the observed ∆ B pattern. During the oval expansion phase of the event, thesimulated ∆ B of both sets fluctuate with higher frequency and magnitude than is ob-served at YKC, thereby explaining the massive spikes in the simulated dB/dt seen dur-ing the same time interval in Figure 7(b). Quantitatively, the Set H simulations exhibitthe best performance with a symmetric signed bias percentage (SSPB; Morley et al., 2018)of ∼ . B from either sets do not differ substan-tially with each other, with a negligible difference of ≤
1% in their respective SSPB. Nei-ther models are able to predict the perturbations during the main phase of the stormbetween 00:00 UT to 09:00 UT, explaining similarly poor performance in predicting the dB/dt values for this magnetometer. Part (d) compares the individual contributions ofthe global current systems - auroral Hall and Pedersen currents, field-aligned currentsand magnetospheric currents, in the ∆ B estimation at YKC and NEW from the Set Hsimulation. At YKC, auroral and field-aligned currents are the dominant current sys-tems driving perturbations in the magnetic field while magnetospheric currents contributenegligibly. The opposite is true at NEW, where the ∆ B variations are mostly driven bychanges in the magnetospheric currents and field aligned currents, with auroral currentsbarely affecting the simulated ∆ B even during the peak driving of the system, indicat-ing minimal contribution. This is further corroborated by the dial plots in Part (e) withthe top row showing the extent of saturated field aligned currents in the SWMF domain –15–anuscript submitted to Space Weather and compares it to the domain boundary of the modeled auroral conductance in the bot-tom row which clearly halts at 60 degree MLat.The comparisons in Figures 7 and 8 indicate that in the modeled ∆ B and dB/dt values, the auroral currents have little or no impact on mid and low latitude magnetome-ter predictions as the auroral oval is not able to extend equatorward to these latitudes.While this is expected during quiet conditions, the impact of auroral currents during ex-treme events can change dynamically with the expansion of the auroral oval, and canextend to much lower latitudes as evidenced by NEW during this event. The impact ofthis shortcoming on predictive skill has been described in further detail in Section 4. dB/dt Comparisons
The results from the binary event analysis performed on the dB/dt predictions showthat changing the auroral conductance in the global model, either by expanding the datasetor by applying the oval adjustments, led to minimal or no improvement in skill score forthe lowest dB/dt threshold, but improved skill for the remaining dB/dt thresholds, withthe most improvement in the highest thresholds. Table 3 presents a re-analysis of theresults from
Pulkkinen2013 , emphasizing the changes in the HSS of dB/dt results, thatwere caused by CMEE and the auroral oval adjustments. In part (a) of the table, theexpansion of dataset results in the improvement of HSS in each threshold for both thelow and high resolution cases, as evidenced by the difference column. This addresses
Welling2017 ’soriginal question, that expansion of the dataset can lead to improvement in dB/dt pre-dictions. In part (b), the HSS improvement caused by oval adjustments to the aurorais more substantial than in part (a), with HSS going up by ∼ . SWPC and
Hi-Res SWPC configurations. The comparison of both RLMand CMEE combined with oval adjustments in case (c) show similar improvements inpredictive skill for the higher dB/dt thresholds when using CMEE with oval adjustments.Figures 9(a) and (b) provide a quantitative picture of HSS improvement in the dB/dt predictions over many more thresholds. In both subplots, the y -axis is HSS, while theincreasing dB/dt thresholds on the x -axis provide a quantitative value of space weatheractivity. As expected, the HSS scores for all models decreased with increasing thresh-old value. However, in the most-extreme thresholds CMEE-driven simulations out-peformRLM-driven simulations, with improvements in the HSS of the same order as previouslyevidenced in Table 3. The HSS values in the highest dB/dt thresholds for the low-resolutionruns of CMEE, in both parts (a) and (b), were either at par or larger than the HSS val-ues for not only the low-resolution but also the high-resolution RLM simulations. Thisis a significant improvement in the skill score due to CMEE, as this provides an alter-nate physics-based remedy that otherwise could only be solved numerically. Naturally,the HSS values of the high-resolution CMEE-driven simulations were the highest at al-most all thresholds. Using this result, we can partially address the science questions posedin Section 1 that the auroral conductance impacts the dB/dt significantly, and that im-provements in the magnitude or pattern of the conductance boosts predictive skill scoresfor strong driving of the system.To better quantify the variation in model performance, the values of all performancemetrics listed in Table 2 were investigated. Table 4 presents these metrics calculated forall model variants at the high dB/dt threshold of 1 . SWPC configuration in the left and the
Hi-Res SWPC configuration in the right,with the worst performance by configuration coloured in orange and the best performancecoloured in blue. For both the
SWPC and
Hi-Res SWPC configurations, the POD andMR improved quite significantly for CMEE and the oval adjustments, indicating thatthe number of hits and misses increased and decreased, respectively. In addition, all skillscore metrics in the latter half of the table, excluding TSS, indicate best performancefor CMEE with oval adjustment variant for both resolutions of the model. The TS and –16–anuscript submitted to
Space Weather F score increased indicating that the number of hits increased. As has been shown inthe previous figure and table, the HSS improves as we switch models to introduce ovaladjustments and expansion of the dataset. However, the opposite occured when look-ing at POFD and FAR values were considered: the application of oval adjustments ledto sharply increased FAR values in both low and high res configurations. While the hitsand true negatives increased significantly and misses decreased, as supported by the PODand MR values, the number of false alarms increased steadily as the conductance coef-ficients were changed and jumped significantly with the application of the oval adjust-ments. This indirectly affected the TSS, which is defined as the difference between thehit rate and miss rate, or mathematically as 1 - (FAR + MR). Since the FAR increased,in spite of the decreased MR, TSS values reduced by more than 0.05 as we switched mod-els. Given that this order of change in skill was similar to what was achievable by chang-ing model resolutions, the increment in false alarms is a significant drawback when us-ing oval adjustments. The aforementioned trend was observed in all dB/dt thresholdsfrom 0.7 nT/s and above, indicating that this was not an isolated case. The performancemetrics for the other thresholds have been presented in the supp. material. B Estimation
Unlike the dB/dt performance quantification using binary event analysis, the us-age of the same procedure on ∆ B values does not help address the science questions posedin Section 1. Figure 10 describes variation in HSS for predicted ∆ B from all model vari-ants against observed values. In comparison to the dB/dt predictions, the change in ∆ B predictions were not nearly as drastic for better or worse. Note that the y-axis in Fig-ures 10(a) and (b) are not the same as in Figures 9(a) and (b); the HSS range spannedin the case of ∆ B is much shorter than in the case of dB/dt . In part (a), the CMEE-driven predictions show deterioration in the HSS values compared to RLM. However, incomparison to the variation in HSS for dB/dt by the expanded dataset, the variation ob-served is minimal. The decrease in HSS values was similar, but lesser, in the Hi-Res
SetF results. In part (b), the variation in ∆ B HSS values are negligible when oval adjust-ments were applied, for both model resolutions. In fact, some higher thresholds in part(b) showed no substantial change in the HSS values with the CMEE-driven simulations.When comparing parts (a) and (b) of Figure 10, the HSS values in part (b) are greaterthan their respective counterpart in part (a) of the figure for thresholds ≥ nT . Thisindicates that while changing coefficients by increasing the dataset caused more varia-tion in the HSS values of individual simulation sets, application of oval adjustments im-proves overall performance regardless of the coefficients used.For a more quantitative explanation of the ∆ B performance, Table 5 presents val-ues of all performance metrics calculated for all model variants at a high ∆ B thresholdof 400 nT. The table is similarly structured to Table 4 with the worst performance ineach configuration coloured orange and the best performance coloured blue. When com-paring the coefficient-driven simulations of RLM and CMEE, substantial variations arenot observed in almost all skill scores with a maximum difference of ∼ .
02 for any givenskill score and resolution. The same is seen with the simulations driven with oval adjust-ments, which also do not vary substantially. However, a significant jump is observed inthe skill scores when comparing the impact of oval adjustments with oval adjusted sim-ulations performing better than only coefficient-driven simulations. For both low andhigh res configurations, TS and F skill scores improve when oval adjustments are ap-plied. This is also seen in the accuracy measures like POD and MR whose values improve,with the POD jumping by a value of ∼ dB/dt metric analysis and in sharpcontrast to the aforementioned performance metrics, the POFD and FAR values are bestfor simulations driven using non-oval adjustment applications. This is similar to the re-sults in Section 3.2, where false alarms increase as we switch conductance models. Sim-ilar to Section 3.2, the trend seen in these performance metrics are not an isolated case –17–anuscript submitted to Space Weather (a) Impact of Dataset Expansion
Threshold
SWPC
Configuration
Hi-Res SWPC
ConfigurationRLM CMEE Difference RLM CMEE Difference .
033 0.624 0.640 +0.016 (b) Effect of Oval Adjustment (OA)
Threshold
SWPC
Configuration
Hi-Res SWPC
ConfigurationCMEE CMEE + Difference CMEE CMEE + Difference (c) Influence of Dataset expansion and OA Combination
Threshold
SWPC
Configuration
Hi-Res SWPC
ConfigurationRLM + CMEE + Difference RLM + CMEE + Difference ± − . + - Ridley Legacy Model, with Auroral Oval AdjustmentsCMEE + - Conductance Model for Extreme Events, with Auroral Oval Adjustments Table 3.
Comparison of Heidke Skill Scores (HSS) for the space weather events listed in Table1(a) at the prescribed four dB/dt thresholds (leftmost column) from
Pulkkinen2013 . (a) Thetop-most table compares HSS for the conductance coefficients of RLM and CMEE; no auroralamelioration added to the model; (b) The middle table compares results simulated using theCMEE using only the empirical conductance coefficients, against another version of the modelthat uses the CMEE coefficients along with the artificial oval adjustments; (c) The bottom-mosttable compares the two empirical models with the auroral oval adjustments. Here, green signifiesimprovement, while red signifies deterioration in prediction value. for this specific threshold, but observed in all thresholds. The performance metrics forthe other thresholds have been presented in the supp. material.The TSS and HSS do not show substantial differences as the conductance is mod-ified, with the maximum difference between skill scores not being more than ∼ .
05. Bycomparison, the difference between the best and the worst HSS performance for the dB/dt is ∼ .
11. The results also show that the best HSS and TSS for the
Hi-Res case are sim-ulations driven by RLM coefficients, which is in direct contrast to the low res case whereRLM coefficients consistently underperform for both TSS and HSS. This contrast is asa result of using the same time forecast window t f as the Pulkkinen2013 on ∆ B predic-tions. The comparison window t f of 20 minutes, used in both this study and the Pulkki-nen2013 study for dB/dt predictions, is not long enough to observe severe variations in –18–anuscript submitted to
Space Weather ∆ B perturbations. As an example, the predicted ∆ B hardly varies over more than twoof the pre-determined thresholds, even during strong driving. In comparison, dB/dt variesover multiple thresholds several times within a t f . This shows that the metrics used inthis study are not totally appropriate to study improvements in ∆ B predictions. Thiscould simply be done by increasing the comparison time window, or by using differenterror or bias metrics. As discussed earlier in Section 3.1 estimation of SSPB in Figure8 for specific magnetometer stations during Event 2 gives a quantitative understandingof the difference. The considerable increase in the frequency and magnitude of dB/dt spikes at YKCwith the application of the oval adjustments in Figure 7(b) is closely associated to thedomain constraints in RIM. As described in Section 2.2.1, while RIM’s simulation do-main spans the ionosphere pole-to-pole, the empirical auroral conductance module is lim-ited with a spatial domain spanning the poles to MLat 60 o . This means that in its presentconfiguration the auroral conductance module, be it RLM or CMEE, is bounded at MLat60 o , with conductance values equatorward of this boundary dropping exponentially andthe aurora being constrained poleward of the boundary. The impact of this boundaryis clearly indicated in Figure 8(d), where auroral currents are the dominant source of ground∆ B in high latitude regions like YKC, but contribute negligibly at mid latitudinal re-gions like NEW.Since application of both the dataset expansion and oval adjustments result in in-creasing the conductance ceiling during strong driving, CMEE allows more magnetosphericcurrents to close more dynamically throughout the ionosphere at any given time. In ad-dition, the oval adjustments enhance conductance in regions of high upward FACs therebychanging the pattern of the auroral conductance and reducing the conductance as a func-tion of distance from the empirically constructed oval. The combined effect of these mod-ifications would result in the auroral horizontal currents in RIM’s domain being estimatedwith increased accuracy. This, in turn, leads to a more accurate estimation of the ∆ B perturbation and subsequently dB/dt , which are both calculated from the Biot-Savartintegral of these current systems (e.g. Yu et al., 2010; Welling, 2019). The conductancemodifications due to the two elements (dataset expansion and oval adjustment) lead tonoisier results in dB/dt , which leads to increased spikes. These spikes, when correct, in-crease the number of hits and when incorrect, increase the number of false alarms. Theemergence of dB/dt spikes in the modeled data during the oval expansion phase in thebottom subplot of Figure 7(b) demarcates why false alarms increase when the oval ad-justment factor is used. In addition to the boundary constraints, false alarms are alsocaused by sudden shifting of the empirically-estimated auroral oval. These shifts are causedas a result of the sensitive dependence of the oval adjustments to changes in FAC pat-terns. Sharp changes in the FAC occuring over time scales in the same order of the cou-pling time cadence cause the empirical estimation of the oval to change rapidly. This briskmovement of the placement of the oval adjustment results in the loci movement of dB/dt spikes, causing unexpected hits and/or false alarms. In all, the aforementioned problemsplace the auroral oval in the wrong spot which lead to dB/dt spikes, perhaps even at theright time, but wrong location hence increasing the false alarms.While an increment in the number of false alarms is a significant drawback, the ad-vantages of using the improved conductance model in the SWMF far outweigh this is-sue. Firstly, the expansion of the dataset in CMEE allows for an increased limit cap onthe magnitude of the conductance which results in generating a more realistic cross po-lar cap potential to be fed back as input to the GM and IM modules. This is essentialwhen conducting numerical experiments investigating the magnetosphere-ionosphere cou-pling. Secondly, the changes in the conductance pattern in CMEE, as a result of the useof nonlinear regression, physically alters the nightside and dayside auroral conductance –19–anuscript submitted to Space Weather
Metric
SWPC
Configuration
Hi-Res SWPC
ConfigurationRLM CMEE RLM + CMEE + RLM CMEE RLM + CMEE + POD 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Table 4.
Performance metrics table for predicted dB/dt at the 1 . nT /s threshold. Listedare all performance metrics defined in Table 2 (Leftmost column) measured for SWMF simula-tions conducted using RLM Coefficients (denoted by ’ RLM ’), CMEE Coefficients (denoted by’
CMEE ’), RLM with oval adjustment (denoted by ’
RLM + ’) and CMEE with oval adjustment(denoted by ’ CMEE + ’) simulated using both the SWPC and
Hi-Res SWPC configurations. Theorange values show the least desirable metric results, while the blue values signify the best resultsfor this threshold.
Metric
SWPC
Configuration
Hi-Res SWPC
ConfigurationRLM CMEE RLM + CMEE + RLM CMEE RLM + CMEE + POD 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Table 5.
Performance metrics table for predicted ∆ B at the 400 nT threshold. Listed are allperformance metrics defined in Table 2 (Leftmost column) measured for SWMF simulations con-ducted using the same variants as in Table 4. The orange values show the least desirable metricresults, while the blue values signify the best results for this threshold. pattern when compared to RLM. Using global modeling, this numerical experiment hasnot only been able to address the question of expanded dataset raised by Welling2017 ,but is also able to discern the impact of ionospheric conductance on space weather fore-casting. Finally, both the magnitude and pattern of ionospheric conductance proves tobe an important quantity in affecting a global model’s dB/dt predictive skill. Given thatthe dB/dt is an important quantity used in the science community and the industry topredict space weather on the ground, accuracy in the ionospheric conductance is impor-tant in our global models. Through this work, the authors present an advanced and moreaccurate auroral conductance model to address this challenge.
In this work, the development of an advanced auroral conductance model, CMEEhas been presented. CMEE has been designed using nonlinear regression to span minute- –20–anuscript submitted to
Space Weather resolution data generated from AMIE for the whole year of 2003 spanning extreme events.It has additional capability to add physics-driven empirical adjustments to improve theauroral conductance to ensure a larger range on conductance values to better predict theconductance for a broad range of activity. In this study, this model has been used in theSWMF to investigate the impact of auroral conductance on space weather prediction.Simulated results were compared against observed global quantities like polar cap po-tential, field aligned current intensity and ground-based magnetic perturbation. Addi-tionally, a quantitative investigation was conducted using a binary event analysis sim-ilar to the
Pulkkinen2013 study and skill scores for dB/dt and ∆ B predictions were com-puted.The investigation showed that application of the increased dataset coupled withoval adjustments led to substantial changes in almost all space weather quantities. CMEEallows the auroral conductance to have an increased range of values, attaining a higherceiling during extreme driving as compared to RLM. Since FACs are largely driven byupstream conditions, they were not drastically impacted by changes in the conductancemodel. However, since the conductance value increased and FACs varied minimally, theCPCP values were lowered with the usage of CMEE and the oval adjustments. Since,auroral horizontal currents directly impact the ground magnetic perturbation ∆ B andits temporal variant dB/dt , the driving of both these quantities were appreciably alteredby the application of both the expanded dataset and oval adjustments. While usage ofthe expanded dataset resulted in a general increase of the modeled dB/dt magnitude,oval adjustments increased the frequency of dB/dt spikes. Neither of these propertieswere able to improve the modeling of the auroral oval expansion. This resulted in theformation of different regimes in the latitudinal contribution to the ∆ B and dB/dt dis-tributions, with negligble contribution of auroral currents in low or mid latitude mag-netometer stations in the modeled output during extreme driving.The results of the binary event analysis conducted on the simulation variants in-dicated that usage of CMEE with oval adjustments yields best performance, with dras-tic improvements in the HSS metric at higher activity thresholds. In addition, most per-formance metrics exhibited favourable changes when applying the CMEE coefficeints and/oroval adjustments, indicating an increase in the number of identified hits and true neg-atives and a decrease in misses. However, the performance metrics also indicated thatthe number of false alarms increased with the application of the oval adjustment. Thiswas caused predominantly because of the brisk movement of the empirically-estimatedoval, and the latitudinal constraint on the auroral conductance which inhibits the ovalfrom expanding beyond MLat 60 ◦ , thereby pushing the auroral currents poleward. Whilethis process increases the number of hits, favourably affecting most performance met-rics, it also hurts metrics like TSS due to increased number of false alarms. The binaryevent analysis of ∆ B predictions do not yield definitive results, exhibiting minimal im-pact on skill scores. This is most likely because the time forecast window of 20 minutes,chosen to study dB/dt forecasts in the original Pulkkinen2013 study, is limited for the∆ B to exhibit significant change in value so as to jump multiple number of thresholdsand therefore produce any meaningful changes in the performance metrics. Outstand-ing shortcomings of the present analysis such as those mentioned above and additionalanalysis like estimation of bias and error metrics for various thresholds are steps thatwe are presently pursuing. In addition, a key drawback of the present method is that themethod of estimating the conductance using AMIE data from times of extreme drivingis inconsistent, since the auroral conductance in AMIE is itself derived using an empir-ical relationship (Ahn et al., 1998). Because validation is a process, continued data-modelcomparisons will be performed in future studies. Further comparisons of the conductanceestimates, field aligned current and potential patterns against measurements by AMIE,SuperDARN and DMSP crossings will be presented. –21–anuscript submitted to Space Weather
The issues causing the misidentification of dB/dt spikes requires a physical solu-tion with numerical modifications to allow the aurora to expand to mid or low latitudesduring extreme events. While this could be done with data, an easier and more novelsolution would be to drive precipitation from the magnetospheric domains. This couldbe done by coupling physics-based precipitative inputs from GM and IM modules to es-timate electron and ion precipitation in the aurora. This is similar to what has been donein studies like Raeder et al. (2001) and Wiltberger et al. (2001). Such an approach al-lows for a novel approach to isolate and understand the impact of individual sources ofauroral conductance. At the same time, the precipitation pattern of the aurora allowsobservational data from extreme events to feature prominently in perceiving the accu-racy of precipitative fluxes at different MLTs and magnetic latitudes. The developmentof such a model is presently being undertaken by the authors to address the aforemen-tioned issues of dataset inconsistencies and oval expansion (Mukhopadhyay et al., 2018,2019).In conclusion, the usage of CMEE designed using an increased dataset coupled withthe application of oval adjustment parameters lead to substantial changes in our dB/dt predictions. With the crucial impact that the auroral conductance imparts on global quan-tities, CMEE would serve as a competent replacement to RLM’s coefficient map. Theusage of the oval adjustments in the SWMF’s auroral conductance estimation is uniqueand compelling in driving future developments of auroral conductance models to acheiveaccuracy in the conductance pattern, in addition to the magnitude. Additionally, as ev-idenced by the skill score analysis, the new model leads to significant improvement inpredictive skill of our space weather model.
