Comprehensive research on sinkhole triggers, data sources, and how to predict them before they happen — using free open-source data + targeted expensive equipment
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.
Sentinel-1 InSAR + DEM + rainfall + geology → XGBoost risk score
Covers: 1,569 km² (all Bangkok)
LiDAR + GPR + soil testing on top 20% risk zones
Covers: ~314 km² (flagged high risk)
Boreholes + ERT + real-time monitoring on confirmed cavities
Covers: ~47 km² (confirmed danger)
Cost saving: ~100× vs blanket expensive survey
| Factor | Data Source | Measurement |
|---|---|---|
| Karst topography (limestone/dolomite) | DMR geological maps | Bedrock type, dissolution rate |
| Soil type (collapsible soils) | LDD soil maps | Cohesion, porosity, grain size |
| Depth to bedrock | Borehole logs, geophysics | Depth in meters |
| Fault lines / fractures | Seismic hazard maps | Proximity to active faults |
| Historical sinkhole events | Government records, news | Event density per km² |
| Natural cavities / caves | Speleological surveys, GPR | Cavity size, depth, stability |
| Factor | Data Source | Measurement |
|---|---|---|
| Groundwater level changes | DGR monitoring wells | Seasonal drawdown rate (m/yr) |
| Groundwater extraction rate | Water permits, well logs | m³/day extraction |
| Rainfall intensity / patterns | TMD weather, rain radar | mm/hr, 7-day accumulation |
| Surface water infiltration | Drainage maps, flood maps | Runoff into sinkhole-prone zones |
| Aquifer recharge rate | Hydrogeological studies | Recharge vs extraction ratio |
| Soil moisture / saturation | SMAP satellite, Sentinel-1 | Volumetric water content |
| Factor | Data Source | Measurement |
|---|---|---|
| Construction / excavation | Building permits, satellite | Proximity + depth of excavation |
| Underground pipe leakage | Water authority leak records | Leak rate, pipe age, material |
| Mining / quarrying | Mining concession maps | Proximity to active mines |
| Heavy traffic / vibration | Traffic density data | Daily vehicle count × weight |
| Land use change | Landsat, Sentinel-2 NDVI | % change in built-up area |
| Dam / reservoir construction | Royal Irrigation Dept | Reservoir-induced seismicity |
| Factor | Data Source | Measurement |
|---|---|---|
| Extreme rainfall events | TMD, CMIP6 climate models | Return period >100mm/day |
| Drought-flood cycling | Historical weather data | Wet-dry cycle frequency |
| Sea level rise (coastal) | Tide gauge, satellite altimetry | mm/year rise |
| Saltwater intrusion | Groundwater salinity monitoring | Chloride concentration |
| Data | Provider | Access |
|---|---|---|
| Geological maps | DMR (กรมทรัพยากรธรณี) | FREE |
| Soil maps | LDD (กรมพัฒนาที่ดิน) | FREE |
| Weather / rainfall | TMD (กรมอุตุนิยมวิทยา) | FREE |
| Groundwater data | DGR (กรมทรัพยากรน้ำบาดาล) | FREE |
| Satellite imagery | Sentinel-1/2 (ESA Copernicus) | FREE |
| Flood maps | GISTDA | FREE |
| Digital Elevation | SRTM 30m / ASTER GDEM | FREE |
| LiDAR DEM (1-5m) | GISTDA / private | $$$ |
| Geophysical surveys | Private companies | $$$ |
| InSAR subsidence | Commercial (TRE ALTAMIRA) | $$$$ |
| Location | Model | Inputs | Accuracy |
|---|---|---|---|
| Florida, USA | Logistic regression + weights-of-evidence | Distance to karst, soil, groundwater, land use | 85-90% AUC |
| Dead Sea | Random Forest + InSAR | Subsidence velocity, shoreline distance, salt depth | 10m resolution |
| Italy | Random Forest susceptibility | Lithology, slope, land use, faults, drainage | 12 factors |
| China Karst | GIS-AHP + Fuzzy Logic | Karst development, soil, groundwater, human activity | 5-level zonation |
| Phase | Action | Timeline |
|---|---|---|
| 1. Data | Download Sentinel-1 (2017-2026) → MintPy processing → subsidence map | 2 weeks |
| 2. Features | DEM derivatives, geology encoding, rainfall aggregation, land use change | 1 week |
| 3. Model | XGBoost baseline → Random Forest → LightGBM → Ensemble with SHAP | 2 weeks |
| 4. Deploy | API endpoint + 3D Twin overlay + daily cron job + LINE bot alerts | 1 week |
SinkAlert Research — BDI Hackathon 2026 · GitHub