Sinkhole susceptibility mapping in Greene county, Missouri through machine learning algorithms

Arip Syaripudin Nur, Yong Je Kim, Boo Hyun Nam, Kyungwon Park,
Sinkhole susceptibility mapping in Greene county, Missouri through machine learning algorithms,
Geodata and AI,
2025,
100035,
ISSN 3050-483X,
https://doi.org/10.1016/j.geoai.2025.100035.
(https://www.sciencedirect.com/science/article/pii/S3050483X25000346)
Abstract: This study utilizes Geographic Information Systems (GIS) and machine learning techniques to analyze the spatial distribution of sinkholes and predict their occurrence in Greene County, Missouri. Greene County is located within the Ozark Plateau physiographic province and is characterized by extensive karst features developed in Mississippian-age carbonate rocks. Sinkhole inventory data were collected from the Missouri Department of Natural Resources, and fifteen potential contributing factors, including geomorphological, and land use variables, were identified. Machine learning models based on logistic regression (LR), random forest (RF), artificial neural network (ANN), and support vector regression (SVR) were employed to evaluate the predictive capability of these factors. The mutual information method was utilized to identify the most influential factors contributing to sinkhole formation, providing insights into the mechanisms driving sinkhole occurrence. The frequency ratio (FR) method was used to visualize the influence of each factor on sinkhole distribution. Evaluation results indicate that the ANN model outperforms the LR (ROC-AUC = 0.783), RF (ROC-AUC = 0.781), and SVR (ROC-AUC = 0.802) models, achieving an area under the ROC curve (ROC-AUC) of 0.817. This demonstrates the effectiveness of ANN in predicting sinkhole occurrences in the study area. This study provides valuable insights into the factors influencing sinkhole distribution and establishes a robust machine learning framework for sinkhole prediction in Greene County, Missouri. The findings can contribute to improved risk assessment and mitigation strategies for karst hazards.
Keywords: Sinkhole susceptibility; GIS; Machine learning; Mutual information method