Acknowledgments
Support for this work has been provided by NASA Grants NNX17AB87G, 80NSSC18K1120,and 80NSSC17K0015, and NSF Grant 1663770. We would like to acknowledge high-performancecomputing support from Pleaides (allocation 1815) provided by NASA’s High-End Com-puting Capability Programme, and Cheyenne (allocation UUSL0016) provided by NCAR’sComputational and Information Systems Laboratory, sponsored by the National ScienceFoundation. Model result data, input files and observation data are available via https://doi.org/10.7302/nwxp-g551. The Space Weather Modeling Framework is maintained by the University of Michi-gan Center for Space Environment Modeling and can be obtained at http://csem.engin.umich.edu/tools/swmf.AMIE Results used in this study are maintained at the University of Michigan’s VirtualModel Repository (VMR; http://vmr.engin.umich.edu/). The authors thank NASA Com-munity Coordinated Modeling Center (CCMC) Staff for providing the magnetometer mea-surements.The authors would like to thank Dr. Meghan Burleigh for reading a draft manuscript.We thank Dr. Shasha Zou, Dr. Robert Robinson, Dr. Steven Morley and Dr. Gabor Tothfor sharing their expertise in the course of this study. A.M. would like to thank Dr. Do-gacan su Ozturk, Dr. Zhenguang Huang, Dr. Natalia Ganjushkina, Ms. Abigail Azari,Mr. Alexander Shane, Mr. Brian Swiger and Mr. Christopher Bert for sharing their ex-pertise during the development of modeling, curve-fitting and validation tools used inthis study.
References
Ahn, B.-H., Richmond, A. D., Kamide, Y., Kroehl, H. W., Emery, B. A., de laBeaujardi´ere, O., & Akasofu, S.-I. (1998, jul). An ionospheric conduc-tance model based on ground magnetic disturbance data.
Journal of Geo-physical Research: Space Physics , (A7), 14769–14780. Retrieved from http://doi.wiley.com/10.1029/97JA03088 doi: 10.1029/97JA03088Anderson, B. J., Korth, H., Waters, C. L., Green, D. L., Merkin, V. G., Barnes, –22–anuscript submitted to Space Weather
R. J., & Dyrud, L. P. (2014). Development of large-scale birkeland currents de-termined from the active magnetosphere and planetary electrodynamics responseexperiment.
Geophysical Research Letters , (9), 3017-3025. Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1002/2014GL059941 doi: 10.1002/2014GL059941Anderson, B. J., Korth, H., Welling, D. T., Merkin, V. G., Wiltberger, M. J.,Raeder, J., . . . Rastaetter, L. (2017). Comparison of predictive estimates ofhigh-latitude electrodynamics with observations of global-scale birkeland currents. Space Weather , (2), 352-373. Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1002/2016SW001529 doi: 10.1002/2016SW001529Axford, W. I., & Hines, C. O. (1961). A Unifying Theory of High-Latitude Geophys-ical Phenomena and Geomagnetic Storms. Canadian Journal of Physics , (10),1433–1464. Retrieved from https://doi.org/10.1139/p61-172 doi: 10.1139/p61-172Boonsiriseth, A., Thorne, R. M., Lu, G., Jordanova, V. K., Thomsen, M. F., Ober,D. M., & Ridley, A. J. (2001, jul). A semiempirical equatorial mapping ofAMIE convection electric potentials (MACEP) for the January 10, 1997, mag-netic storm. Journal of Geophysical Research: Space Physics , (A7), 12903–12917. Retrieved from http://doi.wiley.com/10.1029/1999JA000332 doi:10.1029/1999JA000332Brekke, A., & Moen, J. (1993). Observations of high latitude ionospheric con-ductances. Journal of Atmospheric and Terrestrial Physics , (11), 1493–1512. Retrieved from doi: https://doi.org/10.1016/0021-9169(93)90126-JCarter, J. A., Milan, S. E., Coxon, J. C., Walach, M.-T., & Anderson, B. J. (2016).Average field-aligned current configuration parameterized by solar wind con-ditions. Journal of Geophysical Research: Space Physics , (2), 1294–1307.Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1002/2015JA021567 doi: 10.1002/2015JA021567Chapman, S. (1931, jan). The absorption and dissociative or ionizing effect ofmonochromatic radiation in an atmosphere on a rotating earth. Proceedings ofthe Physical Society , (1), 26–45. Retrieved from http://stacks.iop.org/0959-5309/43/i=1/a=305?key=crossref.46895a3aef390982dcfb99f7afc88ced doi: 10.1088/0959-5309/43/1/305Cid, C., Saiz, E., Guerrero, A., Palacios, J., & Cerrato, Y. (2015). A Carrington-likegeomagnetic storm observed in the 21st century. J. Space Weather Space Clim. , ,A16. Retrieved from https://doi.org/10.1051/swsc/2015017 doi: 10.1051/swsc/2015017Connor, H. K., Zesta, E., Fedrizzi, M., Shi, Y., Raeder, J., Codrescu, M. V., &Fuller-Rowell, T. J. (2016). Modeling the ionosphere-thermosphere responseto a geomagnetic storm using physics-based magnetospheric energy input:OpenGGCM-CTIM results. Journal of Space Weather and Space Climate , ,A25. Retrieved from doi: 10.1051/swsc/2016019De Zeeuw, D. L., Sazykin, S., Wolf, R. A., Gombosi, T. I., Ridley, A. J., & T´oth,G. (2004). Coupling of a global MHD code and an inner magnetospheric model:Initial results. Journal of Geophysical Research: Space Physics , (A12), 1–14.doi: 10.1029/2003JA010366Doherty, P., Coster, A. J., & Murtagh, W. (2004). Space weather effects of Octo-berNovember 2003. GPS Solutions , (4), 267–271. Retrieved from https://doi.org/10.1007/s10291-004-0109-3 doi: 10.1007/s10291-004-0109-3Dungey, J. W. (1963, jan). Interactions of solar plasma with the geomag-netic field. Planetary and Space Science , , 233–237. Retrieved from –23–anuscript submitted to Space Weather doi: 10.1016/0032-0633(63)90020-5Frahm, R. A., Winningham, J. D., Sharber, J. R., Link, R., Crowley, G., Gaines,E. E., . . . Potemra, T. A. (1997, dec). The diffuse aurora: A significant sourceof ionization in the middle atmosphere.
Journal of Geophysical Research: At-mospheres , (D23), 28203–28214. Retrieved from http://doi.wiley.com/10.1029/97JD02430 doi: 10.1029/97JD02430Fuller-Rowell, T. J., & Evans, D. S. (1987). Height-integrated pedersen and hallconductivity patterns inferred from the tiros-noaa satellite data. Journal of Geo-physical Research: Space Physics , (A7), 7606-7618. Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/JA092iA07p07606 doi:10.1029/JA092iA07p07606Gao, Y. (2012). Comparing the cross polar cap potentials measured by superdarnand amie during saturation intervals. Journal of Geophysical Research: SpacePhysics , (A8). Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2012JA017690 doi: 10.1029/2012JA017690Glocer, A., Rasttter, L., Kuznetsova, M., Pulkkinen, A., Singer, H. J., Balch, C., . . .Wing, S. (2016). Community-wide validation of geospace model local k-indexpredictions to support model transition to operations. Space Weather , (7),469-480. Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1002/2016SW001387 doi: 10.1002/2016SW001387Gombosi, T. I., De Zeeuw, D. L., Powell, K. G., Ridley, A. J., Sokolov, I. V., Stout,Q. F., & T´oth, G. (2003). Adaptive mesh refinement for global magnetohy-drodynamic simulation. In J. B¨uchner, M. Scholer, & C. T. Dum (Eds.), Spaceplasma simulation (pp. 247–274). Berlin, Heidelberg: Springer Berlin Heidel-berg. Retrieved from https://doi.org/10.1007/3-540-36530-3 12 doi:10.1007/3-540-36530-3 12Goodman, M. L. (1995, aug). A three-dimensional, iterative mapping procedurefor the implementation of an ionosphere-magnetosphere anisotropic Ohm’s lawboundary condition in global magnetohydrodynamic simulations.
