This study aims to analyze urban ground subsidence in Seoul from a multi-scale spatial perspective and to develop a localized probability-based risk prediction model. While ground subsidence incidents have repeatedly occurred in Seoul, existing manage...
This study aims to analyze urban ground subsidence in Seoul from a multi-scale spatial perspective and to develop a localized probability-based risk prediction model. While ground subsidence incidents have repeatedly occurred in Seoul, existing management approaches have largely relied on aggregated administrative-unit analyses or post-event responses, limiting their capacity to capture localized risk patterns. To address this limitation, this study systematically examines spatial dependence and scale effects of ground subsidence and identifies key influencing factors according to different spatial units.
The analysis covers the entire Seoul metropolitan area, using geocoded point-based subsidence occurrence data. The research proceeds in three stages. First, polygon-based analyses at the administrative-dong level are conducted to identify global spatial patterns and to verify the limitations of aggregated spatial units. Second, spatial regression models are employed to investigate spatial autocorrelation and macro-level influences on subsidence occurrence. Third, a point-based binary logistic regression model is constructed to predict localized subsidence probability, and a ground subsidence risk probability map is generated for Seoul.
The polygon-level analysis reveals significant spatial clustering of ground subsidence occurrences, confirmed through Moran’s I and LISA statistics. However, the results also demonstrate the limitations of aggregated analyses due to the modifiable areal unit problem (MAUP), which obscures intra-area risk heterogeneity. Although spatial regression models confirm strong spatial dependence, macro-scale factors alone are insufficient to explain individual subsidence events.
To overcome these limitations, the study focuses on point-based modeling using localized explanatory variables. Independent variables are constructed by integrating natural environmental factors, urban physical characteristics, and underground infrastructure attributes. In particular, a composite sewer risk index is developed by combining sewer age, diameter, and material characteristics. Variables representing subway line classification, distance to subway lines, and building density are also incorporated.
Through stepwise variable selection, four key variables—composite sewer risk, subway line classification, building density, and distance to subway lines—are identified as significant predictors. The final binary logistic regression model is statistically robust, with an omnibus test significant at the 1% level and a Nagelkerke of 0.324, accounting for approximately 32.4% of the variance in ground subsidence occurrences. The Hosmer–Lemeshow test confirms good model fit, and the overall classification accuracy reaches 85.7%, which is acceptable for rare-event prediction.
Based on the estimated probabilities, a ground subsidence risk probability map is produced for Seoul. High-risk areas are found to be locally concentrated in densely developed zones with aging sewer networks and close proximity to subway lines, rather than being confined within administrative boundaries. This finding underscores the necessity of moving beyond administrative-unit-based management toward risk-oriented spatial governance.
In conclusion, this study empirically demonstrates the limitations of polygon-based analyses and highlights the critical importance of point-based localized modeling for urban ground subsidence risk prediction. The proposed probability-based model and risk map provide a practical analytical foundation for proactive subsidence prevention and targeted urban safety management in Seoul.