To understand why this matters, it helps to start with the quantity being measured. The ionosphere forms where solar and cosmic radiation strip electrons from atoms in the mesosphere, thermosphere and exosphere, creating a plasma of free electrons and ions. The Total Electron Content, or TEC, counts the total number of electrons along a signal path, typically between a Global Navigation Satellite System satellite and a ground receiver. Because GPS and other positioning signals must traverse this charged region, TEC is both a nuisance for navigation accuracy and a remarkably sensitive diagnostic of what is happening overhead. Variations in TEC are driven primarily by space weather, including solar wind, solar flux and geomagnetic storms, but a growing body of research on Lithosphere-Atmosphere-Ionosphere Coupling, or LAIC, suggests that large earthquakes may perturb the ionosphere as well, through mechanisms that propagate energy and electric charge upward from the fault zone before and during rupture.

The idea that earthquakes leave traces in the sky is not new, but it remains contentious. Preparation zones around large faults can span hundreds of kilometers, and proposed coupling mechanisms range from acoustic-gravity waves launched by ground motion to electromagnetic effects associated with stress accumulation and radon release. What has been missing is a rigorous way to separate the seismic component of ionospheric variability from the much larger space-weather background. This is precisely the gap the new study addresses. Rather than simply hunting for anomalies, the researchers built predictive models of TEC using only solar and geomagnetic inputs, then evaluated how well those models captured the TEC behavior observed around the three Japanese earthquakes. Any systematic shortfall or distinctive pattern in the residuals becomes evidence of ionospheric disturbance tied to the seismic events themselves.

The data backbone of the study comes from three Japanese GNSS stations: USUD, AIRA and ISHI. True TEC values for these stations were obtained from IONOLAB, a well-established service for automatic near-real-time estimation of GPS-derived TEC. Solar and geomagnetic drivers were drawn from NASA’s OMNIWeb database and included the solar wind speed, the F10.7 solar flux index, the Disturbance Storm Time index known as Dst, and the Ap index of geomagnetic activity. Earthquake details were compiled from the United States Geological Survey. Together these inputs form a multivariate prediction problem: given the current state of the Sun and the geomagnetic field, what should the TEC be at each station? Deviations between prediction and observation during the earthquake windows then become the object of analysis.

Three machine learning architectures formed the core of the comparison. The first was a Gated Recurrent Unit network, a recurrent neural network designed to retain temporal dependencies in sequential data, trained with the Root Mean Square Propagation optimizer, or RMSProp, which adapts learning rates per parameter and is well suited to the noisy gradients typical of geophysical time series. The second was a Random Forest, an ensemble of decision trees introduced by Leo Breiman in 2001, whose hyperparameters were optimized using a Particle Swarm Optimizer, a bio-inspired search method that mimics the social behavior of bird flocks to explore the parameter space efficiently. The third was XGBoost, the extreme gradient boosting algorithm of Chen and Guestrin, optimized here with a Genetic Algorithm, an evolutionary search technique that iteratively selects, crosses and mutates candidate hyperparameter configurations. These three were benchmarked against NeQuick, a physics-based empirical model of the ionosphere widely used in the GNSS community.

Evaluation rested on four complementary statistical metrics: Root Mean Square Error, Mean Absolute Error, Mean Absolute Percentage Error and the Symmetric Mean Absolute Percentage Error. RMSE penalizes large errors heavily, MAE treats all errors equally, and the percentage-based measures express accuracy relative to the magnitude of the true values, which matters when TEC varies strongly with latitude, season and solar activity. The authors also performed cross-validation on Global Ionospheric Map data for the two best-performing machine learning models, a step that guards against the possibility that strong results reflect overfitting to a single dataset rather than genuine predictive skill. Cross-validation confirmed what the event-based comparisons suggested: XGBoost held its advantage when tested on data it had not been tuned against.

The headline result is that across all three earthquakes, the genetically optimized XGBoost model achieved the lowest RMSE and MAE values of any approach tested, consistently outperforming the GRU network, the swarm-optimized Random Forest and the NeQuick model. Among the remaining methods, Random Forest beat the GRU, an interesting outcome given that the recurrent network is explicitly designed for time series while the tree ensemble is not. The authors suggest that all models were able to capture TEC variations associated with the seismic events, which implies that the ionospheric response to these earthquakes is structured enough to be learned from solar and geomagnetic context, and that deviations from that learned baseline carry information about the lithosphere’s influence on the upper atmosphere.

