TAMC-FRANJAMAR: Reproducible retrospective analysis of network-coherent seismic behavior

TAMC-FRANJAMAR v3.2 — Reproducible Multistation Seismic Analysis Framework

TAMC-FRANJAMAR v3.2 provides a fully reproducible framework for the retrospective analysis of network-coherent behavior in multistation seismic systems.

The central idea of the framework is that the relevant signal is not peak amplitude at individual stations, but the emergence of temporally sustained, network-coherent structure across a distributed seismic network.

This release includes the complete implementation of the TAMC-FRANJAMAR pipeline together with three accompanying technical documents:

1. Main framework paper:
A methodological description of TAMC-FRANJAMAR as a retrospective, reproducible, non-predictive statistical framework for analyzing collective temporal organization in multistation networks. The framework transforms station-level time series into standardized metrics and aggregates them into network-level measures of synchronization, temporal clustering, and collective coherence. Event-centered analyses are compared with matched control periods and evaluated using empirical null models and placebo tests.

2. Worked example — Tohoku 2011:
A detailed technical example applying the framework to the 2011 Mw 9.1 Great Tohoku Earthquake, including matched control-day comparisons, multistation synchronization metrics, aggregate activity timelines, station-level z-scores, and additional comparative cases such as Maule 2010, Wharton Basin 2012, Tehuantepec 2017, and Lorca 2011. This document illustrates how outputs should be interpreted jointly at the system level, rather than as isolated plots or single metrics.

3. Rapid-response example — Miyako 2026:
A rapid-response application of TAMC-FRANJAMAR v3 to the 2026-04-20 M7.4 Miyako, Japan earthquake. The analysis was performed within hours of the event using the same pipeline and fixed parameters, without event-specific tuning. The framework identifies a sharp emergence of network-coherent structure at the time of the earthquake, followed by structured temporal decay. The same event is also recovered in continuous monitoring mode without prior temporal alignment.

Version update:
This version updates the robust precursor analysis script to correct the temporal reference used for matched control windows.

In previous versions, some control-window analyses could inherit the mainshock time of the parent event instead of using the timestamp encoded in the control folder name. This could produce incorrect relative times, including anomalous values such as t ≈ -8000 h.

The updated script now:
- assigns t = 0 from control_XX_YYYYMMDD_HHMMSS folders when analysing controls;
- falls back to timestamp inference from event or control names when events.csv is unavailable;
- writes robust precursor outputs as CSV and JSON for reproducible downstream analysis.

Methodological principles:
- Fixed parameters, with no event-specific tuning;
- Matched control windows;
- Empirical null models;
- Placebo tests and robustness checks;
- System-level interpretation of multistation coherence;
- Retrospective and reproducible analysis.

These constraints are intended to ensure consistent statistical evaluation, reduce overfitting, and support falsifiable interpretation. Negative, weak, ambiguous, or null-compatible outcomes are treated as informative results rather than failures of the method.

Rapid-response and monitoring context:
An experimental, non-predictive monitoring interface implementing the same pipeline is available at:

https://franjamar-monitor.streamlit.app/

The monitoring configuration uses the same multistation coherence logic in a continuous setting, enabling exploratory near-real-time visualization of network-level seismic behavior. Monitoring outputs are diagnostic and exploratory; they are not operational alerts or earthquake forecasts.

Quick start:
- conda create -n tamc python=3.10 -y
- conda activate tamc
- pip install -r requirements.txt
- python full_pipeline_franjamarv3.py

Reproducibility:
All analyses are designed to be reproducible under fixed parameters, consistent data ingestion, standardized station-level processing, matched control windows, and documented null-model evaluation.

Disclaimer:
TAMC-FRANJAMAR is diagnostic, retrospective, and non-predictive. It does not estimate future earthquake probability, timing, location, or magnitude. It is intended for reproducible exploratory and retrospective analysis of network-coherent seismic behavior, not for earthquake prediction or operational forecasting.

Related works:

Related project: GNGS

A complementary project from the same research line is GNGS — Geomagnetic Network Geometry and Synchrony, a local, reproducible and traceable framework for studying the spatial and temporal organization of geomagnetic disturbances across distributed ground-magnetometer networks.

GNGS analyses network synchrony, hierarchical structure, Peak–Valley asymmetry and multistation geomagnetic behaviour, providing both HISTORICAL and RECENT operational modes.

GNGS v1.0:
https://doi.org/10.5281/zenodo.22021222

GNGS is descriptive and diagnostic and does not provide earthquake or seismic-event predictions.