چكيده لاتين
Land subsidence, as an environmental hazard, has caused significant damage to infrastructure, facilities, and historical monuments. As the overexploitation of groundwater leads us closer to the depletion of aquifers, the extent of subsidence increases. The Isfahan-Borkhar Plain, due to its arid climate and unique topographical features with diverse elevations, has not been immune to this hazard. Various factors, such as groundwater levels, elevation, and soil composition, have been reported in previous studies as influencing land subsidence. However, very few studies have focused on modeling land subsidence based on these factors. In recent years, the use of remote sensing techniques, particularly radar interferometry (InSAR), has become common among earth science researchers for estimating land subsidence rates. This thesis aims to estimate the land subsidence rate using the InSAR technique while analyzing its influencing factors. In this study, the land subsidence rate in the Isfahan-Borkhar Plain from 2019 to 2023 was estimated using the Small Baseline Subset (SBAS) interferometry method applied to a time series of Sentinel-1 radar images. The results indicated a maximum displacement rate of 116.8 [mm / year], with a cumulative displacement of 506.29 mm. The findings also revealed that the highest rate of subsidence during this four-year period occurred in the southern part of the Isfahan-Borkhar Plain. Subsequently, the correlation between various influencing factors and land subsidence was examined. These factors included topographical features (elevation, slope, aspect, and topographic moisture), geological parameters (distance to faults and lithology), land cover conditions (vegetation and land use/land cover), and groundwater levels. Following this, land subsidence was modeled using two approaches. A fuzzy model was developed to predict land subsidence based on the most influential factors. This model demonstrated high prediction accuracy for the test data. Machine Learning Algorithms: Using decision tree (DT), random forest (RF), and extreme gradient boosting (XGBoost) algorithms, a land subsidence susceptibility map was created for the study area, categorized into five classes: very high, high, moderate, low, and very low subsidence probability. The overall accuracy of the maps produced by these methods was 75.42% for XGBoost, 90.58% for DT, and 95.63% for RF. Additionally, based on the ranking of the RF algorithm, elevation, groundwater level, lithology and distance to Qom-Zefreh fault were the most important factors, while vegetation cover and land cover were the least significant factors influencing land subsidence in the Isfahan-Borkhar Plain. Based on the best-performing map generated by the RF algorithm, the highest probability of land subsidence was observed in the central and eastern parts of the region, while the lowest probability was identified in the northwestern and western areas, as well as parts of the northeastern Isfahan-Borkhar Plain.
