Drainage proximity and sinkhole occurrence in Sivrihisar (Central Turkey): A comparative analysis of linear, Poisson, and negative binomial regression models

Authors

• Bilge Bingül, Eskişehir Technical University, Department of Remote Sensing and Geographical Information Systems, Eskişehir, TürkiyeFollow
• Emrah Pekkan, Eskişehir Technical University, Institute of Earth and Space Sciences, Eskişehir, TürkiyeFollow
• Resul Çömert, Eskişehir Technical University, Institute of Earth and Space Sciences, Eskişehir, TürkiyeFollow

Highlights

• Sinkhole frequency increases significantly with proximity to drainage networks
• Data overdispersion invalidates standard Poisson models
• Negative Binomial Regression outperforms linear and Poisson models
• Hydrochemistry reveals dolomite dissolution, not gypsum
• First comprehensive inventory maps 104 cover-collapse sinkholes

Abstract

This study investigates the relationship between sinkhole occurrence and distance to drainage in the Sivrihisar region (Central Turkey) and evaluates the suitability of different regression approaches for modeling clustered count data in karst terrains. A comprehensive inventory of 104 sinkholes developed within the Neogene lacustrine limestones of the Akpınar Formation was compiled using official records, remote sensing analyses, and detailed field surveys. Sinkhole occurrences were analyzed relative to a drainage network derived from a high-resolution Digital Surface Model and grouped by proximity to drainage lines. Linear Regression (LM), Poisson Regression (PR), and Negative Binomial Regression (NBR) models were comparatively applied to quantify the relationship between sinkhole frequency and drainage proximity. The results reveal a statistically significant negative relationship: sinkhole occurrences are strongly concentrated near drainage networks and decrease rapidly with increasing distance. Exploratory data analysis indicates pronounced overdispersion in the sinkhole count data, with variance substantially exceeding the mean, thereby violating the PR model's assumptions. Consistent with this data structure, 10-fold cross-validation demonstrates that the NBR model provides the most reliable predictive performance by explicitly accounting for clustering and excess variance, outperforming both LM and PR models. Hydrogeochemical observations supported the geological interpretation, indicating that sinkhole development is associated with carbonate dissolution in the Akpınar Formation. Overall, the findings emphasize that appropriate statistical model selection is pivotal for reliable spatial assessment of sinkhole probability and highlight the dominant influence of drainage-controlled hydrological processes on sinkhole distribution at the regional scale.

DOI

https://doi.org/10.5038/ijs2602

Creative Commons License

This work is licensed under a Creative Commons Attribution-Noncommercial-Share Alike 4.0 License.

Recommended Citation

Bingül, B., Pekkan, E., Çömert, R., 2026. Drainage proximity and sinkhole occurrence in Sivrihisar (Central Turkey): A comparative analysis of linear, Poisson, and negative binomial regression models. International Journal of Speleology, 55(2), ijs2602. https://doi.org/10.5038/ijs260