AnnalesGeophysicae , (8), 843–853. Retrieved from https://doi.org/10.1007/s00585-995-0843-z doi: 10.1007/s00585-995-0843-zHaiducek, J. D., Welling, D. T., Ganushkina, N. Y., Morley, S. K., & Ozturk, D. S.(2017). SWMF Global Magnetosphere Simulations of January 2005: GeomagneticIndices and Cross-Polar Cap Potential. Space Weather , (12), 1567–1587. Re-trieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1002/2017SW001695 doi: 10.1002/2017SW001695Hanssen, A. W., & Kuipers, W. J. A. (1965). On the relationship between the fre-quency of rain and various meteorological parameters. Meded. Verh. , , 2 – 15.Hartinger, M. D., Xu, Z., Clauer, C. R., Yu, Y., Weimer, D. R., Kim, H., . . . Willer,A. N. (2017). Associating ground magnetometer observations with current orvoltage generators. Journal of Geophysical Research: Space Physics , (7), 7130-7141. Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1002/2017JA024140 doi: 10.1002/2017JA024140Heidke, P. (1926). Berechnung des Erfolges und der G¨ute derWindst¨arkevorhersagen im Sturmwarnungsdienst. Geografiska Annaler , , 301–349.Herrmann, L. R. (1976). Laplacian-Isoparametric Grid Generation Scheme. Journalof the Engineering Mechanics Division , (5), 749–907.Honkonen, I., Rast¨atter, L., Grocott, A., Pulkkinen, A., Palmroth, M., Raeder,J., . . . Wiltberger, M. (2013, may). On the performance of global magneto-hydrodynamic models in the Earth’s magnetosphere. Space Weather , (5),313–326. Retrieved from http://doi.wiley.com/10.1002/swe.20055 doi:10.1002/swe.20055Iijima, T., & Potemra, T. A. (1976). The amplitude distribution of fiel- –24–anuscript submitted to Space Weather daligned currents at northern high latitudes observed by Triad.
Jour-nal of Geophysical Research-Space Physics , (13), 2165–2174. Retrievedfrom http://onlinelibrary.wiley.com/doi/10.1029/JA081i013p02165/abstract{%}5Cnpapers3://publication/doi/10.1029/JA081i013p02165 doi:10.1029/JA081i013p02165Jolliffe, I. T., & Stephenson, D. B. (2012). Forecast Verification: A Practitioner’sGuide in Atmospheric Science . John Wiley & Sons. Retrieved from https://books.google.com/books?hl=en{&}lr={&}id=DCxsKQeaBH8C{&}pgis=1
Kaeppler, S. R., Hampton, D. L., Nicolls, M. J., Strmme, A., Solomon, S. C.,Hecht, J. H., & Conde, M. G. (2015). An investigation comparing ground-based techniques that quantify auroral electron flux and conductance.
Journalof Geophysical Research: Space Physics , (10), 9038-9056. Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1002/2015JA021396 doi: 10.1002/2015JA021396Khachikjan, G. Y., Koustov, A. V., & Sofko, G. J. (2008). Dependence of super-darn cross polar cap potential upon the solar wind electric field and magnetopausesubsolar distance. Journal of Geophysical Research: Space Physics , (A9). Re-trieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2008JA013107 doi: 10.1029/2008JA013107Kihn, E. A., & Ridley, A. J. (2005). A statistical analysis of the assimila-tive mapping of ionospheric electrodynamics auroral specification. Journalof Geophysical Research: Space Physics , (A7). Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2003JA010371 doi:10.1029/2003JA010371Knight, S. (1973). Parallel electric fields. Planetary and Space Science . doi: 10.1016/0032-0633(73)90093-7Korth, H., Zhang, Y., Anderson, B. J., Sotirelis, T., & Waters, C. L. (2014). Sta-tistical relationship between large-scale upward field-aligned currents and electronprecipitation.
Journal of Geophysical Research: Space Physics , (8), 6715–6731.Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1002/2014JA019961 doi: 10.1002/2014JA019961Le, G., Lhr, H., Anderson, B. J., Strangeway, R. J., Russell, C. T., Singer, H., . . .Torbert, R. B. (2016). Magnetopause erosion during the 17 march 2015 mag-netic storm: Combined field-aligned currents, auroral oval, and magnetopauseobservations. Geophysical Research Letters , (6), 2396-2404. Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1002/2016GL068257 doi: 10.1002/2016GL068257Liemohn, M. W., Ganushkina, N. Y., De Zeeuw, D. L., Rastaetter, L., Kuznetsova,M., Welling, D. T., . . . van der Holst, B. (2018). Real-time swmf at ccmc: As-sessing the dst output from continuous operational simulations. Space Weather , (10), 1583-1603. Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2018SW001953 doi: 10.1029/2018SW001953Liemohn, M. W., McCollough, J. P., Jordanova, V. K., Ngwira, C. M., Morley,S. K., Cid, C., . . . Vasile, R. (2018). Model evaluation guidelines for geomag-netic index predictions. Space Weather , (12), 2079-2102. Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2018SW002067 doi: 10.1029/2018SW002067Liemohn, M. W., Ridley, A. J., Brandt, P. C., Gallagher, D. L., Kozyra, J. U.,Ober, D. M., . . . DeMajistre, R. (2005, dec). Parametric analysis of nightsideconductance effects on inner magnetospheric dynamics for the 17 April 2002storm. Journal of Geophysical Research , (A12), A12S22. Retrieved from http://doi.wiley.com/10.1029/2005JA011109 doi: 10.1029/2005JA011109Lu, G., Siscoe, G. L., Richmond, A. D., Pulkkinen, T. I., Tsyganenko, N. A.,Singer, H. J., & Emery, B. A. (1997). Mapping of the ionospheric field- –25–anuscript submitted to Space Weather aligned currents to the equatorial magnetosphere.
