Earthquake Forecasting: Alternative Precursors Explored

The quest to predict earthquakes—specifically, the ability to deterministically specify the time, location, and magnitude of impending large shallow events—remains one of the most contentious and elusive challenges in the geosciences. For over a century, seismology has oscillated between fervent optimism, driven by new models or observational technologies, and deep pessimism, precipitated by the failure of those models to prevent catastrophic loss of life. Established The current official consensus, maintained by agencies such as the United States Geological Survey (USGS) and the Japan Meteorological Agency, is that reliable short-term prediction is not currently possible; under some theoretical frameworks it may be inherently unachievable because of the chaotic nature of rupture nucleation (U.S. Geological Survey, n.d.). Consequently the operational focus has shifted toward Probabilistic Seismic Hazard Assessment (PSHA), Earthquake Early Warning, and—most relevant here—time-dependent forecasting.

This prevailing narrative of "impossibility" is challenged by a parallel stream of research at the intersection of geophysics, atmospheric science, heliophysics, and statistical physics. Contested This body of work argues that the failure to predict may stem not only from inherent randomness in the crust but also from an overly narrow focus on mechanical seismic precursors such as foreshocks and crustal strain. Over the last two decades, and particularly with dedicated satellite missions like DEMETER, Swarm, and CSES, many studies have reported non-seismic anomalies in the lithosphere, atmosphere, and ionosphere before large earthquakes. The evidential status of those anomalies remains uneven: co-seismic coupling is well established, but pre-seismic diagnostic skill still depends heavily on retrospective filtering, source discrimination, and—rarely—prospective testing.

This report is a feasibility study of these alternative methodologies. It examines the Lithosphere-Atmosphere-Ionosphere Coupling (LAIC) model, the controversial correlations between space weather and seismicity, the historical "Blot Hypothesis" of deep-focus precursors, ground-based electromagnetic methods, and the statistical machinery used to validate any of them. The throughline is a single discipline: a candidate signal earns operational status only when it adds out-of-sample skill over the seismicity-rate and seasonal baselines that already exist—and, as the closing sections show, none has yet done so prospectively.

1.1 The Semantic Shift: Prediction versus Forecasting

To evaluate feasibility rigorously, one must disentangle the terminology that confuses the public and policymakers. Established "Earthquake prediction" is classically defined as a statement specifying the time, location, and magnitude of a future event with narrow uncertainties (e.g., "a magnitude 7.0 will strike Tokyo on Friday"). This strict definition has set a bar that no current technology can clear. "Earthquake forecasting," by contrast, is a probabilistic assessment—akin to modern meteorology—defining an increased likelihood of an event over a specific time window and spatial area (e.g., "a 60% probability of M > 6.0 in the Nankai Trough within 30 days").

The feasibility explored here concerns the latter: moving from the multi-decadal probabilities of PSHA maps—useful for building codes but useless for evacuation—to time-dependent forecasts on the scale of days to weeks. That intermediate-term capability, if achieved, would be a genuine advance in disaster risk reduction.

1.2 The Historical Context of "Impossibility"

The pessimism of modern mainstream seismology is rooted in the "dilatancy-diffusion" era of the 1970s, when researchers believed volume changes in rock before failure would inevitably produce detectable precursors such as changes in the Vp/Vs ratio and groundwater fluctuations. Established The apparent success of the 1975 Haicheng prediction, based on foreshocks and groundwater anomalies, seemed to validate the approach; the failure to anticipate the devastating 1976 Tangshan earthquake shattered that confidence.

By the 1990s the theoretical landscape shifted toward Self-Organized Criticality (SOC). Contested In the SOC view the crust sits perpetually near a critical state where any small perturbation can trigger rupture, and the final size of an earthquake is set by the heterogeneous stress field encountered as the rupture propagates, not by the nucleation. If that model holds strictly, a magnitude 8 begins exactly like a magnitude 1, and magnitude-scaling precursors are physically impossible—the strong-skeptic position later argued in Science by Geller et al. (1997).

