Land Subsidence Susceptibility Mapping Research Articles

Land Subsidence Susceptibility Mapping in Ca Mau Province, Vietnam, Using Boosting Models

The Ca Mau Peninsula, situated in the Mekong Delta of Vietnam, features low-lying terrain. In addition to the challenges posed by climate change, land subsidence in the area is exacerbated by the overexploitation of groundwater and intensive agricultural practices. In this study, we assessed the land subsidence susceptibility in the Ca Mau Peninsula utilizing three boosting machine learning models: AdaBoost, Gradient Boosting, and Extreme Gradient Boosting (XGB). Eight key factors were identified as the most influential in land subsidence within Ca Mau: land cover (LULC), groundwater depth, digital terrain model (DTM), normalized vegetation index (NDVI), geology, soil composition, distance to roads, and distance to rivers and streams. The dataset includes 2011 points referenced from the Persistent Scattering SAR Interferometry (PSI) method, of which 1011 points are subsidence points and the remaining are non-subsidence points. The sample points were split, with 70% allocated to the training set and 30% to the testing set. Following computation and execution, the three models underwent evaluation for accuracy using statistical metrics such as the receiver operating characteristic (ROC) curve, area under the curve (AUC), specificity, sensitivity, and overall accuracy (ACC). The research findings revealed that the XGB model exhibited the highest accuracy, achieving an AUC and ACC above 0.88 for both the training and test sets. Consequently, XGB was chosen to construct a land subsidence susceptibility map for the Ca Mau Peninsula. In addition, 31 subsidence points measured by leveling surveys between 2005 and 2020, provided by the Department of Survey, Mapping and Geographic Information Vietnam, were used for validating the land subsidence susceptibility from the XGB method. The findings indicate a 70.9% accuracy rate in predicting subsidence susceptibility compared to the leveling measurement points.

Land subsidence susceptibility mapping based on InSAR and a hybrid machine learning approach

Land subsidence (LS) due to natural processes or human activity can irreparably damage the environment. This study employed the quasi-permanent scatterer method to detect areas with and without subsidence in the period from 2018 to 2020. In addition, 12 factors affecting subsidence were selected to detect LS-prone areas. Information gain ratio (IGR) and frequency ratio methods were used to determine the importance and weighting of various factors and sub-factors affecting subsidence. Novel approaches, including the standard adaptive-network-based fuzzy inference system (ANFIS) algorithm and its integration with the particle swarm optimization (PSO) algorithm, yielded LS maps. The models’ predictive performance was assessed using statistical indexes such as the root mean square error (RMSE), area under the receiver operating characteristic curve (AUROC) and confusion matrix criteria (e.g., sensitivity, specificity, precision, accuracy, and recall). Finally, Shapley additive explanations approach was used to explore the importance of features in modeling. The findings showed that the subsidence pattern was V-shaped, averaging 321 mm/year. Ground-leveling and interferometric synthetic aperture radar measurements revealed a 0.93 correlation coefficient with a σ = 1.43 mm/year deformation rate. Based on IGR analysis, aquifer media, the decline of the water table, and aquifer thickness played pivotal roles in LS occurrences. In addition, the ANFIS-PSO model classified approximately 50.26 % of the study area as high and very high susceptibility. The AUROC values of ANFIS-PSO and ANFIS models for the training dataset were 0.912 and 0.807, respectively. For the testing dataset, the ANFIS-PSO model produced a higher AUROC value of 0.863, while the ANFIS model had a value of 0.771. In addition, the RMSE values for the ANFIS-PSO model were lower. Given its remarkable accuracy, the ANFIS-PSO model was deemed suitable for evaluating subsidence susceptibility in the study area.

Evaluating Machine Learning-Based Approaches in Land Subsidence Susceptibility Mapping