Journal of GeophysicalResearch: Space Physics , (A7), 14467–14476. Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/97JA00744 doi:10.1029/97JA00744Merkin, V. G., Milikh, G., Papadopoulos, K., Lyon, J., Dimant, Y. S., Sharma,A. S., . . . Wiltberger, M. (2005, nov). Effect of anomalous electron heating onthe transpolar potential in the LFM global MHD model. Geophysical ResearchLetters , (22), n/a–n/a. Retrieved from http://doi.wiley.com/10.1029/2005GL023315 doi: 10.1029/2005GL023315Merkin, V. G., Sharma, A. S., Papadopoulos, K., Milikh, G., Lyon, J., & Goodrich,C. (2005, sep). Global MHD simulations of the strongly driven magnetosphere:Modeling of the transpolar potential saturation. Journal of Geophysical Research:Space Physics , (A9). Retrieved from http://doi.wiley.com/10.1029/2004JA010993 doi: 10.1029/2004JA010993Merkine, V. G., Papadopoulos, K., Milikh, G., Sharma, A. S., Shao, X., Lyon, J.,& Goodrich, C. (2003, dec). Effects of the solar wind electric field and iono-spheric conductance on the cross polar cap potential: Results of global MHDmodeling. Geophysical Research Letters , (23), n/a–n/a. Retrieved from http://doi.wiley.com/10.1029/2003GL017903 doi: 10.1029/2003GL017903Moen, J., & Brekke, A. (1993, may). The solar flux influence on quiet time con-ductances in the auroral ionosphere. Geophysical Research Letters , (10), 971–974. Retrieved from http://doi.wiley.com/10.1029/92GL02109 doi: 10.1029/92GL02109Morley, S. K., Brito, T. V., & Welling, D. T. (2018). Measures of Model Perfor-mance Based On the Log Accuracy Ratio. Space Weather , (1), 69–88. Re-trieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1002/2017SW001669 doi: 10.1002/2017SW001669Mukhopadhyay, A. (2017). Statistical Comparison of Magnetopause Distances andCPCP Estimation by Global MHD Models (Tech. Rep.). Retrieved from https://ccmc.gsfc.nasa.gov/RoR{ }WWW/SWREDI/contest-presentations/2017/Statistical{ }Comparison{ }of{ }MP{ }Distances{ }and{ }CPCP{ }CCMC{ }Contest{ }2{ }AgnitM.pdf doi: https://doi.org/10.1002/essoar.10502157.1Mukhopadhyay, A., Welling, D., Burleigh, M., Ridley, A., Liemohn, M., Anderson,B., & Gjerloev, J. (2019, dec). Conductance in the Aurora: Influence of Magneto-spheric Contributors. In
Agu fall meeting abstracts (Vol. 2019, pp. SA41B–3169).Retrieved from https://ui.adsabs.harvard.edu/abs/2019AGUFMSA41B3169M/abstract doi: doi.org/10.1002/essoar.10502150.1Mukhopadhyay, A., Welling, D., Liemohn, M., Zou, S., & Ridley, A. (2018, dec).Challenges in Space Weather Prediction: Estimation of Auroral Conductance.In
Agu fall meeting abstracts (Vol. 2018, pp. SA33B–3462). Retrieved from https://ui.adsabs.harvard.edu/abs/2018AGUFMSA33B3462M/abstract
Newell, P. T., Sotirelis, T., & Wing, S. (2009, sep). Diffuse, monoenergetic, andbroadband aurora: The global precipitation budget.
Journal of GeophysicalResearch: Space Physics , (A9). Retrieved from http://doi.wiley.com/10.1029/2009JA014326 doi: 10.1029/2009JA014326Ohtani, S., Wing, S., Merkin, V. G., & Higuchi, T. (2014, jan). Solar cycledependence of nightside field-aligned currents: Effects of dayside ionosphericconductivity on the solar wind-magnetosphere-ionosphere coupling. Jour-nal of Geophysical Research: Space Physics , (1), 322–334. Retrieved from http://doi.wiley.com/10.1002/2013JA019410 doi: 10.1002/2013JA019410Ozturk, D. S., Zou, S., & Slavin, J. A. (2017). IMF By effects on ground mag-netometer response to increased solar wind dynamic pressure derived fromglobal MHD simulations. Journal of Geophysical Research: Space Physics . doi:10.1002/2017JA023903 –26–anuscript submitted to
Space Weather
Perlongo, N. J., Ridley, A. J., Liemohn, M. W., & Katus, R. M. (2017, apr). Theeffect of ring current electron scattering rates on magnetosphere-ionospherecoupling.
Journal of Geophysical Research: Space Physics , (4), 4168–4189. Retrieved from http://doi.wiley.com/10.1002/2016JA023679 doi:10.1002/2016JA023679Powell, K. G., Roe, P. L., Linde, T. J., Gombosi, T. I., & Zeeuw, D. L. D.(1999). A Solution-Adaptive Upwind Scheme for Ideal Magnetohydrodynam-ics. Journal of Computational Physics , (2), 284–309. Retrieved from doi: https://doi.org/10.1006/jcph.1999.6299Pujol, J. (2007). The solution of nonlinear inverse problems and the Levenberg-Marquardt method. Geophysics , (4), W1–W16. Retrieved from https://doi.org/10.1190/1.2732552 doi: 10.1190/1.2732552Pulkkinen, A., Kuznetsova, M., Ridley, A., Raeder, J., Vapirev, A., Weimer, D.,. . . Chulaki, A. (2011, feb). Geospace Environment Modeling 2008-2009Challenge: Ground magnetic field perturbations. Space Weather , (2), n/a–n/a. Retrieved from http://doi.wiley.com/10.1029/2010SW000600 doi:10.1029/2010SW000600Pulkkinen, A., Rast¨atter, L., Kuznetsova, M., Singer, H., Balch, C., Weimer, D., . . .Weigel, R. (2013, jun). Community-wide validation of geospace model groundmagnetic field perturbation predictions to support model transition to opera-tions. Space Weather , (6), 369–385. Retrieved from http://doi.wiley.com/10.1002/swe.20056 doi: 10.1002/swe.20056Raeder, J., McPherron, R. L., Frank, L. A., Kokubun, S., Lu, G., Mukai, T., . . .Slavin, J. A. (2001). Global simulation of the Geospace Environment Model-ing substorm challenge event. Journal of Geophysical Research-Space Physics , (A1), 381–395. doi: 10.1029/2000ja000605Richmond, A. D., & Kamide, Y. (1988). Mapping electrodynamic features of thehigh-latitude ionosphere from localized observations - Technique. Journal of Geo-physical Research , (A6), 5741–5759. doi: 10.1029/JA093iA06p05741Ridley, A. J., De Zeeuw, D. L., Gombosi, T. I., & Powell, K. G. (2001). Usingsteady state MHD results to predict the global state of the magnetosphere-ionosphere system. Journal of Geophysical Research , (A12), 30067. Re-trieved from http://adsabs.harvard.edu/abs/2001JGR...10630067R doi:10.1029/2000JA002233Ridley, A. J., Gombosi, T. I., & De Zeeuw, D. L. (2004). Ionospheric control of themagnetosphere: conductance. Annales Geophysicae , (2), 567–584. Retrievedfrom https://hal.archives-ouvertes.fr/hal-00317238/ doi: 10.5194/angeo-22-567-2004Ridley, A. J., Gombosi, T. I., Sokolov, I. V., T´oth, G., & Welling, D. T. (2010, aug).Numerical considerations in simulating the global magnetosphere. Annales Geo-physicae , (8), 1589–1614. Retrieved from doi: 10.5194/angeo-28-1589-2010Ridley, A. J., & Liemohn, M. W. (2002, aug). A model-derived storm timeasymmetric ring current driven electric field description. Journal of Geophys-ical Research: Space Physics , (A8), SMP 2–1–SMP 2–12. Retrieved from http://doi.wiley.com/10.1029/2001JA000051 doi: 10.1029/2001JA000051Robinson, R. M., Vondrak, R. R., Miller, K., Dabbs, T., & Hardy, D. (1987,mar). On calculating ionospheric conductances from the flux and energyof precipitating electrons. Journal of Geophysical Research , (A3), 2565.Retrieved from http://doi.wiley.com/10.1029/JA092iA03p02565 doi:10.1029/JA092iA03p02565Robinson, R. M., Zhang, Y., Anderson, B. J., Zanetti, L. J., Korth, H., & Fitzmau-rice, A. (2018). Statistical relations between field-aligned currents and precip- –27–anuscript submitted to Space Weather itating electron energy flux.