Hypothesis Yet this view is not universal. The "critical point" hypothesis holds that while the exact moment of failure is stochastic, the approach of a large system toward instability can be marked by accelerated moment release and the synchronization of diverse physical fields. It is within this framework—not against it—that non-seismic precursors find their physical justification. The remainder of this report weighs the evidence for each.

Hypothesis The LAIC model is the most comprehensive physical framework attempting to explain how a subterranean process could generate signals in the upper atmosphere and ionosphere. Unlike purely statistical correlations, LAIC proposes a sequential cause-and-effect chain transporting energy from the ground to geospace, uniting thermal hotspots, electromagnetic emissions, and ionospheric irregularities under one mechanism (Picozza, Conti & Sotgiu 2021). Its physical plausibility is real; its operational reliability is what the rest of this section interrogates.

2.1 The Physical Mechanism: The Ionization Pathway

Hypothesis The central engine of LAIC is the ionization of the lower atmosphere over the preparation zone. As stress accumulates, micro-fracturing (dilatancy) increases crustal permeability, opening pathways for gases—CO₂, CH₄, H₂, and critically the radioactive noble gas Radon-222—to escape to the surface. Radon-222 (half-life 3.8 days) is a potent ionizer through alpha decay; a pulse released from a stressed fault can generate large numbers of air ions in the planetary boundary layer.

This initial ionization is proposed to trigger a cascade:

1. Cluster-ion generation: new air ions attach to water molecules, forming heavy cluster ions.
2. Latent-heat release: condensation onto ions is exothermic, predicted to appear as increased Outgoing Longwave Radiation (OLR) and Surface Latent Heat Flux over the epicentral area—a candidate physical cause for reported "thermal anomalies."
3. Conductivity change: a dense cloud of heavy cluster ions alters the air column's conductivity.
4. Global-electric-circuit perturbation: the Earth behaves as a spherical capacitor with a near-surface field of ~100–120 V/m; the local conductivity change disrupts this circuit.
5. Ionospheric coupling: through electric-field penetration, the disturbance is proposed to propagate into the E and F layers, altering electron density and Total Electron Content (TEC). Because the ionosphere is a magnetized plasma, such irregularities drift along field lines, so a satellite anomaly may be shifted toward the magnetic equator relative to the epicenter.

2.2 Observational Evidence: The Thermal Branch

Contested The thermal anomaly is one of LAIC's most testable predictions. Historically, thermal-infrared anomalies before earthquakes were dismissed as coincidental weather; LAIC supplies a specific physical cause (latent heat of ion hydration). Retrospective analyses of major events—e.g., the 2008 Wenchuan and 2015 Nepal earthquakes—have reported localized OLR increases of roughly 20–100 W/m² appearing 2–14 days before the mainshock, often stationary relative to the fault rather than migrating like weather systems (Picozza, Conti & Sotgiu 2021).

The reliability of thermal precursors is complicated by competing heat sources—anthropogenic emissions, seasonal variation, soil moisture—so background-subtraction algorithms are required to isolate any transient signal. Higher-resolution sensors sharpen case studies without resolving causation: ECOSTRESS thermal-monitoring time series over the restless Campi Flegrei caldera illustrate how better sensors improve anomaly detection without, on their own, establishing a seismic cause (Piscini & Fidani 2025). Contested The strongest recent geochemical case is a helium-isotope anomaly recorded in groundwater before the 2024 Noto Peninsula earthquake by a pre-existing monitoring network—valuable precisely because the instrumentation predated the event, though still a single case rather than a validated rule (Kagoshima et al. 2025).

2.3 The Ionospheric Branch: Satellite Missions and Data

Contested The upper branch of LAIC—ionospheric perturbation—has been the focus of dedicated space missions, because the ionosphere can amplify weak ground-level electric fields through plasma instabilities. The evidence here is a genuine split between positive case studies and negative large-N tests.

2.3.1 The DEMETER Mission (2004–2010)

The French DEMETER satellite was the first spacecraft designed for seismo-ionospheric coupling, measuring ULF/ELF/VLF waves and plasma parameters in low Earth orbit. Contested A statistical analysis of the whole dataset found a weak but significant decrease in ULF wave intensity and ion density over earthquake zones roughly four hours before the shock; however, the signal-to-noise ratio was too low to forecast individual events, and the four-hour timing sits awkwardly against LAIC's "days to weeks" prediction.

