A paradigm for developing earthquake probability forecasts based on geoelectric data
Pith reviewed 2026-05-24 22:22 UTC · model grok-4.3
The pith
Geoelectric signals carry statistically significant information about impending large earthquakes that machine learning can convert into probability forecasts.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
The authors show that geoelectric signals exhibit a seismoelectric relationship with future earthquakes, demonstrated through significance tests on an alarm-based model and binary classification applied to joint station data, allowing machine learning to quantify probabilistic forecasts of large earthquakes.
What carries the argument
An improved alarm-based model combined with binary classification on multi-station geoelectric data, optimized over frequency bands.
If this is right
- Removing the time parameter for coarse-graining improves the model's applicability.
- Joint stations method extends the single station approach for better detection.
- Optimal frequency bands maximize the signal-to-noise ratio for earthquake-related signals.
- Significance tests confirm the predictive potential of the geoelectric signals.
- Machine learning can be used to extract the relationship for probabilistic forecasts.
Where Pith is reading between the lines
- The approach could be tested on datasets from different seismic regions to assess generalizability.
- Integration with other geophysical precursors might enhance forecast accuracy.
- If the relationship holds, it suggests physical mechanisms linking electric signals to stress changes before quakes.
Load-bearing premise
The geoelectric signals identified by the model genuinely precede and predict large earthquakes rather than arising from unrelated noise or biases in data selection.
What would settle it
A replication study on independent geoelectric and earthquake data that applies the same significance tests but finds no statistical link after correcting for multiple comparisons would falsify the claim.
read the original abstract
We examine the precursory behavior of geoelectric signals before large earthquakes by means of an algorithm including an alarm-based model and binary classification. This algorithm, introduced originally by Chen and Chen [Nat. Hazards., 84, 2016], is improved by removing a time parameter for coarse-graining of earthquake occurrences, as well as by extending the single station method into a joint stations method. We also determine the optimal frequency bands of earthquake-related geoelectric signals with the highest signal-to-noise ratio. Using significance tests, we also provide evidence of an underlying seismoelectric relationship. It is appropriate for machine learning to extract this underlying relationship, which could be used to quantify probabilistic forecasts of impending earthquakes, and to get closer to operational earthquake prediction.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript presents an improved version of an alarm-based model combined with binary classification to detect precursory geoelectric signals before large earthquakes. Building on Chen and Chen (2016), the authors remove a time parameter for coarse-graining earthquake occurrences, extend the approach from single-station to joint-station analysis, identify optimal frequency bands via highest signal-to-noise ratio, and invoke significance tests to claim evidence of an underlying seismoelectric relationship. They conclude that machine learning can extract this relationship to produce probabilistic earthquake forecasts.
Significance. If the statistical evidence for a genuine seismoelectric relationship survives scrutiny, the work could supply a concrete data-driven pathway toward probabilistic forecasts in a field where operational prediction remains elusive. The joint-station extension and SNR-based band selection are practical methodological steps that could be adopted by others working with electromagnetic precursors.
major comments (1)
- [Abstract] Abstract: The central claim that 'significance tests... provide evidence of an underlying seismoelectric relationship' rests on tests performed after data-driven selection of optimal frequency bands (by highest SNR) and after extension to joint-station combinations. The manuscript does not state whether any multiplicity correction (Bonferroni, FDR, or permutation-based) was applied to the family of tests over bands and station groupings. Without such correction the reported significance levels are inflated and cannot securely rule out noise or post-selection artifacts, directly undermining the evidential foundation for the downstream machine-learning recommendation.
minor comments (1)
- [Abstract] The abstract invokes 'significance tests' without naming the test statistic, the exact null hypothesis, sample sizes, or p-value threshold; these details should be supplied in the methods section for reproducibility.
Simulated Author's Rebuttal
We thank the referee for the constructive comment on the statistical robustness of our significance tests. We address the concern directly below and will revise the manuscript to strengthen the evidential claims.
read point-by-point responses
-
Referee: [Abstract] Abstract: The central claim that 'significance tests... provide evidence of an underlying seismoelectric relationship' rests on tests performed after data-driven selection of optimal frequency bands (by highest SNR) and after extension to joint-station combinations. The manuscript does not state whether any multiplicity correction (Bonferroni, FDR, or permutation-based) was applied to the family of tests over bands and station groupings. Without such correction the reported significance levels are inflated and cannot securely rule out noise or post-selection artifacts, directly undermining the evidential foundation for the downstream machine-learning recommendation.
Authors: We agree that selecting optimal frequency bands via highest SNR and extending to joint-station combinations constitutes a family of tests, and that the absence of multiplicity correction (e.g., FDR or permutation-based) means the reported significance levels may be inflated. The submitted manuscript did not apply or discuss such a correction. In revision we will (i) explicitly state the number of tests performed, (ii) apply a permutation-based multiplicity correction to the significance tests, and (iii) update the abstract and discussion to reflect the corrected evidence for the seismoelectric relationship. This change directly addresses the referee's concern and bolsters the foundation for the machine-learning recommendation. revision: yes
Circularity Check
No significant circularity; empirical application with statistical tests on new data
full rationale
The paper extends its authors' 2016 algorithm by removing a time parameter and adding a joint-stations method, selects frequency bands by highest SNR on the data, and reports significance tests as evidence for a seismoelectric relationship before recommending ML for forecasts. This workflow applies an existing method to fresh observations and performs statistical checks rather than deriving a result that equals its inputs by construction. No equations reduce to tautologies, no fitted parameters are relabeled as independent predictions, and the cited prior work supplies only the base procedure while the new evidence claim rests on the tests themselves. Self-citation occurs but is not load-bearing for the central statistical claim.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.