Geophysical Research Letters , (17), 8738-8745.Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2018GL078718 doi: 10.1029/2018GL078718Roederer, J. G. (1970). Dynamics of Geomagnetically Trapped Radiation (Vol. 2) .Berlin, Heidelberg: Springer Berlin Heidelberg. doi: https://doi.org/10.1007/978-3-642-49300-3Schunk, R., & Nagy, A. (2009).
Ionospheres: Physics, Plasma Physics, and Chem-istry (2nd ed.). Cambridge University Press. doi: 10.1017/CBO9780511635342Sorkine, O., Cohen-Or, D., Lipman, Y., Alexa, M., R¨ossl, C., & Seidel, H.-P. (2004).Laplacian surface editing. In
Proceedings of the 2004 eurographics/acm siggraphsymposium on geometry processing (p. 175184). New York, NY, USA: Associationfor Computing Machinery. Retrieved from https://doi.org/10.1145/1057432.1057456 doi: 10.1145/1057432.1057456T´oth, G., Meng, X., Gombosi, T. I., & Rast¨atter, L. (2014). Predicting the timederivative of local magnetic perturbations.
Journal of Geophysical Research: SpacePhysics , (1), 310–321. Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1002/2013JA019456 doi: 10.1002/2013JA019456T´oth, G., Sokolov, I. V., Gombosi, T. I., Chesney, D. R., Clauer, C. R., De Zeeuw,D. L., . . . K´ota, J. (2005, dec). Space Weather Modeling Framework: A new toolfor the space science community. Journal of Geophysical Research , (A12),A12226. Retrieved from http://doi.wiley.com/10.1029/2005JA011126 doi:10.1029/2005JA011126T´oth, G., van der Holst, B., Sokolov, I. V., De Zeeuw, D. L., Gombosi, T. I.,Fang, F., . . . Opher, M. (2012). Adaptive numerical algorithms in spaceweather modeling. Journal of Computational Physics , (3), 870–903. doi:10.1016/j.jcp.2011.02.006Waters, C. L., Anderson, B. J., Green, D. L., Korth, H., Barnes, R. J., & Van-ham¨aki, H. (2020). Science data products for ampere. In M. W. Dunlop &H. L¨uhr (Eds.), Ionospheric multi-spacecraft analysis tools: Approaches for de-riving ionospheric parameters (pp. 141–165). Cham: Springer InternationalPublishing. Retrieved from https://doi.org/10.1007/978-3-030-26732-2 7 doi: 10.1007/978-3-030-26732-2 7Welling, D. T. (2019). Magnetohydrodynamic models of b and their use in gicestimates. In
Geomagnetically induced currents from the sun to the power grid (p. 43-65). American Geophysical Union (AGU). Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1002/9781119434412.ch3 doi: 10.1002/9781119434412.ch3Welling, D. T., Anderson, B. J., Crowley, G., Pulkkinen, A. A., & Rast¨atter, L.(2017, jan). Exploring predictive performance: A reanalysis of the geospacemodel transition challenge.
Space Weather , (1), 192–203. Retrieved from http://doi.wiley.com/10.1002/2016SW001505 doi: 10.1002/2016SW001505Welling, D. T., Ngwira, C. M., Opgenoorth, H., Haiducek, J. D., Savani, N. P.,Morley, S. K., . . . Liemohn, M. (2018). Recommendations for next-generationground magnetic perturbation validation. Space Weather , (12), 1912-1920.Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2018SW002064 doi: 10.1029/2018SW002064Welling, D. T., & Ridley, A. J. (2010). Exploring sources of magnetospheric plasmausing multispecies MHD. Journal of Geophysical Research: Space Physics , (A4). Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2009JA014596 doi: 10.1029/2009JA014596Wilks, D. S. (2011). Statistical methods in the atmospheric sciences (3rd ed.). Aca-demic Press.Wiltberger, M., Merkin, V., Zhang, B., Toffoletto, F., Oppenheim, M., Wang,W., . . . Stephens, G. K. (2017, may). Effects of electrojet turbulence on –28–anuscript submitted to
Space Weather a magnetosphere-ionosphere simulation of a geomagnetic storm.
Journal ofGeophysical Research: Space Physics , (5), 5008–5027. Retrieved from http://doi.wiley.com/10.1002/2016JA023700 doi: 10.1002/2016JA023700Wiltberger, M., Wang, W., Burns, A. G., Solomon, S. C., Lyon, J. G., & Goodrich,C. C. (2004). Initial results from the coupled magnetosphere ionosphere ther-mosphere model: magnetospheric and ionospheric responses. Journal of At-mospheric and Solar-Terrestrial Physics , (15), 1411–1423. Retrieved from doi: https://doi.org/10.1016/j.jastp.2004.03.026Wiltberger, M., Weigel, R. S., Lotko, W., & Fedder, J. A. (2001). Modeling sea-sonal variations of auroral particle precipitation in a global-scale magnetosphere-ionosphere simulation. Journal of Geophysical Research-Space Physics , (A1),381–395. doi: 10.1029/2008JA013108Wolf, R. A., Harel, M., Spiro, R. W., Voigt, G.-H., Reiff, P. H., & Chen, C.-K.(1982, aug). Computer simulation of inner magnetospheric dynamics for themagnetic storm of July 29, 1977. Journal of Geophysical Research , (A8),5949. Retrieved from http://doi.wiley.com/10.1029/JA087iA08p05949 doi:10.1029/JA087iA08p05949Yagou, H., Ohtake, Y., & Belyaev, A. (2002, July). Mesh smoothing via mean andmedian filtering applied to face normals. In Geometric modeling and processing.theory and applications. gmp 2002. proceedings (p. 124-131). doi: 10.1109/GMAP.2002.1027503Yu, Y., Jordanova, V. K., Ridley, A. J., Albert, J. M., Horne, R. B., & Jef-fery, C. A. (2016). A new ionospheric electron precipitation module cou-pled with ram-scb within the geospace general circulation model.
Journalof Geophysical Research: Space Physics , (9), 8554-8575. Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1002/2016JA022585 doi: 10.1002/2016JA022585Yu, Y., Ridley, A. J., Welling, D. T., & T´oth, G. (2010, aug). Including gap re-gion field-aligned currents and magnetospheric currents in the MHD calculationof ground-based magnetic field perturbations. Journal of Geophysical Research:Space Physics , (A8). Retrieved from http://doi.wiley.com/10.1029/2009JA014869 doi: 10.1029/2009JA014869Zhang, B., Lotko, W., Brambles, O., Wiltberger, M., & Lyon, J. (2015). Elec-tron precipitation models in global magnetosphere simulations. Journal ofGeophysical Research: Space Physics , (2), 1035–1056. Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1002/2014JA020615 doi: 10.1002/2014JA020615 –29–anuscript submitted to Space Weather
Figure 3.