2.3.2 The CSES Mission (2018–Present)

The China Seismo-Electromagnetic Satellite (CSES / Zhangheng-1), launched in 2018, carries more sensitive instrumentation. Contested Retrospective analyses report electron-density and electron-temperature anomalies exceeding background thresholds 1–5 days before the February 2023 Turkey M7.8/M7.5 doublet, and synchronous plasma-parameter enhancements before the January 2024 Noto Peninsula M7.6 earthquake. These are retrospective, single-event observations: persuasive individually, but each selected after the earthquake was known.

2.3.3 The Swarm Constellation

ESA's three-satellite Swarm mission, primarily geomagnetic, has been used to cross-validate CSES findings. Contested Machine-learning analyses of Swarm data have reported magnetic anomalies preceding events such as the 2017 Mexico and 2020 Croatia earthquakes, sometimes months in advance—an intriguing but unconfirmed long-preparation signature. A large 2014–2024 study correlating Swarm magnetic-field and TEC data against M4+ earthquakes extends this record while underscoring that most reported associations remain weak and retrospective (Semlali et al. 2025).

Contested The decisive tests are the large-N ones, and they lean negative. A global analysis spanning 129 stations and 57.8 million TEC measurements found no consistent pre-earthquake TEC anomaly (Cullen et al. 2024). Apparent single-station anomalies are largely reproduced by randomized surrogate catalogues once solar and geomagnetic effects are removed (Ikuta & Oba 2022), and at least one widely cited TEC enhancement has been shown to be an artifact of the analysis method rather than a seismic signal (Eisenbeis & Occhipinti 2021). Positive case studies persist—e.g., a reported TEC precursor associated with the 2024 Noto earthquake (Nayak et al. 2024)—but the contrast between abundant positive case studies and negative population tests is the defining tension of the ionospheric branch.

2.4 Challenges to Feasibility and Implementation

Established Even granting the physics, operational LAIC forecasting faces formidable hurdles:

1. Signal specificity and source discrimination. The ionosphere is dominated by solar activity and lower-atmosphere weather. A geomagnetic storm can mask any subtle pre-seismic signal entirely; separating a "seismic" perturbation from a "solar" one demands multi-parameter analysis and real-time space-weather indices (Dst, Kp). During a storm, seismo-ionospheric forecasting reliability falls to near zero.
2. Magnetic conjugacy. Field-line geometry can map a ground-level electric field to the ionosphere at a different latitude—or the opposite hemisphere—complicating any attempt to trace an observed anomaly back to a specific epicenter.
3. The false-alarm problem. Ionospheric anomalies also occur in seismically quiet periods. A high False Alarm Ratio is the central barrier to public use: a system that warns on every anomaly quickly loses trust.

The conservative reading, shared by the field's own critical reviews, is that TEC, TID, radon, thermal, and EM signals belong in research and diagnostic contexts—shown with provenance and contamination flags and benchmarked against seismic baselines—rather than as standalone warnings (Picozza, Conti & Sotgiu 2021).

Contested The proposition that solar activity—flares, coronal mass ejections, the solar wind—could trigger earthquakes is among the most controversial hypotheses in geophysics. Established The USGS position is blunt: "it has never been demonstrated that there is a causal relationship between space weather and earthquakes" (U.S. Geological Survey, n.d.). The skepticism is grounded in energy scales—solar-terrestrial stresses are orders of magnitude smaller than tectonic stresses. The counter-argument is that, in a crust poised near criticality, even a minute, sharply-timed perturbation might tip a fault already at failure.

3.1 The Reverse Piezoelectric Effect Hypothesis

Hypothesis The leading proposed mechanism is the reverse piezoelectric effect. Quartz, abundant in continental crust, deforms when an electric field is applied to it. The triggering chain runs: a solar event compresses the magnetosphere → a geomagnetic storm produces rapid magnetic-field fluctuations → Faraday induction drives geomagnetically induced currents and telluric electric fields through the crust → in quartz-rich fault zones those fields impose a small mechanical strain. The strain is minute (estimated 10⁻⁸–10⁻⁷ in fractional terms); advocates argue that for a fault already at the breaking point, a sudden electromagnetic "jerk" could nonetheless overcome static friction.