Land subsidence (LS) due to natural and human-driven forces (e.g., earthquakes and overexploitation of groundwater) has detrimental and irreversible impacts on the environmental, economic, and social aspects of human life. Thus, LS hazard mapping, monitoring, and prediction are important for scientists and decision-makers. This study evaluated the performance of seven machine learning approaches (MLAs), comprising six classification approaches and one regression approach, namely (1) classification and regression trees (CARTs), (2) boosted regression tree (BRT), (3) Bayesian linear regression (BLR), (4) support vector machine (SVM), (5) random forest (RF), (6) logistic regression (LogR), and (7) multiple linear regression (MLR), in generating LS susceptibility maps and predicting LS in two case studies (Semnan Plain and Kashmar Plain in Iran) with varying intrinsic characteristics and available data points. Multiple input variables (slope, aspect, groundwater drawdown, distance from the river, distance from the fault, lithology, land use, topographic wetness index (TWI), and normalized difference vegetation index (NDVI)), were used as predictors. BRT outperformed the other classification approaches in both case studies, with accuracy rates of 75% and 74% for Semnan and Kashmar plains, respectively. The MLR approach yielded a Mean Square Error (MSE) of 0.25 for Semnan plain and 0.32 for Kashmar plain. According to the BRT approach, the variables playing the most significant role in LS in Semnan Plain were groundwater drawdown (20.31%), distance from the river (17.11%), land use (14.98%), NDVI (12.75%), and lithology (11.93%). Moreover, the three most important factors in LS in Kashmar Plain were groundwater drawdown (35.31%), distance from the river (23.1%), and land use (12.98%). The results suggest that the BRT method is not significantly affected by data set size, but increasing the number of training set data points in MLR results in a decreased error rate.

Land subsidence susceptibility mapping: a new approach to improve decision stump classification (DSC) performance and combine it with four machine learning algorithms.

Land subsidence is a worldwide threat. In arid and semiarid lands, groundwater depletion is the main factor that induce the subsidence resulting in environmental damages and socio-economic issues. To foresee and prevent the impact of land subsidence, it is necessary to develop accurate maps of the magnitude and evolution of the subsidences. Land subsidence susceptibility maps (LSSMs) provide one of the effective tools to manage vulnerable areas and to reduce or prevent land subsidence. In this study, we used a new approach to improve decision stump classification (DSC) performance and combine it with machine learning algorithms (MLAs) of naïve Bayes tree (NBTree), J48 decision tree, alternating decision tree (ADTree), logistic model tree (LMT), and support vector machine (SVM) in land subsidence susceptibility mapping (LSSSM). We employ data from 94 subsidence locations, among which 70% were used to train learning hybrid models and the other 30% were used for validation. In addition, the models' performance was assessed by ROC-AUC, accuracy, sensitivity, specificity, odd ratio, root-mean-square error (RMSE), kappa, frequency ratio, and F-score techniques. A comparison of the results obtained from the different models reveals that the new DSC-ADTree hybrid algorithm has the highest accuracy (AUC = 0.983) in preparing LSSSMs as compared to other learning models such as DSC-J48 (AUC = 0.976), DSC-NBTree (AUC = 0.959), DSC-LMT (AUC = 0.948), DSC-SVM (AUC = 0.939), and DSC (AUC = 0.911). The LSSSMs generated through the novel scientific approach presented in our study provide reliable tools for managing and reducing the risk of land subsidence.

Land subsidence susceptibility mapping in urban settlements using time-series PS-InSAR and random forest model

Land subsidence in urban settlements is globally becoming prevalent and severe due to sea level rise and accelerated construction. However, few studies have analyzed the susceptibility of land subsidence in urban settlements and the subsidence rate thresholds that have a great impact on the reliability of the land subsidence susceptibility map (LSSM). This work aims to provide a novel LSSM framework for decision makers to conduct risk control in regional urban settlements. Herein, the COSMO-SkyMed Synthetic Aperture Radar (SAR) data from July 2016 to June 2021 was acquired for remote sensing interpretation in Wuhan, China. Persistent Scatterer Interferometric Synthetic Aperture Radar (PS-InSAR) technology, combined with geostatistical analysis, was employed to map regional land subsidence rates. A total of 12 impact factors, identified through Pearson correlation coefficient (PCC) and multicollinearity tests, were selected as input features for the machine learning model. The performance of the model was assessed using Receiver Operating Characteristic (ROC) curves, and the segmentation of PS points for land subsidence rate was discussed. Furthermore, the correspondence between the measured ground-level observations and the predicted LSSM was compared. The results demonstrate that the random forest (RF) model outperforms other models in the test set, achieving an Area Under the Curve (AUC) of 0.940. The optimal threshold of −10 mm/year is proposed for segmenting PS points, which exhibits the strongest reliability in the distribution characteristics of the susceptibility index (SI). Combined with a high-performance hybrid model, this work is expected to provide a promising reference for land subsidence susceptibility prediction in similar urban settlements worldwide.