There are several reasons why a gradient-boosted tree model might outperform a recurrent network in this setting. XGBoost builds an additive ensemble of shallow trees, each correcting the errors of its predecessors, and this architecture handles heterogeneous input features, such as the mix of solar wind speed, flux measurements and geomagnetic indices used here, without requiring the inputs to be sequenced or normalized to the same temporal scale. Boosting also tends to be robust with modest training sets, which is a real constraint in earthquake studies where the number of usable large events is inherently limited. The genetic algorithm’s global search over hyperparameters, including tree depth, learning rate and regularization strength, may have found configurations that gradient-based tuning would miss. The GRU, by contrast, must learn its temporal filters from data, and with only a handful of earthquake windows available for training, the recurrent model may simply have had too little sequence data to exploit its architectural advantage.

The broader significance of the work lies in the convergence of two research communities. Space-weather modelers have long sought better TEC prediction, because unmodeled ionospheric delay is one of the largest error sources in precise positioning and satellite navigation. Seismologists, meanwhile, have spent decades searching for reliable precursory signals, and ionospheric anomalies have been reported before earthquakes in Japan, Peru, Türkiye, Morocco, Cyprus, Alaska and the Himalayas, using ground GNSS networks and satellites such as DEMETER. By showing that machine learning models trained on space-weather inputs can characterize the expected ionospheric state with high fidelity, the study provides a cleaner statistical framework for asking whether specific anomalies exceed what space weather alone can explain. The LAIC hypothesis gains a sharper test, and navigation science gains better models in the same stroke.

Cautions remain, and the authors are careful not to overclaim. Three earthquakes, however well studied, do not establish a universal precursor signature, and the field has a long history of anomalies that looked compelling for one event but failed to generalize. The coupling physics connecting fault rupture to ionospheric electron content is still debated, and confounding factors such as volcanic activity, which Japan has in abundance, complicate attribution. What this study demonstrates is methodological progress: a validated, cross-checked machine learning pipeline that predicts TEC more accurately than a standard empirical model during seismic episodes, with XGBoost and genetic optimization at the top of the leaderboard. If future work extends this framework to more events, more regions and longer archives of GNSS data, the dream of reading warning signs from the ionosphere before the ground moves may edge closer to operational reality. For now, the sky above Japan has proven to be a measurable, modelable witness to the violence below.

Subject of Research: Machine learning analysis and prediction of ionospheric TEC anomalies associated with the 2024 Japan earthquakes

Article Title: Analysis of ionospheric TEC anomalies and prediction using XGBoost, GRU and RF algorithms associated with 2024 Japan Earthquakes (Mw > 7.0)

Article References: Mukesh, R., Kiruthiga, S., Rubashri, J., Fathima, S. R., Priyavarshini, P., Dass, S. C., Sahanaa, A. R. S., Safana, S., Karthick, S., & Ratnam, D. V. (2026). Analysis of ionospheric TEC anomalies and prediction using XGBoost, GRU and RF algorithms associated with 2024 Japan Earthquakes (Mw > 7.0). Earth Science Informatics, 19(11), Article 184. https://doi.org/10.1007/s12145-026-02220-9

Image Credits: AI Generated

DOI: 10.1007/s12145-026-02220-9

Keywords: ionosphere, total electron content, earthquake precursors, XGBoost, genetic algorithm, GRU, Random Forest, PSO, RMSProp, NeQuick, Japan earthquakes, lithosphere-atmosphere-ionosphere coupling

Injection-induced earthquakes, or IIEs, have become a growing concern in regions where oil and gas operations pump large volumes of fluid into the subsurface. Most induced events are small, but some have reached magnitudes capable of causing damage, prompting jurisdictions to adopt operational safeguards. The most common of these is the traffic-light system, which calls for injection rates to be reduced or halted altogether when earthquakes of increasing size occur near an operation. The logic behind such systems rests on two assumptions: that smaller foreshocks can serve as warning signs of a larger event to come, and that stopping injection quickly enough can prevent that larger event from occurring. Yet until now, scientists have lacked a clear explanation for why some induced earthquakes are preceded by swarms of foreshocks while others arrive with little or no advance notice, and how foreshock behavior connects to the physical processes that culminate in a mainshock.