3.2 Statistical Evidence and Correlations

Contested Large-scale statistical studies have tried to move this beyond speculation, and the picture is genuinely mixed. A 20-year analysis pairing SOHO solar data with the ISC-GEM catalogue reported a significant correlation between solar-proton density and large (M > 5.6) earthquakes at a ~24-hour lag, with a chance probability quoted below 10⁻⁵ (Marchitelli et al. 2020).

That correlation has not gone unchallenged—and the challenges are the stronger work. Re-analysing 333 M ≥ 7 events (2000–2022), Akhoondzadeh & De Santis (2022) found solar/geomagnetic anomalies in the day before roughly a third of events—but found the same rate in 100 synthetic surrogate catalogues, concluding the apparent link is statistically indistinguishable from chance. An interacting-point-process study placed any genuine proton-flux contribution at an upper bound near 28% of seismic intensity—directional but modest (Lyubushin & Rodionov 2025). And the broadest canonical test of solar-terrestrial triggering found the effect to be statistically insignificant across the global catalogue (Love & Thomas 2013). Other surveys keep the door ajar to a weak space-weather influence under quiet conditions (Sorokin & Novikov 2024), but the weight of rigorous evidence is that a strong, exploitable space-weather trigger has not been demonstrated.

3.3 The Magnitude Problem and Laboratory Validation

Established The primary quantitative critique is decisive on its own terms. Stress changes from geomagnetically induced currents are estimated in Pascals; the stress drop in an earthquake is in MegaPascals—six orders of magnitude larger. The standard reductio is the tides: solid-Earth and ocean tides impose kilopascal-scale stresses twice daily, yet their measurable correlation with seismicity is small. If far larger tidal stresses barely register, far smaller space-weather stresses should register less.

Hypothesis Laboratory work offers a narrow rejoinder: spring-block experiments suggest that the nature of a perturbation can matter as much as its magnitude, with sharp electromagnetic pulses able to trigger slip in faults held at 98–99% of failure stress, apparently by momentarily reducing friction rather than by adding static stress. This keeps the hypothesis alive in principle, but the gap between a benchtop fault at the edge of failure and a heterogeneous crustal fault remains the unbridged problem.

Contested A distinct strand concerns the predictive relationship between deep-focus earthquakes (depth > 300 km) and subsequent shallow events—the historical "Blot Hypothesis" of energy migrating upward through a subducting slab.

4.1 The Legacy of Claude Blot

In the 1960s–70s the French geophysicist Claude Blot, working on Pacific subduction zones, observed apparent regularities in seismic sequences and proposed that a deep "energy phenomenon" migrates upward over months to a year before triggering shallow earthquakes and eruptions, with empirical time-delay formulas. Hypothesis Blot's rigid, deterministic formulas were largely rejected as lacking a mechanism. Established Stress transfer itself is mainstream: a large deep-focus event alters the stress field throughout the slab, and viscoelastic relaxation can redistribute stress over weeks to years. Contested Whether that coupling yields a usable deep-to-shallow precursor on Blot's timescales is far less settled—the broad association is plausible and loosely consistent with his observed delays, but the deterministic precision Blot claimed is considered overstated and the predictive skill has not been independently established.

4.2 The "Omega-Theory" and Global Synchronization

Hypothesis A modern, highly theoretical evolution is the "Omega-Theory" of Jure Žalohar, commercialized by Quantectum. Drawing on Cosserat (micropolar) continuum mechanics, it holds that earthquakes are synchronized by global "tectonic waves" that propagate slowly through the crust, with events occurring where these waves constructively interfere into "synchronization clouds."

