Identification of areas prone to subsidence risk and factors affecting it in the Kerdi Shirazi Plain using GLM and Cforest models

Extended Abstract
Introduction and Goal
Over the past few decades, subsidence has become a major global problem. Predicting and spatial modeling of land subsidence and identifying areas prone to subsidence are essential to reduce the negative effects of this environmental problem, given the intensification of this phenomenon in the country. Given the threats and destructive effects of land subsidence on water and soil resources, the need to manage this phenomenon and prevent its spread is a key issue in the sustainable development of the country. Subsidence studies are essential to gain insight, identify research gaps, improve methodology, and ensure that new research contributes to the existing knowledge base. Therefore, given the high importance of the Kerdi Shirazi Plain due to the presence of the Mourkerdi Forest Reserve, its biodiversity, as well as its agricultural status, and given that subsidence is expanding in this region, identifying areas prone to subsidence risk is essential to combat this phenomenon and reduce the damages caused by it. The main objective of this study is to develop a spatial model for subsidence risk using GLM machine learning model in the study area. For this purpose, in this study, for the first time, a machine learning model is used to identify areas prone to land subsidence risk in the Kerdi Shirazi Plain, and Also, the contribution and relative importance of various factors controlling subsidence will be quantitatively determined using the Cforest machine model.
Materials and Methods
In order to prepare a land subsidence risk map in the study area, first a database related to the factors controlling this phenomenon and an inventory map of subsidence in the area were prepared through field visits and data collection related to the presence and absence of subsidence in the ArcGIS environment. After identifying the most important factors controlling subsidence, finally, the relationship between the effective variables and the points of presence and absence of subsidence was examined through the GLM machine learning model and the output of the prediction model (values 0 to 1) was classified into five subsidence risk classes including very low risk (0 - 0.2), low (0.2 - 0.4), medium (0.4 - 0.6), high (0.6 - 0.8) and very high (0.8 - 1) and presented as a subsidence risk map. Finally, the relative importance of each of the effective and controlling factors of this method is determined using one of the Cforest machine models, which is the best model for determining the parameters controlling various hazards, especially subsidence, and is based on the amount of greater efficiency and less error. This model was used and determined using relative importance compared to other models presented.
Results and Discussion
According to the results of evaluating the performance of the GLM model for predicting the risk of subsidence using AUC, or the area under the ROC curve, the number 0.99 indicates that the GLM model has excellent performance in identifying subsidence points. According to the results of this model, 2180 and 441 hectares of the total area are in the very low and low subsidence sensitivity classes, while 402, 447 and 1113 hectares of the total area are in the medium, high and very high subsidence sensitivity classes, respectively. Also, 24.3% of the total study area has a very high susceptibility to subsidence risk.The central parts of the region with agricultural and horticultural land uses and forest land uses that are adjacent to agricultural and horticultural lands and share a common aquifer have a very high and high risk of land subsidence. Also, according to the results of the relative importance of variables controlling land subsidence risk, three main variables including land use, groundwater level, and groundwater drawdown are the most important factors controlling subsidence risk in the study area. The results of this study showed for the first time that subsidence risk can be a serious threat to forest lands, especially forests in arid and semi-arid regions.
Conclusion and Suggestions
According to the results, three main variables including land use type, groundwater level and groundwater decline are the most important factors controlling the risk of subsidence in the study area, which indicates the intensification of groundwater exploitation in order to develop agricultural and horticultural activities in the study area. Therefore, in order to reduce the negative effects of land subsidence, it is recommended to prevent any activity that leads to intensification of groundwater exploitation and to carry out watershed management activities, including flood spreading, upstream in the region in order to recharge the aquifer of the study area.