Land subsidence susceptibility mapping: comparative assessment of the efficacy of the five models.

Land subsidence (LS) as a major geological and hydrological hazard poses a major threat to safety and security. The various triggers of LS include intense extraction of aquifer bodies. In this study, we present an LS inventory map of the Daumeghan plain of Iran using 123 LS and 123 non-LS locations which were identified through field survey. Fourteen LS causative factors related to topography, geology, hydrology, and anthropogenic characteristics were selected based on multi-collinearity test. Based on the results, five susceptibility maps were generated employing models and input data. The LS susceptibility models were evaluated and validated using the receiver operating characteristic (ROC) curve and statistical indices. The results indicate that the LS susceptibility maps produced have good accuracy in predicting the spatial distribution of LS in the study area. The result showed that the optimization models BA and GWO were better than the other machine learning algorithm (MLA). In addition, The BA model has 96.6% area under of ROC (AUROC) followed by GWO (95.8%), BART (94.5%), BRT (93.1%), and SVR (92.7%). The LS susceptibility maps formulated in our study can serve as a useful tool for formulating mitigation strategies and for better land-use planning.

A fuzzy logic approach for land subsidence susceptibility mapping: the use of hydrogeological data

A fuzzy logic approach for land subsidence susceptibility mapping: the use of hydrogeological data

Land Subsidence Susceptibility Mapping Using Interferometric Synthetic Aperture Radar (InSAR) and Machine Learning Models in a Semiarid Region of Iran

Most published studies identify groundwater extraction as the leading cause of land subsidence (LS). However, the causes of LS are not only attributable to groundwater extraction. Other land-use practices can also affect the occurrence of LS. In this study, radar interferometric techniques and machine learning (ML) models were used for the prediction, susceptibility zoning, and prioritization of influential variables in the occurrence of LS in the Bakhtegan basin. The LS rate was characterized by applying an interferometric synthetic aperture radar (InSAR). The recursive feature elimination (RFE) method was used to detect and select the dominant combination of indicators to prepare an LS susceptibility map. Three ML models, including random forest (RF), k-nearest neighbors (KNN), and classification and regression trees (CART), were used to develop predictive models. All three models had acceptable performance. Among the ML models, the RF model performed the best (i.e., Nash–Sutcliffe efficiency, Kling–Gupta efficiency, correlation coefficient, and percent bias metrics of 0.76, 0.78, 0.88, and 0.70 for validating phase, respectively). The analysis conducted on all three ML model outputs showed that high and very high LS susceptibility classes were located on or near irrigated agricultural land. The results indicate that the leading cause of land LS in the study region is not due to groundwater withdrawals. Instead, the distance from dams and the proximity to anticlines, faults, and mines are the most important identifiers of LS susceptibility. Additionally, the highest probability of LS susceptibility was found at distances less than 18 km from synclines, 6 to 13 km from anticlines, 23 km from dams, and distances less than 20 to more than 144 km from mines. The validated methods presented in this study are reproducible, transferrable, and recommended for mapping LS susceptibility in semiarid and arid climate zones with similar environmental conditions.

LAND SUBSIDENCE SUSCEPTIBILITY MAPPING USING MACHINE LEARNING ALGORITHMS

Abstract. Land subsidence (LS) is one of the most challenging natural disasters that has potential consequences such as damage to infrastructures and buildings, creating sinkholes, and leading to soil destruction. To mitigate the damages caused by LS, it is necessary to determine the LS-prone areas. In this paper, LS susceptibility was assessed for Kashan Plain in Iran using Random Forest (RF) and XGBoost machine learning algorithms. For the susceptibility analysis, twelve influential factors including elevation, slope, aspect, curvature, topographic wetness index (TWI), groundwater drawdown (GWD), normalized difference vegetation index (NDVI), distance to stream (DtS), distance to road (DtR), distance to fault (DtF), lithology, and land use were taken into account. 291 LS points were used in this study which was divided into two parts of 70% and 30% for training and testing the models, respectively. The prediction power of the models and their produced LS susceptibility maps (LSSMs) were validated using the Root Mean Square Error (RMSE), R-Squared (R2), and Mean Absolute Error (MAE) values. The results showed that the XGBoost had a higher R² equal to 0.9032 compared to that of the RF which was equal to 0.8355. XGBoost model had an RMSE equal to 0.3764 cm compared to that of the RF model which was equal to 0.4906 cm. MAE for the XGBoost model was 0.1217 cm and for the RF model was 0.3050 cm. Therefore, the achieved results proved that XGBoost had better performance in this research for predicting LS values based on the measured ones.