This sits on the fringe of mainstream science. The "tectonic waves" central to the theory have not been detected by standard geodetic networks (GPS/InSAR) at the amplitudes required to trigger faults directly. As of 2026 there is no independent, peer-reviewed prospective evaluation of the Quantectum/Omega-Theory forecasts, and no Quantectum model is registered with the Collaboratory for the Study of Earthquake Predictability (CSEP), which requires forecasts to be locked in advance of the events they target; the reported skill rests on self-assessment against the company's own held-out data, without a third-party reference null.

4.3 Lessons from Laboratory "Quakes"

Established In the laboratory, rigorous support for time-to-failure prediction has emerged from rock mechanics. Machine-learning models trained on the continuous acoustic-emission "noise" of a sample can predict the timing of stick-slip "lab quakes" with high accuracy, detecting subtle shifts in the variance and amplitude of the signal long before failure. Hypothesis This is a genuine proof-of-concept that a preparation phase is not silent and carries information about impending failure. The open challenge—unsolved—is detecting the crustal-scale equivalent of those acoustic emissions (micro-seismicity or electromagnetic noise) in the field, where the signal is far weaker and the noise far larger.

Contested Where satellites trade coverage for revisit time, ground-based electromagnetic monitoring offers continuous, high-resolution data—and a correspondingly long, unresolved controversy.

5.1 ULF Magnetic Anomalies

Ultra-low-frequency (0.01–10 Hz) magnetic anomalies are among the most-cited ground precursors. Contested The famous case is the 1989 Loma Prieta (M6.9) earthquake: a Stanford magnetometer at Corralitos reportedly recorded a large rise in ULF activity two weeks before, spiking hours before the mainshock (Fraser-Smith et al. 1990). That observation launched the field of seismo-electromagnetics. It has also not survived scrutiny intact: a later re-analysis attributed the Corralitos signal largely to instrumental and external sources rather than a tectonic precursor (Thomas, Love & Johnston 2009). Broader statistical reviews of observatory ULF data report correlations modestly better than chance for M > 5 events within ~100 km, with anomalies typically 1–2 weeks before—suggestive, but well short of operational reliability, and entangled with the difficulty of removing solar-terrestrial noise.

5.2 The VAN Method and Seismic Electric Signals (SES)

Contested The most persistent ground-based prediction effort is the VAN method (Varotsos, Alexopoulos, Nomicos) in Greece, which monitors "seismic electric signals"—transient geoelectric-potential changes—via dipole networks, and claims numerous successful predictions over three decades. The debate has run for more than thirty years: critics argue the signals are indistinguishable from industrial electrical noise and that the "successes" exploit broad time-and-space windows; supporters argue the chance probability of the hits is vanishingly small. VAN remains one of the few long-running operational prediction attempts, and one of the least resolved.

Established The transition from hypothesis to operation requires rigorous validation. It is not enough to show a precursor existed before one earthquake; one must show it reliably appears before most of them and—crucially—stays quiet when no earthquake follows.

6.1 The Molchan Error Diagram: Measuring Skill

The standard tool for evaluating a prediction algorithm is the Molchan error diagram, which plots the miss rate (earthquakes that occurred with no warning) against the fraction of time under alarm.

• Random guessing falls on the diagonal: alarm 50% of the time and you catch ~50% of events by chance.
• Skillful prediction must fall well below the diagonal—catching a high fraction of events while keeping alarm time low.

Contested Studies of ULF, ionospheric, and synchronization-based precursors typically land "better than random" (probability gains of ~1.5–2.0), confirming the signals carry some physical information. Established But performance plateaus: catching 80% of events might require being on alert 30–40% of the time. For a city, four months a year of "high alert" is operational paralysis—the false-alarm problem is what keeps these methods out of public evacuation decisions.

6.2 The Global Earthquake Forecasting System (GEFS)

Established Recognizing that single precursors are insufficient, the Global Earthquake Forecasting System project (2015–2018), funded largely by the reinsurance industry, attempted to integrate ground signals (radon, groundwater, ULF) and space signals (ionosphere, thermal) into unified, time-dependent probability maps. Its conclusion was sobering: no single signal was robust enough for reliable operational prediction, and multi-parameter observation must be paired with rigorous statistical testing rather than simple signal stacking.

