1- , ra.norouzi@yahoo.com
Abstract: (141 Views)
Objective: Over the past two decades, land subsidence has emerged as a significant geomorphological hazard and one of the most critical environmental crises in Iran, causing irreversible damage to many plains each year. Among its primary current causes is the excessive and unregulated extraction of groundwater. The Eshtehard Plain, recognized as one of the industrial and agricultural hubs of Alborz province, is no exception. Due to severe groundwater depletion, it has been officially declared a critical zone by the Ministry of Energy. The objective of this study is to model the risk of land subsidence in this plain using the Random Forest algorithm and to analyze the contributing factors influencing its occurrence
Methods: In this study, twelve independent spatial layers were utilized, including: digital elevation model (DEM), distance to rivers, distance to qanats, distance to wells, distance to faults, groundwater depth, drainage density, soil type, lithology, land use, topographic wetness index (TWI), and solar radiation. The dependent layer consisted of subsidence zones. The Random Forest model was implemented in the R software environment. Two key importance measures—Mean Decrease Accuracy and Mean Decrease Gini—were employed to rank, assess the significance of, and assign weights to the contributing factors of land subsidence. Finally, model performance was evaluated using three complementary metrics: Accuracy, Kappa, and AUCResults: The results demonstrated that the Random Forest model achieved high accuracy in classifying land subsidence risk. Model evaluation showed strong performance with an overall accuracy of 0.963, a Kappa coefficient of 0.611, and an AUC value of 0.955, indicating that the model is highly effective for spatial risk zoning of land subsidence. The most influential variables in subsidence occurrence were identified as groundwater depth, distance to wells, geology, and land use. Furthermore, more than 65% of the study area was categorized as high-risk and very high-risk, reflecting the critical condition of the Eshtehard Plain. Notably, the share of urban land use has shown a steady increase from 2011 to 2023, with a significant spike in 2023, where increased population concentration has placed additional pressure on groundwater resources, leading to an intensification of subsidence in affected areas
Conclusions: The Random Forest algorithm successfully modeled the spatial distribution of land subsidence risk with high accuracy. This method can serve as an effective tool for informed decision-making in groundwater resource management, sustainable development planning, and hazard mitigation in similar regions.
Methods: In this study, twelve independent spatial layers were utilized, including: digital elevation model (DEM), distance to rivers, distance to qanats, distance to wells, distance to faults, groundwater depth, drainage density, soil type, lithology, land use, topographic wetness index (TWI), and solar radiation. The dependent layer consisted of subsidence zones. The Random Forest model was implemented in the R software environment. Two key importance measures—Mean Decrease Accuracy and Mean Decrease Gini—were employed to rank, assess the significance of, and assign weights to the contributing factors of land subsidence. Finally, model performance was evaluated using three complementary metrics: Accuracy, Kappa, and AUCResults: The results demonstrated that the Random Forest model achieved high accuracy in classifying land subsidence risk. Model evaluation showed strong performance with an overall accuracy of 0.963, a Kappa coefficient of 0.611, and an AUC value of 0.955, indicating that the model is highly effective for spatial risk zoning of land subsidence. The most influential variables in subsidence occurrence were identified as groundwater depth, distance to wells, geology, and land use. Furthermore, more than 65% of the study area was categorized as high-risk and very high-risk, reflecting the critical condition of the Eshtehard Plain. Notably, the share of urban land use has shown a steady increase from 2011 to 2023, with a significant spike in 2023, where increased population concentration has placed additional pressure on groundwater resources, leading to an intensification of subsidence in affected areas
Conclusions: The Random Forest algorithm successfully modeled the spatial distribution of land subsidence risk with high accuracy. This method can serve as an effective tool for informed decision-making in groundwater resource management, sustainable development planning, and hazard mitigation in similar regions.
Send email to the article author
| Rights and permissions | |
 | This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. |
