Cross-validation with a Decade of Data from Both Countries | 2026-04-05
Taiwan M6.0+ x 18
Japan M6.4+ x 17
| USGS + Open-Meteo Data
If you only look at Taiwan's data, you'll find that "rainfall in the 15 days before earthquakes was 54% higher than normal" — which seems remarkable. But this could simply be because Taiwan's major earthquakes tend to occur in winter and spring, which are naturally rainy seasons.
Data from a single country cannot distinguish between "real signals" and "seasonal coincidences."
If a weather indicator is a genuine earthquake precursor, it should show the same pattern across different countries with different climates.
Taiwan and Japan are ideal comparison subjects:
If an indicator shows +54% in Taiwan but -30% in Japan → spurious correlation (seasonally driven)
If an indicator is elevated in both countries → worth serious investigation
| Season | Taiwan M6.0+ | Japan M6.4+ |
|---|---|---|
| Winter (Dec-Feb) | 44% (8/18) | 47% (8/17) |
| Spring (Mar-May) | 33% (6/18) | 24% (4/17) |
| Summer (Jun-Aug) | 6% (1/18) | 18% (3/17) |
| Autumn (Sep-Nov) | 17% (3/18) | 12% (2/17) |
| Winter+Spring Total | 78% | 71% |
Below is the complete comparison of the 15 days before earthquakes vs. control groups. Green bars = Taiwan, red bars = Japan. Dark = pre-earthquake, light = control group.
| Indicator | Taiwan Pre-quake | Taiwan Control | Diff | Japan Pre-quake | Japan Control | Diff | Match? |
|---|---|---|---|---|---|---|---|
| Avg Pressure | 1016.3 | 1013.5 | +2.8 | 1011.0 | 1008.9 | +2.1 | ✓ |
| Pressure Range | 10.7 | 9.3 | +15% | 19.0 | 15.7 | +21% | ✓✓ |
| 15-Day Rainfall | 91.6mm | 59.3mm | +54% | 49.3mm | 70.7mm | -30% | ✗ |
| Rainy Days | 9.2 | 7.4 | +24% | 4.9 | 7.3 | -33% | ✗ |
| Avg Humidity | 83.9% | 80.1% | +3.8% | 74.2% | 78.7% | -4.5% | ✗ |
| Avg Temperature | 22.2°C | 26.5°C | -4.3°C | 8.9°C | 15.4°C | -6.5°C | ✓ |
| Indicator | Taiwan | Japan | Interpretation |
|---|---|---|---|
| Elevated Pressure Range | +15% | +21% | Most consistent and research-worthy finding across both countries |
| Winter-Spring Preference | 78% | 71% | Seasonal tectonic forces? Or statistical coincidence? |
| Slightly Higher Pressure | +2.8 hPa | +2.1 hPa | May reflect winter-spring high pressure systems |
| Indicator | Taiwan | Japan | Interpretation |
|---|---|---|---|
| Rainfall | +54% | -30% | Purely seasonal effect. Taiwan's winters are rainy; Japan's Pacific coast is dry in winter |
| Humidity | +3.8% | -4.5% | Same as above, seasonally driven |
| Rainy Days | +24% | -33% | Same as above |
This is the power of cross-country comparison: looking only at Taiwan's data, you might think "more rain before earthquakes" is a discovery. Adding Japan's data immediately refutes that "discovery."
Taiwan's pre-earthquake pressure range was 15% higher than controls; Japan's was 21% higher. Both countries showed consistent direction, with Japan's signal being stronger.
Japan's most extreme cases:
Taiwan's highest was 17.8 hPa for the Sep 2022 Taitung M6.9 (Japan's values are more extreme because mid-to-high latitude pressure systems are more intense).
Possible physical mechanism: Rapid atmospheric pressure changes (such as passing low-pressure systems) exert small but measurable forces on the crust. When a fault is already near its critical stress state, these minor external force changes could become the "last straw" that triggers an earthquake. This hypothesis has been investigated in geophysics but has not yet reached consensus.
Taiwan M6.0+ winter-spring: 78%, Japan M6.4+ winter-spring: 71%. Both countries consistent.
Possible explanations:
Taiwan -4.3°C, Japan -6.5°C. But this is entirely explained by the "winter-spring preference" — winter and spring are naturally cold. Temperature itself is unlikely to be an earthquake trigger.
Taiwan alone: pre-earthquake rainfall +54%, looks significant.
Adding Japan: pre-earthquake rainfall -30%. Completely opposite direction.
Conclusion: The rainfall difference is a seasonal effect, not an earthquake precursor. Taiwan's winters are rainy (northeast monsoon), while Japan's Pacific coast is dry in winter. Since major earthquakes prefer winter-spring → Taiwan appears to have more rain, Japan appears to have less.
Taiwan: +3.8%. Japan: -4.5%. Again, opposite directions.
Conclusion: The humidity difference is also a seasonal effect, same reasoning as rainfall.
These two cases perfectly demonstrate why cross-country comparison matters: data from a single country can easily produce seemingly plausible but actually nonexistent "patterns."
Largest in a decade
15-day pre-quake rainfall: 16.7mm (very little)
Pressure range: 13.6 hPa
Weather conditions: Unremarkable
→ The best counterexample for weather-earthquake independence
Japan's most devastating in recent years
15-day pre-quake rainfall: 130.4mm (incl. snowmelt)
Pressure range: 21.0 hPa
Weather conditions: Intense
→ The best candidate for the snowfall-induced earthquake hypothesis
Even within the same theoretical framework, different major earthquakes have completely different "weather backstories."
Taiwan's M7.4 occurred during the calmest weather → weather is not a necessary condition.
Japan's Noto M7.5 occurred after heavy snowfall → weather may be one of the triggering factors.
A reasonable conclusion: Earthquakes are primarily driven by tectonic forces. Weather is at most the "last straw" — when a fault is already near critical state, changes in pressure or precipitation may accelerate its occurrence, but they are not the fundamental cause.
Cross-country comparison clearly refutes rainfall and humidity as potential earthquake precursors (opposite directions in both countries).
However, pressure range was elevated in both countries (Taiwan +15%, Japan +21%), making it the only cross-country consistent signal.
This does not mean "pressure changes cause earthquakes" — a more likely interpretation is: when a fault is already in a critical state, intense pressure changes may be one of many minor triggering factors.
Data sources:
USGS Earthquake Catalog API |
Open-Meteo Historical Weather API
Further reading:
Taiwan Full Report |
Japan Full Report