It is worth being precise about how that effort ended. The GEFS organizers' own synthesis concluded that most non-seismic precursor studies, taken individually, do not provide strong enough evidence, and that a systematic search of the DEMETER archive did not reveal a robust precursory pattern (Sornette, Ouillon, Mignan & Freund 2021). No operational successor to GEFS exists; the most ambitious private effort, QuakeFinder, ended data collection in 2023. Established What actually became operational was not precursor-based at all—it was seismicity-rate forecasting. Operational Earthquake Forecasting built on Epidemic-Type Aftershock Sequence (ETAS) models is now the expert first choice in Italy, New Zealand, and the United States (Mizrahi et al. 2024, Reviews of Geophysics) and has been validated over a decade of prospective use. This is the bar everything above must clear: a non-seismic precursor is useful only if it adds out-of-sample skill on top of a seismicity-rate baseline, prospectively—a test the field has not yet passed.

Established The feasibility of forecasting large shallow earthquakes is not a binary "possible/impossible." It is a spectrum of probability gain, and the honest position is that the gain available from non-seismic precursors today is small, fragile, and unproven prospectively.

7.1 From Single Precursors to an Evidence-Gated Ensemble

Hypothesis The most defensible path forward is an evidence-gated ensemble rather than a deterministic precursor stack. No single layer is clean; each has false positives, false negatives, and contamination pathways. A useful system asks, for every layer, whether it adds skill after the strongest baselines are already in hand:

• Baseline seismicity — ETAS, EEPAS, foreshock clustering, geodetic strain, and catalogue history define the benchmark every alternative must beat. Established
• Coupling and proxy signals — TEC/TID, radon, thermal, hydrochemical, Swarm, CSES, and ground EM can supply diagnostic context if their contamination controls are explicit. Contested
• Deep-process context — Blot-style deep-earthquake signals can remain visible where they are measured and backtested, but deterministic migration claims should not be promoted without independent validation. Hypothesis

A signal graduates from "research context" to "forecast method" only if it improves prospective or pseudo-prospective scoring against clear baselines, keeps false-alarm cost tolerable, and preserves its skill after correlated inputs are ablated. That is a high bar, and it is the right one.

7.2 The Future: AI and New Missions

Contested The future of the field lies in larger datasets, careful baselines, and machine learning. New instrumentation keeps arriving: CSES-02 (Zhangheng-1 02) launched on 14 June 2025 into a Sun-synchronous orbit phased 180° from CSES-01, roughly halving the constellation's global revisit time to ~2.5 days and adding an ionospheric photometer and a combined vector-plus-scalar magnetometer; its seismic-precursor results are not yet published.

Established That optimism must be tempered by how machine learning has actually performed on real catalogues. In a controlled benchmark, none of five neural point-process models outperformed a standard ETAS model at earthquake forecasting (Stockman, Lawson & Werner 2024, EarthquakeNPP). A cross-disciplinary review found that a large share of AI–earthquake studies report no baseline comparison, with many high-accuracy claims traceable to data leakage, and concluded there is still no indispensable correlation between earthquakes and the ionosphere (Zhang, Wen, Sornette & Zhan 2025). The failure mode is concrete: in 2026 a Scientific Reports paper claiming high machine-learning prediction accuracy for Los Angeles was retracted because its data were split without respecting time order, inflating the reported accuracy (retraction note, 2026). The promise of AI is real, but so is the discipline it demands: chronological validation, an honest baseline, and prospective rather than retrospective testing.

7.3 A Contemporary Test Case: The 2025 Myanmar Earthquake

Contested The field's current state shows clearly on a single large event. Before the 28 March 2025 Mw 7.7 Myanmar earthquake, one study reported a total-electron-content change beginning about 36 minutes before the mainshock, scaling with magnitude roughly as prior events would predict (Heki & Zhan 2026, Geophysical Journal International). Yet the event occurred during the recovery phase of a geomagnetic storm, and an independent analysis could not rule out that the pre-event ionospheric signal was storm-recovery rather than seismic (Obasanjo et al. 2026). Two lessons follow. First, a 36-minute lead—even if entirely real—is not an evacuation window. Second, the same event produced both a precursor claim and its confounder. That is the field in miniature.

