Multi-Source Magnetosphere-Ionosphere Current System Inversion

B. Stephenson & Argus | May 2026 | Pipeline Session 70

1. The Problem: Three Partial Views of One System

The magnetosphere-ionosphere (M-I) current system is a single coupled entity that we observe through three fundamentally different instruments, each measuring a different physical quantity:

• SuperMAG (~50 ground magnetometers): measures δB, the magnetic perturbation at Earth's surface. This is the Biot-Savart integral of ALL currents above the station — field-aligned (FAC), Hall, Pedersen, and magnetospheric. The measurement conflates multiple current systems into a single vector.
• GPS TEC (2.5°×5° global grid): measures total electron content, the column-integrated ionospheric density. TEC is related to conductance (σ_H, σ_P) which controls how much current flows for a given electric field: J = σ · E.
• SWMF/BATS-R-US (12 virtual magnetometers): provides the full MHD forward model with Biot-Savart decomposition into FAC, Hall, Pedersen, and magnetospheric components. Available for 6 events from CCMC.

Each source alone is underdetermined: SuperMAG can't uniquely decompose δB into current systems, TEC can't directly measure currents, and the model has conductance biases. But together they form an overdetermined system.

2. The Overdetermination Argument

Discretize the auroral ionosphere into voxels: 6 latitude bins × 72 longitude bins × 3 altitude layers = 1,296 unknowns. Count the constraints:

The physics provides the critical coupling: electron density (TEC) determines conductance (σ), conductance and electric field determine current (J = σ·E), current determines δB (Biot-Savart). These links reduce the effective degrees of freedom because density unknowns and current unknowns are not independent.

3. The SWMF Rosetta Stone

BATS-R-US provides the full forward model: for each timestep at each virtual magnetometer, it reports δB decomposed into FAC, Hall, Pedersen, and magnetospheric contributions. This decomposition is the Rosetta Stone that lets us calibrate observational inversions.

3.1 The FAC-Hall Cancellation

A critical finding from the SWMF analysis (Pipeline Session 70, prior work): FAC and Hall contributions to ground δB partially cancel. FAC alone shows strong spatial coherence (Moran's I ≈ 0.15), but the Hall contribution has opposite polarity, driving the total toward zero. This cancellation makes ground δB a poor indicator of individual current system strength — but the E-component of δB, which is FAC-dominated, preserves more spatial structure.

3.2 Calibration Strategy

At the 4 triple-overlap events (where all three sources exist), we:

1. Run SECS inversion on SuperMAG data
2. Compare SECS-derived currents to SWMF decomposition at the 12 overlap stations
3. Compute TEC-derived conductance, compare to SWMF's internal conductance
4. Calibrate the TEC→σ conversion and SECS regularization

Then apply the calibrated method to the 20 dual-source events where no SWMF exists.

4. The Pipeline: TEC → Conductance → Current → δB

4.1 TEC to Conductance

We convert VTEC to Hall/Pedersen conductance using a two-component model:

• Solar EUV contribution (Moen & Brekke 1993): Σ_P ≈ 0.53 + 2.08·cos(χ)^0.55 mho, where χ is solar zenith angle.
• Auroral enhancement: excess TEC above the solar baseline indicates particle precipitation enhancing E-region density. We model ΔΣ_P ≈ 0.4 · ΔTEC · exp(-(λ-67°)²/128), with a Gaussian auroral zone weighting.

4.2 SECS Inversion

Spherical Elementary Current Systems (Amm & Viljanen 1999) decompose ionospheric currents into divergence-free elementary systems placed on a grid. Each DF-SECS produces a known magnetic signature at ground, forming a linear transfer matrix:

B_ground = T · I_SECS

We solve this via Tikhonov regularization: minimize ||T·I - B_obs||² + α·||I||². The regularization parameter α is scaled relative to max(diag(T^TT)) to handle the unit mismatch (T values ~10^-13 T/A).

4.2b Ground Induction Correction

Ground magnetometers don't just see ionospheric currents — the time-varying external field induces telluric currents in the conducting Earth. These produce a secondary magnetic field that enhances the observed horizontal components:

B_obs = B_external · (1 + Q)

The amplification factor Q depends on the skin depth δ relative to the ionospheric source height h (110 km):

Q = h / (h + δ)    where    δ = √(2 / μ₀σω)

Higher conductivity σ means shorter skin depth, which places the image currents closer to the surface, producing stronger enhancement. Since each magnetometer sits on fixed geology, Q is a per-station constant at a given frequency band — compute once, apply forever.

4.3 Conductance-Constrained Inversion

When TEC data is available, we add a conductance penalty: minimize ||T·I - B||² + α||I||² + βΣ(I_j/σ_j)². Higher conductance at a SECS position permits larger currents, while low conductance regions are penalized for large currents.

5. Results: The Six Pairwise Comparisons

5.1 SM ↔ TEC: Time-Lagged Cross-Correlation

The raw Spearman correlation between TEC and δB spatial coherence at zero lag is near zero (r ≈ 0.04 from prior analysis). But this doesn't mean they're unrelated — it means there's a time lag in the causal chain.