The stakes of this gap in understanding are not abstract. In Western Canada, a surge of induced seismicity over the past decade has drawn sustained attention from regulators, operators, and researchers alike. Events felt by residents near communities such as Fox Creek in Alberta drew international scientific scrutiny after being linked to hydraulic fracturing operations, and similar concerns have arisen in other jurisdictions where wastewater disposal or stimulation activities have coincided with earthquake swarms. The economic and social consequences are tangible: operational shutdowns carry significant costs for industry, while residents living near active wells face repeated shaking, concerns about structural damage, and questions about long-term risk. A more reliable way to anticipate when small events might escalate into damaging ones would therefore carry value well beyond the scientific community.

To address those questions, Bei Wang and colleagues turned to a decade of high-quality seismic recordings from western Canada, a region that has experienced a well-documented rise in induced seismicity tied to fluid injection. From that dataset, the team examined foreshock activity preceding 77 IIEs of magnitude 3 or greater. The scale and quality of the records allowed the researchers to detect and characterize foreshocks that would have gone unnoticed in less sensitive monitoring, providing an unusually detailed statistical foundation for analyzing how foreshock sequences develop in space and time before larger ruptures.

The results were striking in their consistency. Roughly 92% of the earthquakes in the study — 71 of the 77 events — were preceded by identifiable foreshocks. In other words, detectable precursory seismicity was the norm rather than the exception for injection-induced earthquakes of significant magnitude in this region. That finding carries immediate practical weight, because it supports the basic premise underlying traffic-light protocols: that smaller earthquakes accompanying injection operations are, far more often than not, part of a build-up toward a larger event rather than isolated, self-limiting rumblings.

The near-ubiquity of foreshocks also reframes how small events during injection should be interpreted. Under many current protocols, a magnitude threshold triggers a response — caution, reduced injection, or shutdown — but the underlying assumption has often been that most small earthquakes will simply die out on their own. The new results suggest that in this region, at least, small events during injection are statistically likely to be early expressions of a developing rupture sequence. That does not mean every foreshock presages a damaging mainshock, but it does mean the base rates that inform protocol design should be revisited with this high foreshock-precedence figure in mind.

But the study went beyond simply counting foreshocks. Wang and colleagues found that foreshock productivity — how many foreshocks occur, and how vigorous the sequences are — along with their spatial and temporal patterns, reflected the interplay of three key factors: the fluid injection itself, the seismogenic index of the affected volume of rock, and the stress state of the faults involved. In essence, the character of foreshock activity encodes information about how close a fault is to failure, how susceptible the surrounding rock is to seismic rupture, and how the injected fluid is perturbing the system. That interplay helps explain why foreshock behavior varies from one induced earthquake to another: different combinations of injection history, rock properties, and fault stress conditions produce different precursory signatures.

The seismogenic index, a concept drawn from statistical seismology, is worth unpacking for readers unfamiliar with it. In broad terms, it captures how prone a given volume of rock is to producing seismicity in response to a given amount of fluid-induced perturbation. Two sites receiving similar injection volumes can behave very differently: one may respond with abundant small earthquakes, while the other remains nearly silent because its faults are oriented unfavorably or its stress conditions are far from failure. By tying foreshock productivity to this index, the study provides a physical rationale for why precursory activity is so variable — and why a warning threshold tuned in one basin may misfire in another.

Perhaps the most consequential contribution of the work is the identification of three distinct rupture nucleation models describing how injection-induced earthquakes get underway. The first is a fluid-driven cascade, in which injected fluid drives a chain of small seismic events that progressively leads to a larger rupture. The second is fluid-driven pre-slip with an intact source asperity, in which fluids promote aseismic — silent, non-radiating — slip on parts of a fault while a locked patch, or asperity, remains intact until it finally breaks in the mainshock. The third is injection-weakened pre-slip, in which fluid injection weakens the fault sufficiently that pre-slip develops and ultimately cascades into rupture. Together, these models illustrate the diverse ways in which fluids can promote aseismic slip and transfer stress through the fault system before a mainshock strikes.