7.4 Final Verdict

Established The official position on deterministic earthquake prediction has not materially changed: exact short-term prediction remains unsupported. What is evolving is the operational forecasting discipline—time-dependent probabilities, prospective testing, and transparent communication of uncertainty. Alternative precursors remain scientifically important because they may describe real lithosphere-atmosphere-ionosphere coupling, but their current role is hypothesis generation and diagnostic context, not warning. A forecast should come from validated probability gain over an honest baseline, never from the presence of an anomaly alone.

Established This report is a literature review, and it inherits the literature's limitations:

• Publication and selection bias. Positive case studies are far easier to publish than null results, and most reported precursors are identified after the earthquake is known. The large-N tests that correct for this lean negative (Cullen et al. 2024; Akhoondzadeh & De Santis 2022).
• Mechanism vs. detectability. Several mechanisms (radon ionization, reverse piezoelectricity, stress transfer) are physically plausible but produce signals near or below the noise floor of current instruments and amid strong solar, atmospheric, and anthropogenic confounders.
• The energy-scale gap remains unresolved. No proposed space-weather mechanism closes the six-order-of-magnitude gap between induced stresses and seismic stress drops; the strongest defense is qualitative (timing/criticality), not quantitative.
• Prospective testing is rare. Almost no non-seismic precursor has been evaluated in a locked, pre-registered, prospective framework against a seismicity-rate baseline. Until that is routine, "better than random" in retrospect should not be read as operational skill.
• Open questions worth watching. Whether CSES-02's improved revisit and instrumentation yields a prospectively testable ionospheric signal; whether geochemical networks (e.g., helium-isotope monitoring) deliver repeatable, pre-instrumented cases; and whether leakage-safe machine learning can extract a signal that genuinely beats ETAS.