Example Fitting of the Conductance Model for Extreme Events (CMEE) - (a) Rep-resentative Line Plot of Auroral Conductance (Hall or Pedersen, in siemens) vs. Field AlignedCurrents (FACs, Upward or Downward, in µA/m ) through Equation 1 denoting the three re-gions of interest - low and high FAC bins used to estimate the values of A and A , while theregion in between these bins defining the curve using regression of A . (b) An example log-logplot of the AMIE data showing the scatter of Hall Conductance versus Upward Field AlignedCurrents, at magnetic latitude of 68 o and magnetic local time (MLT) 23 in the nightside auroralzone. Alongside the data spread, the regression line is plotted in red with the dot-dashed linesexhibiting the low and high FAC bins. (c) The distribution of AMIE data from 2003 showingthe scatter of Hall Conductance versus all Field Aligned Currents plotted along the line plots ofRLM and CMEE, denoted in blue and red respectively, at 68 o magnetic latitude and 23 MLT.Note this distribution plot is in linear scale compared to the similar plot part (b), which is inlogarithmic scale. –30–anuscript submitted to Space Weather
Figure 4.
A comparison of Hall conductance values from different conductance model vari-ants. Dial plots from (left to right) simulation sets A, B, C and D at time instances during Event1 (Epoch 1, Top Row) when
Kp <
4, (Epoch 2, Second Row) when 4 ≤ Kp <
8, and (Epoch 3,Third Row) when Kp ≥
8. (Bottom Subplot) Comparison of Kp from the Kyoto Observatory (inblack) against simulated Kp from simulationsets A (in red), B (in blue), C (in gold) and D (ingreen). Additionally, the plot background is coloured by the Kp , green signifying Kp <
4, yellowsignifying 4 ≤ Kp <
8, and red signifying Kp ≥ Space Weather
Figure 5.
Time series comparison of integrated field aligned currents (iFACs) for Events5 spanning the storm main phase from AMPERE (gray line) and the eight simulation sets ofthe SWMF. Goal of each frame: Top Frame (a) illustrates the impact of dataset expansion oniFACs by comparing Sets A (in red), B (in blue), E (in gold) and D (in green). Middle Frame(b) displays the effect of oval adjustments by comparing Sets B (in light blue), D (in blue), F(in light green) and H (in green). Bottom Frame (c) presents the combined influence of datasetexpansion and oval adjustments by comparing Sets C (in red), D (in blue), G (in gold) and H (ingreen). The plot background is coloured by the Kp , green signifying Kp <
4, yellow signifying4 ≤ Kp <
8, and red signifying Kp ≥
8. –32–anuscript submitted to
Space Weather
Figure 6.
Time series comparison of cross polar cap potential (CPCP) for Event 3 compar-ing observations from AMIE, SuperDARN, and the eight configurations of the SWMF. Tracesshow AMIE in solid black, SuperDARN in dashed black, with the difference region between thedatasets coloured gray. The SWMF simulations are coloured similarly to Figure 5. Goal of eachframe: Top Frame (a) illustrates the impact of dataset expansion on iFACs by comparing (i)Sets A & B in upper panel, and (ii) Sets E & D in bottom panel. Middle Frame (b) displays theeffect of oval adjustments by comparing (i) Sets B & D in upper panel, and (ii) F & H (in green)in bottom panel. Bottom Frame (c) presents the combined influence of dataset expansion andoval adjustments by comparing (i) Sets C & D in top panel, and (ii) G & H in bottom panel.The plot background is coloured by the Kp , green signifying Kp <
4, and yellow signifying4 ≤ Kp <
8. –33–anuscript submitted to
Space Weather
Figure 7.
Impact of changes to the auroral conductance on dB/dt predictions - (a) (Left) Lo-cation of Yellowknife (YKC) and Newport (NEW) magnetometer stations mapped in geographiccoordinates with the SWMF auroral boundary demarcated using the thick blue line. (Right)Raw dB/dt observations at a 1-minute cadence at YKC and NEW. (Bottom) Expansion of theauroral oval as seen through DMSP F16 auroral radiance maps and the magnetometer stationsat Yellowknife (YKC) and Newport (NEW). The dialplots on top are demarcated by blue, green,yellow and red dot-dashed lines in the line plots, in increasing order of their timestamps. (b)Comparison of max-filtered predicted dB/dt from
Hi-Res
SWMF simulations against similarlyfiltered dB/dt observations at Yellowknife (YKC). Goal of each panel: Top panel (i) shows im-pact of coefficients by comparing simulation sets E (in red) and F (in blue). Middle panel (ii)illustrates the impact of oval adjustments by comparing sets F (in light blue) and H (in blue).Bottom panel (iii) compares sets G (in red) and H (in blue). Observations are shown as a thick,grey curve. (c) Comparison of max-filtered predicted dB/dt from sets G (in red) and H (in blue)against observations (thick, grey curve). The dot-dashed lines in the line plots are markers of thethresholds used in the
Pulkkinen2013 study for their event-based analysis. The background ofthe line plots are coloured by Kp , similarly to Figure 5. The dark shaded background regions aretimes when the respective magnetometer was in the nightside.–34–anuscript submitted to Space Weather
Figure 8.
Impact of changes to the auroral conductance on ∆ B predictions - (a) (Left) Lo-cation of Yellowknife (YKC) and Newport (NEW) magnetometer stations mapped in geographiccoordinates with the SWMF auroral boundary demarcated using the thick blue line. (Right) Raw∆ B observations at a 1-minute cadence at YKC and NEW. (b) Comparison of predicted ∆ B from Hi-Res
SWMF simulations against observations at YKC, and (c) at NEW. Both subplotscompare results from simulation sets G (in red) and H (in blue) against observations (in black).(d) Comparing contribution of individual current sources in the simulated ∆ B at (i) YKC and(ii) NEW. The contributions from Hall currents are in blue, Pedersen currents in light blue, FACsin red, and MHD in orange. The background of the line plots are coloured by Kp , similarly toFigure 5. The dark shaded background regions are times when the respective magnetometerwas in the nightside. (e) Dial plots of modelled FACs (top row) and Hall Conductance (bottomrow) in the Northern hemisphere from simulation set H at the same time instances as the DMSPpasses in Figure 7. –35–anuscript submitted to Space Weather
Figure 9.
Heidke Skill Score (HSS) Performance of all SWMF simulation variants at ascend-ing dB/dt predictions for all events from Table 1(a). (a) Comparison of simulation sets A (inred), B (in blue), E (in yellow) and F (in green) illustrating the impact of dataset expansion. (b)Comparison of simulation sets C (in red), D (in blue), G (in yellow) and H (in green) displayingthe overall impact of dataset expansion with oval adjustments. Note the y-axis in (a) and (c)does not start at zero. –36–anuscript submitted to
Space Weather
Figure 10.