We compute cross-correlation between mean auroral TEC anomaly (excess above solar baseline) and median |δB| at all SuperMAG stations, sweeping lags from -6 to +6 hours at 30-minute resolution:

5.2 SM ↔ SWMF: Model-Observation Comparison

At the 12 shared virtual magnetometer stations:

• Halloween 2003: RMSE = 299 nT, r(N) = 0.43 — SWMF captures pattern but overestimates amplitude
• Aug 2001: RMSE = 38 nT, r(N) = -0.16 — moderate storm, poor amplitude correlation
• Aug 2005: RMSE = 49 nT, r(N) = 0.55 — best model performance

5.3 Resolution Sensitivity

We test effective resolution via two methods:

Point Spread Function (PSF): Place a unit current at each SECS position, compute synthetic δB, invert, measure FWHM of recovered peak. Result: median FWHM = 1,400 km with the current 41-station SuperMAG network.

Withheld-Station Test: Train SECS on N stations, predict δB at withheld stations, measure RMSE. Key finding:

6. Animated M-I Visualization

The pipeline generates animated polar stereographic projections for each event, overlaying all available data sources. Below: Halloween 2003 at two epochs.

Green/yellow colormap: TEC-derived Hall conductance (Σ_H in mho). Orange arrows: SuperMAG δB vectors. Blue arrows: SECS-inverted equivalent ionospheric current. Gray circles: 50°, 60°, 70°, 80°N latitude lines.

7. The Corrected-BATS Prediction

A key implication of this work: BATS-R-US with TEC-corrected conductance should outperform uncorrected GAMERA, despite GAMERA's superior numerics (7th-order vs 2nd-order spatial reconstruction).

The reasoning: conductance is the dominant error source in MHD ground perturbation prediction. GAMERA improves the MHD solution (sharper current sheets, less numerical diffusion) but uses the same simplified conductance model as BATS-R-US. If we replace the model conductance with TEC-observed conductance, we fix the biggest error — the remaining numerical error (which GAMERA would reduce) is secondary.

8. Path to Higher Resolution

Current resolution (~1,400 km) is limited by SuperMAG station spacing (~500-1000 km in the auroral zone) and the SECS grid density. The path forward:

8.1 Dense Magnetometer Arrays (~200 km)

IMAGE (Scandinavia), CARISMA (Canada), and other regional arrays have ~100-200 km station spacing. Incorporating these would immediately improve the PSF to ~200 km in their coverage regions.

8.2 High-Resolution TEC (~50-100 km)

The CORS network provides ~1800 GPS receivers across North America at ~50-100 km spacing. With raw RINEX data (slant TEC, not GIM), this enables GPS ionospheric tomography with 3D altitude resolution. Combined with dense magnetometers, the overdetermined inversion could achieve ~50 km effective resolution.

8.3 MMS Tetrahedron Calibration (~10-100 km)

The MMS (Magnetospheric Multiscale) mission has 4 spacecraft in a tetrahedron at 10-100 km separation, measuring current density via the curlometer technique (j = ∇×B/μ₀). When MMS passes through the M-I current sheet during our 2015 events, the in-situ current measurement provides a "ground truth" at scales far below what the ground network can resolve. This calibrates the point spread function and tests whether TEC-constrained inversion can recover fine structure.

9. Time-Lag Variation as Storm Diagnostic

The time lag between TEC and δB is not constant — it varies with storm phase and driver mechanism. This variation is itself a diagnostic:

• Short lag (~0-30 min, TEC leads): Fast precipitation event. E-region ionization builds quickly, current responds promptly. Typical of substorm onset.
• Long lag (~2-6 hr, TEC leads): Gradual conductance buildup from sustained particle precipitation during prolonged southward IMF. The Aug 2001 result (+5.5h).
• Negative lag (δB leads TEC): Electric-field driven. Enhanced convection drives currents before precipitation enhances ionization. Or: post-storm decay phase where activity drops but ionization persists.
• Anti-correlation: Negative-phase storm. Extreme activity depletes ionospheric density through thermospheric composition changes. The Bastille Day result.

10. What's Novel

1. Joint TEC + SuperMAG inversion with MHD decomposition as prior — individual techniques exist; the combination is new.
2. Moran's I correlogram as structural constraint on inversion — spatial autocorrelation used as a validation metric, not just a diagnostic.
3. Time-lag cross-correlation between TEC anomaly and ground δB revealing driver-mechanism-dependent delays.
4. SWMF decomposition as Rosetta Stone for observational inversion calibration — transfer learning from model to observations.
5. Conductance-corrected MHD prediction — replacing model conductance with TEC-observed conductance to improve ground perturbation accuracy.

11. Pipeline Architecture

Seven Python modules, all in scripts/:

24 events processed in ~37 seconds. 288 animated SVG frames generated, all <500 KB.

12. Next Steps

1. Higher-resolution TEC (15-min IGS rapid products) for finer time-lag analysis
2. Incorporate IMAGE/CARISMA dense arrays for sub-500 km resolution
3. Raw RINEX GPS data for slant-TEC tomography (3D density reconstruction)
4. MMS conjunction search for in-situ current calibration at 2015 events
5. GAMERA runs for head-to-head comparison: corrected BATS vs uncorrected GAMERA
6. AMPERE satellite FAC data as independent constraint on field-aligned currents
7. Systematic time-lag/storm-phase analysis across all 24 events

Generated by the multi-source M-I inversion pipeline, Session 70. Pipeline: 4 data loaders + SECS inversion + TEC conductance + 6-pair combination analysis.