• Akhoondzadeh, M., & De Santis, A. (2022). Is the apparent correlation between solar-geomagnetic activity and occurrence of powerful earthquakes a casual artifact? Atmosphere, 13(7), 1131. https://doi.org/10.3390/atmos13071131
• Cullen, L., Smith, A. W., Galib, A. H., Varshney, D., Brown, E. J. E., & Chi, P. J. (2024). A global analysis of pre-earthquake ionospheric anomalies. arXiv. https://doi.org/10.48550/arXiv.2401.01773
• Eisenbeis, J., & Occhipinti, G. (2021). The TEC enhancement before seismic events is an artifact. Journal of Geophysical Research: Space Physics, 126. https://doi.org/10.1029/2020JA028733
• Fraser-Smith, A. C., Bernardi, A., McGill, P. R., Ladd, M. E., Helliwell, R. A., & Villard, O. G. (1990). Low-frequency magnetic field measurements near the epicenter of the Ms 7.1 Loma Prieta earthquake. Geophysical Research Letters, 17(9), 1465–1468. https://doi.org/10.1029/GL017i009p01465
• Geller, R. J., Jackson, D. D., Kagan, Y. Y., & Mulargia, F. (1997). Earthquakes cannot be predicted. Science, 275(5306), 1616. https://doi.org/10.1126/science.275.5306.1616
• Heki, K., & Zhan, W. (2026). Ionospheric changes immediately before the 2025 March 28 Myanmar earthquake. Geophysical Journal International. https://doi.org/10.1093/gji/ggag124
• Ikuta, R., & Oba, R. (2022). How credible are earthquake predictions based on TEC variations? Journal of Geophysical Research: Space Physics, 127. https://doi.org/10.1029/2021JA030151
• Kagoshima, T., Sano, Y., Takahata, N., Kawamoto, Y., Shibata, T., & Li, Y. (2025). Helium isotope anomaly in groundwater prior to the 2024 Noto Peninsula earthquake. Nature Communications. https://doi.org/10.1038/s41467-025-65717-9
• Love, J. J., & Thomas, J. N. (2013). Insignificant solar-terrestrial triggering of earthquakes. Geophysical Research Letters, 40, 1165–1170. https://doi.org/10.1002/grl.50211
• Lyubushin, A., & Rodionov, E. (2025). Quantitative assessment of the trigger effect of proton flux on seismicity. Entropy, 27(5), 505. https://doi.org/10.3390/e27050505
• Marchitelli, V., Harabaglia, P., Troise, C., & De Natale, G. (2020). On the correlation between solar activity and large earthquakes worldwide. Scientific Reports, 10, 11495. https://doi.org/10.1038/s41598-020-67860-3
• Mizrahi, L., Dallo, I., van der Elst, N. J., Christophersen, A., Spassiani, I., & Werner, M. J. (2024). Developing, testing, and communicating earthquake forecasts: Current practices and future directions. Reviews of Geophysics, 62. https://doi.org/10.1029/2023RG000823
• Nayak, K., Romero-Andrade, R., Sharma, G., López-Urías, C., Trejo-Soto, M. E., & Vidal-Vega, A. I. (2024). Evaluating ionospheric total electron content (TEC) variations as precursors to seismic activity. Atmosphere, 15(12), 1492. https://doi.org/10.3390/atmos15121492
• Obasanjo, P., Okoh, D., Awuor, A., Baki, P., & Ochieng, G. (2026). Multi-scale GNSS-TEC and GIM investigation of ionospheric perturbations triggered by the 28 March 2025 Myanmar earthquake preceded by geomagnetic storm. Frontiers in Astronomy and Space Sciences, 13, 1791135. https://doi.org/10.3389/fspas.2026.1791135
• Picozza, P., Conti, L., & Sotgiu, A. (2021). Looking for earthquake precursors from space: A critical review. Frontiers in Earth Science, 9, 676775. https://doi.org/10.3389/feart.2021.676775
• Piscini, A., & Fidani, C. (2025). A novel algorithm for thermal monitoring using ECOSTRESS time series: The case of Campi Flegrei, Naples, Italy. Remote Sensing Letters. https://doi.org/10.1080/2150704X.2025.2459213
• Retraction note: Improving earthquake prediction accuracy in Los Angeles with machine learning. (2026). Scientific Reports. https://doi.org/10.1038/s41598-026-49689-4
• Semlali, B.-E. B., Molina, C., Park, H., & Camps, A. (2025). Global correlation of Swarm satellites' magnetic field and TEC data with M4+ earthquakes between 2014 and 2024. Advances in Space Research. https://doi.org/10.1016/j.asr.2025.02.065
• Sornette, D., Ouillon, G., Mignan, A., & Freund, F. (2021). Preface to the Global Earthquake Forecasting System (GEFS) special issue. The European Physical Journal Special Topics, 230, 1–5. https://doi.org/10.1140/epjst/e2020-000242-4
• Sorokin, V., & Novikov, V. (2024). Possible interrelations of space weather and seismic activity: An implication for earthquake forecast. Geosciences, 14(5), 116. https://doi.org/10.3390/geosciences14050116
• Stockman, S., Lawson, D., & Werner, M. (2024). EarthquakeNPP: A benchmark for earthquake forecasting with neural point processes. arXiv. https://doi.org/10.48550/arXiv.2410.08226
• Thomas, J. N., Love, J. J., & Johnston, M. J. S. (2009). On the reported magnetic precursor of the 1989 Loma Prieta earthquake. Physics of the Earth and Planetary Interiors, 173(3–4), 207–215. https://doi.org/10.1016/j.pepi.2008.11.014
• U.S. Geological Survey. (n.d.). Can you predict earthquakes? https://www.usgs.gov/faqs/can-you-predict-earthquakes
• U.S. Geological Survey. (n.d.). Do solar flares or magnetic storms (space weather) cause earthquakes? https://www.usgs.gov/faqs/do-solar-flares-or-magnetic-storms-space-weather-cause-earthquakes
• Zhang, Y., Wen, C., Sornette, D., & Zhan, C. (2025). Integrating artificial intelligence and geophysical insights for earthquake forecasting: A cross-disciplinary review. arXiv. https://doi.org/10.48550/arXiv.2502.12161