← Back to SinkAlert Dashboard

🕳️ Sinkhole Causation & Probability Forecasting

Comprehensive research on sinkhole triggers, data sources, and how to predict them before they happen — using free open-source data + targeted expensive equipment

💡 Core Strategy: Medical Screening Model

Don't survey every square kilometer with expensive equipment. Use free satellite data + ML to screen the entire city, then deploy LiDAR + boreholes only on high-risk zones.

🔍 Tier 1 — Broad Screen

$0

Sentinel-1 InSAR + DEM + rainfall + geology → XGBoost risk score

Covers: 1,569 km² (all Bangkok)

🎯 Tier 2 — Targeted

~$5K/km²

LiDAR + GPR + soil testing on top 20% risk zones

Covers: ~314 km² (flagged high risk)

🔬 Tier 3 — Deep

~$50K+

Boreholes + ERT + real-time monitoring on confirmed cavities

Covers: ~47 km² (confirmed danger)

Cost saving: ~100× vs blanket expensive survey

1. Geological Factors (Natural)

FactorData SourceMeasurement
Karst topography (limestone/dolomite)DMR geological mapsBedrock type, dissolution rate
Soil type (collapsible soils)LDD soil mapsCohesion, porosity, grain size
Depth to bedrockBorehole logs, geophysicsDepth in meters
Fault lines / fracturesSeismic hazard mapsProximity to active faults
Historical sinkhole eventsGovernment records, newsEvent density per km²
Natural cavities / cavesSpeleological surveys, GPRCavity size, depth, stability

2. Hydrological Factors

FactorData SourceMeasurement
Groundwater level changesDGR monitoring wellsSeasonal drawdown rate (m/yr)
Groundwater extraction rateWater permits, well logsm³/day extraction
Rainfall intensity / patternsTMD weather, rain radarmm/hr, 7-day accumulation
Surface water infiltrationDrainage maps, flood mapsRunoff into sinkhole-prone zones
Aquifer recharge rateHydrogeological studiesRecharge vs extraction ratio
Soil moisture / saturationSMAP satellite, Sentinel-1Volumetric water content

3. Human-Induced Factors

FactorData SourceMeasurement
Construction / excavationBuilding permits, satelliteProximity + depth of excavation
Underground pipe leakageWater authority leak recordsLeak rate, pipe age, material
Mining / quarryingMining concession mapsProximity to active mines
Heavy traffic / vibrationTraffic density dataDaily vehicle count × weight
Land use changeLandsat, Sentinel-2 NDVI% change in built-up area
Dam / reservoir constructionRoyal Irrigation DeptReservoir-induced seismicity

4. Climate Change Amplifiers

FactorData SourceMeasurement
Extreme rainfall eventsTMD, CMIP6 climate modelsReturn period >100mm/day
Drought-flood cyclingHistorical weather dataWet-dry cycle frequency
Sea level rise (coastal)Tide gauge, satellite altimetrymm/year rise
Saltwater intrusionGroundwater salinity monitoringChloride concentration

5. Thailand-Specific Data Sources

DataProviderAccess
Geological mapsDMR (กรมทรัพยากรธรณี)FREE
Soil mapsLDD (กรมพัฒนาที่ดิน)FREE
Weather / rainfallTMD (กรมอุตุนิยมวิทยา)FREE
Groundwater dataDGR (กรมทรัพยากรน้ำบาดาล)FREE
Satellite imagerySentinel-1/2 (ESA Copernicus)FREE
Flood mapsGISTDAFREE
Digital ElevationSRTM 30m / ASTER GDEMFREE
LiDAR DEM (1-5m)GISTDA / private$$$
Geophysical surveysPrivate companies$$$
InSAR subsidenceCommercial (TRE ALTAMIRA)$$$$

6. Technology Stack for Forecasting

Remote Sensing Pipeline

ML/AI Stack

Visualization

7. Models to Learn From

LocationModelInputsAccuracy
Florida, USALogistic regression + weights-of-evidenceDistance to karst, soil, groundwater, land use85-90% AUC
Dead SeaRandom Forest + InSARSubsidence velocity, shoreline distance, salt depth10m resolution
ItalyRandom Forest susceptibilityLithology, slope, land use, faults, drainage12 factors
China KarstGIS-AHP + Fuzzy LogicKarst development, soil, groundwater, human activity5-level zonation

8. Bangkok Implementation Plan

PhaseActionTimeline
1. DataDownload Sentinel-1 (2017-2026) → MintPy processing → subsidence map2 weeks
2. FeaturesDEM derivatives, geology encoding, rainfall aggregation, land use change1 week
3. ModelXGBoost baseline → Random Forest → LightGBM → Ensemble with SHAP2 weeks
4. DeployAPI endpoint + 3D Twin overlay + daily cron job + LINE bot alerts1 week

SinkAlert Research — BDI Hackathon 2026 · GitHub