The distinction among these models matters because each implies a different warning signature. A fluid-driven cascade should announce itself through an accelerating sequence of detectable small earthquakes clustered near the injection point. Pre-slip models, by contrast, can unfold largely beneath the detection threshold of seismic networks, with foreshocks appearing only sporadically as the silent slip loads stress onto locked patches. An injection-weakened fault may show a more gradual drift toward instability. Recognizing which model best describes a developing sequence — and the study shows that more than one pathway can operate even within a single region — is a step toward interpreting real-time monitoring data with physical rather than purely statistical rules.

The recognition that fluids can drive aseismic processes in the run-up to an induced earthquake is particularly important. Foreshocks recorded at the surface are only the audible portion of a broader deformation process; silent slip, accelerated by fluid pressure changes, can load stress onto locked fault patches without generating any seismic signal at all. This means that seismic monitoring alone may provide an incomplete picture of the state of a fault during injection operations. The authors accordingly conclude that risk mitigation for induced earthquakes should be informed by combined seismic and geodetic monitoring — pairing earthquake detection with measurements of ground deformation that can reveal aseismic slip in progress.

Geodetic techniques such as satellite-based radar interferometry and continuous GPS stations can detect millimeter-to-centimeter scale deformation of the ground surface, deformation that often accompanies slow slip on faults at depth. Integrating such measurements with dense seismic arrays would give operators and regulators a two-channel view of fault behavior: the seismic channel capturing radiating events, and the geodetic channel capturing the silent loading that may precede them. The challenge, as the authors acknowledge, is that geodetic data are not routinely collected or ingested in real time by most current monitoring programs, making this recommendation as much an operational roadmap as a scientific conclusion.

The implications for traffic-light protocols are significant. The finding that more than 90% of significant induced earthquakes in western Canada had foreshocks validates the use of small-event thresholds as warning triggers. At the same time, the study shows that foreshock productivity and patterns vary according to the seismogenic index and fault stress state, meaning that a uniform, one-size-fits-all threshold may perform differently from one region, formation, or fault to another. A traffic-light scheme calibrated for one geological setting could under-warn in another, or impose costly operational shutdowns where precursory activity does not actually signal a large rupture. The authors argue that IIE risk mitigation should instead be tailored to regional and local conditions.

The study also carries implications for fundamental earthquake science. Foreshocks preceding natural tectonic earthquakes remain an active and contested research area, with debate over whether they reflect a genuine nucleation process or are simply part of a self-similar earthquake cascade. Injection-induced sequences offer a valuable natural laboratory for these questions because the timing and location of the forcing — the injection — are comparatively well constrained. By classifying foreshock sequences into three physically distinct nucleation models, the study demonstrates that multiple pathways to rupture can operate even within a single region and dataset, underscoring the complexity of how earthquakes begin.

Several caveats accompany the findings. The analysis draws on a decade of recordings from western Canada, and the extent to which the 92% foreshock-precedence figure and the three nucleation models apply to other regions with different geology, injection practices, or monitoring networks remains to be established. Detecting foreshocks also depends heavily on the sensitivity and density of the seismic network; regions with sparser instrumentation would likely identify a lower fraction of events with precursors, so cross-regional comparisons must be made carefully. In addition, while the study links foreshock patterns to the seismogenic index and fault stress state, translating those insights into operational decision rules for individual injection sites will require further work, including real-time integration of geodetic data that many current monitoring programs do not routinely collect.

Nevertheless, the research marks a meaningful advance in the science of induced seismicity. By demonstrating that foreshocks are nearly ubiquitous before significant injection-induced earthquakes in western Canada, and by articulating concrete physical models for how fluids shepherd faults toward rupture, Wang and colleagues have provided both a diagnostic framework and a practical directive. The combination of seismic and geodetic monitoring, tailored to the seismogenic character of each region and fault, offers a path toward traffic-light systems that are more predictive and less blunt — a development with clear stakes for communities and industries operating where the subsurface is increasingly called upon to absorb vast quantities of injected fluid.