Imminent earthquake forecasting on the basis of Japan INTERMAGNET stations, NEIC, NOAA and Tide code data analysis

This research presents one possible way for imminent prediction of earthquake magnitude, depth and epicenter coordinates by solving the inverse problem using a data acquisition network system for monitoring, archiving and complex analysis of geophysical variables precursors. Among many possible precursors the most reliable are the geoelectromagnetic field, the boreholes water level, the radon surface concentration, the local heat flow, the ionosphere variables, the low frequency atmosphere and Earth core waves. In this study only geomagnetic data are used. Within the framework of geomagnetic quake approach it is possible to perform an imminent regional seismic activity forecasting on the basis of simple analysis of geomagnetic data which use a new variable Schtm with dimension surface density of energy. Such analysis of Memambetsu, Kakioka, Kanoya (Japan, INTERMAGNET) stations and NEIC earthquakes data, the hypothesis that the predicted earthquake is this with bigest value of the variable Schtm permit to formulate an inverse problem (overdetermined algebraic system) for precursors signals like a functions of earthquake magnitude, depth and distance from a monitoring point. Thus, in the case of data acquisition network system existence, which includes monitoring of more than one reliable precursor variables in at least four points distributed within the area with a radius of up to 700 km, there will be enough algebraic equations for calculation of impending earthquake magnitude, depth and distance, solving the overdetermined algebraic system.

💡 Research Summary

The paper proposes a method for “imminent” earthquake forecasting that relies exclusively on geomagnetic observations from three Japanese INTERMAGNET stations—Memambetsu (MMB), Kakioka (KAK) and Kanoya (KNY). The authors introduce a new scalar quantity, denoted SchtM, which they describe as a surface energy density associated with a forthcoming earthquake. SchtM is calculated from the earthquake’s magnitude, focal depth and the distance between the epicenter and a monitoring station, although the exact functional form is obscured by garbled symbols and is not reproducibly defined.

The core of the methodology is the identification of a “geomagnetic quake” (or geomagnetic precursor) as a positive jump in a daily geomagnetic signal. This signal is constructed by taking minute‑averaged components of the geomagnetic vector, computing hourly standard deviations for each component, normalising these by the hourly vector magnitude, and then forming a geometric sum across the 24 hours. The daily signal (GmSig_day) is compared to the previous day’s value, corrected for the NOAA A‑index (a proxy for global geomagnetic activity). A positive difference is termed a geomagnetic quake.

In parallel, the authors compute the times of daily tidal extremes (maximum, minimum and inflection points) using a publicly available Earth‑tide software. For each earthquake they calculate DayDiff, the absolute time difference between the earthquake occurrence and the nearest tidal extreme. They claim that the distribution of DayDiff for all earthquakes (global catalogue 1981‑2013, M ≥ 3) follows a Gaussian with width W_all = 4.46 ± 0.22 days. When they restrict the sample to earthquakes that have the largest SchtM values (the “predicted” earthquakes), the Gaussian width narrows (e.g., W_pr ≈ 3.7 days for the three Japanese stations). This narrowing is presented as evidence that large SchtM events are more tightly coupled to tidal extremes and therefore more predictable.

The authors then assemble a set of linear‑like equations linking the observed precursor signal (the daily geomagnetic jump) to the unknown earthquake parameters (magnitude, depth, and distance to each station). For the three stations they identify 16 earthquakes that are simultaneously “predicted” by at least two stations, yielding 32 equations (each station provides a signal, each earthquake provides magnitude and depth, and each station‑earthquake pair contributes a distance term). They treat this as an over‑determined algebraic system and solve it using the REGN program, producing a set of digital parameters A that define the explicit form of the precursor‑signal function PrecSigTh(Mag,Depth,Distance). The Fortran implementation of this function is provided in an appendix.

A key claim is that with a network of at least four monitoring points and at least four independent precursor variables (e.g., geomagnetic, radon, water level, heat flow) the system would contain enough equations (2 + 2·G ≤ P·G) to solve for the four unknown earthquake attributes (time window, magnitude, depth, epicenter coordinates). Since the present study uses only one precursor (geomagnetic) and three stations, the authors acknowledge that the system is under‑determined for practical forecasting, but argue that the concept demonstrates feasibility.

Critical assessment:

1. Definition and Physical Basis of SchtM – The paper never provides a clear derivation of SchtM. The symbols in the formula are corrupted, the units are loosely described as J km⁻², and there is no discussion of how the magnitude, depth and distance combine to produce a physically meaningful surface energy density. Without a transparent definition, the reproducibility of the method is impossible.

2. Geomagnetic Signal Construction – Using hourly standard deviations of geomagnetic components is a crude proxy for magnetic variability. The method does not account for diurnal variations, magnetospheric storms, or local noise sources. The A‑index correction only removes a global solar‑wind contribution; it does not eliminate regional disturbances that could masquerade as a precursor.

3. Statistical Validation – The Gaussian fits to DayDiff are presented without goodness‑of‑fit statistics (R², p‑values) or confidence intervals beyond the quoted width. The selection of “predicted” earthquakes based on the highest SchtM values introduces a strong selection bias; the reported high percentages (e.g., 97 % for MMB) simply reflect that the same events are being re‑identified, not an independent forecast success rate.

4. Over‑determined System Feasibility – The algebraic system described (2 + 2·G ≤ P·G) assumes that each precursor variable contributes an independent equation. In reality, geomagnetic data from nearby stations are highly correlated, and the linear‑like relationship between the precursor signal and the three earthquake parameters is not justified. The authors do not explore multicollinearity, parameter identifiability, or error propagation.

5. Lack of Independent Test Set – All analyses are retrospective, using the same catalogue that supplied the SchtM values. No out‑of‑sample or real‑time prediction experiment is reported, making it impossible to assess true predictive skill.

6. Implementation Details – The Fortran code and the REGN routine are mentioned but not described; the reader cannot evaluate convergence criteria, regularisation, or sensitivity to initial guesses.

In summary, the paper introduces an intriguing concept—linking geomagnetic variability, tidal forcing, and a novel surface‑energy metric—to anticipate seismic activity. However, the methodological foundations are insufficiently rigorous, the statistical evidence is weak, and the proposed inverse‑problem framework remains largely theoretical. Future work should (i) provide a mathematically sound definition of SchtM with clear physical justification, (ii) incorporate multiple, independently validated precursors, (iii) perform robust statistical testing with blind validation datasets, and (iv) develop a realistic inversion scheme that accounts for measurement uncertainties and parameter correlations. Only then can the promise of a multi‑precursor, over‑determined forecasting network be credibly assessed.