Spatially Validated Machine Learning for Sinkhole Susceptibility Mapping in Abandoned Coalfields

Underground cavity failures pose a major risk in former mining areas, where sinkholes can unexpectedly Damage residential areas and infrastructure. This study maps the susceptibility to mining-induced sinkholes in abandoned underground coal mining areas in Bohemia, Czech Republic, using machine learning and identifies the site conditions associated with their occurrence. The dataset comprises 184 documented sinkholes and 368 control points, with training and prediction restricted to mapped mining areas. Six machine learning models were evaluated using spatial cross-validation to limit overly optimistic performance estimates caused by spatial dependence. Random forest achieved the best discrimination (AUC=0.77) and was interpreted using SHAP values and partial dependence plots, while a model-independent contrast between sinkhole and control locations characterized the associated site conditions. A bootstrap ensemble of this model was used to generate a susceptibility map and quantify its predictive variability, and the map's correspondence with the documented inventory was assessed using the frequency ratio. The results offer practical guidance for spatial planning and hazard mitigation by indicating where sinkholes are most likely to occur. Note: This is a preprint of a manuscript submitted to Bulletin of Engineering Geology and the Environment. It has not yet been peer reviewed.

Authors

Institutions

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-17
DOI
https://doi.org/10.5281/zenodo.24301226
Primary Topic
Geotechnical and Mining Engineering
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Spatially Validated Machine Learning for Sinkhole Susceptibility Mapping in Abandoned Coalfields

Reza Taherdangkoo, Martin Herbst, Heinz Konietzky, Stefan Wutzler
3 links
Zenodo (CERN European Organization for Nuclear Research)
Geotechnical and Mining Engineering
preprint

Spatially Validated Machine Learning for Sinkhole Susceptibility Mapping in Abandoned Coalfields

Reza Taherdangkoo, Martin Herbst, Heinz Konietzky, Stefan Wutzler
preprint en

Abstract

Underground cavity failures pose a major risk in former mining areas, where sinkholes can unexpectedly Damage residential areas and infrastructure. This study maps the susceptibility to mining-induced sinkholes in abandoned underground coal mining areas in Bohemia, Czech Republic, using machine learning and identifies the site conditions associated with their occurrence. The dataset comprises 184 documented sinkholes and 368 control points, with training and prediction restricted to mapped mining areas. Six machine learning models were evaluated using spatial cross-validation to limit overly optimistic performance estimates caused by spatial dependence. Random forest achieved the best discrimination (AUC=0.77) and was interpreted using SHAP values and partial dependence plots, while a model-independent contrast between sinkhole and control locations characterized the associated site conditions. A bootstrap ensemble of this model was used to generate a susceptibility map and quantify its predictive variability, and the map's correspondence with the documented inventory was assessed using the frequency ratio. The results offer practical guidance for spatial planning and hazard mitigation by indicating where sinkholes are most likely to occur. Note: This is a preprint of a manuscript submitted to Bulletin of Engineering Geology and the Environment. It has not yet been peer reviewed.

Zenodo (CERN European Organization for Nuclear Research)
TU Bergakademie Freiberg (DE)
Geotechnical and Mining Engineering
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon