🧪 Tectonic Risk Validation — Hypothesis Rejected

Empirical test using 10 real Bangkok sinkhole sites vs 1,000 random control points. All data real. No fabrication.

1. The Test

Element	Description
Test group	10 real Bangkok sinkhole incidents (Din Daeng, Lat Phrao, Bang Kapi, Ramkhamhaeng, Sukhumvit, Sathon, Ratchadaphisek, Bang Sue, Chatuchak, Khlong Toei)
Control group	1,000 random points uniformly distributed within Bangkok bounding box (100.40–100.70°E, 13.65–13.90°N)
Metric	Minimum distance (km) from each point to nearest GEM-mapped fault trace within Thailand region
Test	Mann-Whitney U (non-parametric, doesn't assume normal distribution)
Fault source	GEM Global Active Faults Database — 4,232 coordinate points in Thailand region
Null hypothesis	Sinkhole sites are no closer to faults than random Bangkok locations

2. Results

3. Per-Site Distances

Site	District	Lat, Lon	Fault Distance
1	Bang Sue	13.810, 100.527	114.9 km
2	Sathon	13.720, 100.529	117.6 km
3	Din Daeng	13.769, 100.552	118.5 km
4	Chatuchak	13.803, 100.560	118.5 km
5	Ratchadaphisek	13.786, 100.570	120.0 km
6	Sukhumvit	13.738, 100.561	120.3 km
7	Khlong Toei	13.713, 100.558	120.8 km
8	Lat Phrao	13.803, 100.614	124.2 km
9	Ramkhamhaeng	13.753, 100.607	124.7 km
10	Bang Kapi	13.762, 100.642	128.2 km

Observation: Range is only 114.9–128.2 km — just 13.3 km spread across all 10 sites. Every point in Bangkok sits at essentially the same distance from the basin-bounding faults. This makes fault proximity useless as a point-level discriminator.

4. Why It Failed

Reason	Explanation
Faults are at basin edge, not in the city	The 18 CPFZ fault traces mapped within 200 km of Bangkok are basin-bounding faults — they define the edge of the Chao Phraya Basin. All of central Bangkok is 110–130 km from them. There is no meaningful spatial variation within the city.
GEM database only has major faults	The GEM GAF-DB maps faults at regional/global scale. It doesn't include smaller intra-basin lineaments — the subsurface fractures that DMR's airborne geophysics discovered WITHIN the Central Plain. Those micro-faults might be relevant but aren't in open data.
City-scale causes dominate	Bangkok sinkholes are triggered by local factors: pipe leakage, construction, groundwater pumping, soft clay consolidation. These vary at the street/block level, not at the 100+ km basin scale.

5. What IS Actually Useful from the Tectonic Research

The fault-controlled basin does matter — but as a systemic factor, not a point-proximity factor:

What works	How to use it
Basin = sinkhole-prone everywhere	All of Bangkok sits in a fault-block basin with soft marine clay over deep aquifers. This is a city-wide constant — it explains WHY Bangkok has sinkholes at all, but doesn't tell you WHERE.
Differential subsidence along fault blocks	If we can intersect InSAR subsidence with the 18 CPFZ fault traces, blocks settling at different rates might reveal active fracture zones. This is a different analysis — not "distance to fault" but "change in subsidence rate across fault." Requires running the InSAR×faults computation.
Earthquake-triggered sinkholes	Only 1 M3+ quake within 200 km of BKK in 26 years. This factor is irrelevant for Bangkok — the city is seismically quiet.
Regional applicability	If SinkAlert expands to Kanchanaburi, Tak, or Chiang Rai (areas crossed by active faults like Three Pagodas and Mae Chan), fault proximity WOULD matter because those faults are WITHIN populated areas. Bangkok is the exception — the faults are 100+ km away.

6. What We're Keeping vs Discarding

Factor	Decision	Reason
f₁ — Point distance to fault	DISCARDED	p = 0.45 — no predictive power in Bangkok
f₂ — Local earthquake density	DISCARDED	Only 1 quake in 26 years near BKK — zero signal
f₃ — Fault slip rate	DISCARDED	Data missing for Thailand in GEM; no variation within BKK anyway
f₄ — Differential subsidence across faults	PENDING	Theoretically sound but needs InSAR×fault intersection computation — not yet tested
f₅ — Fault slip type	DISCARDED	All Bangkok-proximate faults are dextral strike-slip — no variation, no signal
f₆ — GW extraction rate	PENDING	Already partially captured by InSAR subsidence. DGR well data would refine.
f₇ — Fracture zone mask	DISCARDED	500 m buffer around faults 100+ km away has zero effect on city-center points

7. What We Actually Learned

1. Good science means publishing negative results. The research was thorough, the data was real, the test was rigorous — and the hypothesis was rejected. That's valuable. It prevents wasting time building a feature that doesn't work.
2. The basin IS fault-controlled — but at the basin scale. The Chao Phraya Fault Zone created the geological conditions that make Bangkok sinkhole-prone. That's a basin-level explanation, not a point-level predictor.
3. Bangkok's insulation from active faults is a feature, not a bug. Unlike Tokyo, Istanbul, or Los Angeles, Bangkok's sinkholes are NOT fault-triggered. They're hydrogeological — pipe leakage + soft clay + groundwater pumping. That's where the model should focus.
4. For regional expansion, this analysis becomes valuable. If SinkAlert covers Chiang Rai or Kanchanaburi — where active faults run through populated areas — re-run this test with those locations.

8. Data Provenance

Source	What	Verified
GEM GAF-DB	gem_active_faults.geojson (11.7 MB, 16,195 features worldwide)	✅ Downloaded 2026-07-19
GEM GAF-DB (Thailand extract)	194 fault traces, 4,232 coordinate points	✅ Extracted 2026-07-19
SinkAlert real incidents	sinkalert_bangkok_real.csv — 10 verified sinkhole sites with district names	✅ Existing SinkAlert data
Random control points	1,000 uniform-random points in Bangkok bbox (seed=42)	✅ Generated for this test
Statistical method	Mann-Whitney U (scipy-compatible manual implementation)	✅ Standard non-parametric test

← Tectonic Research  |  ← Risk Score v2 (proposed — see what was discarded)  |  SinkAlert Dashboard