Application of novel hybrid model for land subsidence susceptibility mapping

Land subsidence (LS) is one of the challenging geo‐hazards in several regions of the world that is mainly caused by the overexploitation of groundwater and subsequent collapse of underground cavities. The prediction of LS is difficult due to limitations in monitoring, proper surveys, and knowledge related to the functioning and behaviour of the phenomenon. The application of remote sensing (RS) and global positioning system (GPS) for the proper location identification of LS in the ground and its implementation in susceptibility mapping can help to better constrain this geohazard. However, proper LS susceptibility maps to identify LS prone areas and to mitigate this problem in a way are generally lacking. Several methodologies have been developed for preparing LS susceptibility maps through RS techniques coupled with statistical and machine learning (ML) algorithms. Furthermore, different regions have unique natural and anthropogenic causes for the occurrence of LS, such as subsidence due to groundwater over‐exploitation, road construction, consolidation of alluvial soil, mining activities, and so forth, resulting in severe economic losses. The LS susceptibility mapping (LSSM) can help to prevent natural and human catastrophes and reduce the economic damages. ML techniques are becoming increasingly proficient at modelling such occurrences and have wide application in LSSM. This study compares the performances of single and hybrid ML models to predict and map the susceptible areas prone to LS. A combination of several algorithms helps to improve the overall results. In order to achieve this, the gradient‐boosting machine‐evolutionary genetic algorithm (GBM‐EG) was adopted in this study for LS assessment and compared to three algorithms, namely random forest, artificial neural networks, and alternating decision tree, that stand as commonly used benchmarks. This case study on LSSM was carried out in the Abozidabad watershed, Isfahan Province, Iran, and a total of 155 LS locations were identified by using GPS. Measures of 12 conditioning factors were compiled for each of the sites. These data were input into the models in five different training‐to‐validation configurations of sample ratios: D1 (90%:10%), D2 (80%:20%), D3 (70%:30%), D4 (60%:40%), and D5 (50%:50%). The performance of the algorithms were evaluated with the receiver operating characteristic curve and seed cell area index (SCAI). The results show that the best area under curve value (0.989) and SCAI value were achieved by the GBM‐EG hybrid model in an 80:20 ratio. The results demonstrate that LSSMs can be effectively produced to support planning and land use for management decisions.

Land-subsidence susceptibility mapping: assessment of an adaptive neuro-fuzzy inference system–genetic algorithm hybrid model

Land-subsidence (LS) is a common geo-hazard in many regions of the world. The overexploitation of subsurface and ground water, followed by collapse of underground cavities, is one of the mechanisms that cause LS. Optimal prediction of LS is difficult due to limitations in monitoring, surveying and information about the mechanism and process of LS. Therefore, LS susceptibility mapping (LSSM) is crucial to prevent and reduce the economic damage of natural and manmade catastrophes. This study in the Isfahan Province, Iran, aimed to map LS susceptibility using a hybrid model of adaptive neuro-fuzzy inference system–genetic algorithm (ANFIS–GA) and three stand-alone machine-learning (ML) models: support vector machine, maximum entropy and ANFIS. In total, 155 LS historical points were collected, and 12 LS conditioning factors (LSCFs) were compiled for each site. These factors were elevation, slope, aspect, topographic wetness index, plan curvature, distance to stream, drainage density, groundwater withdrawal, lithology, distance to fault, normalized difference vegetation index and land use and land cover. These data were input into the models in five training-to-validation sample ratios, designated D1 (90:10), D2 (80:20), D3 (70:30), D4 (60:40) and D5 (50:50). The importance of these variables to the occurrence of LS was calculated using a random forest algorithm. Algorithm performance was evaluated through receiver operating characteristic-area under the curve analysis and the results show ANFIS–GA to be the optimal model, with average training and validation AUC values of 0.9644 and 0.9416, respectively. The results demonstrate that LSSMs can be effectively prepared to support decision makers in land use planning and allocation through the development of suitable management strategies.

Land subsidence susceptibility mapping using PWRSTFAL framework and analytic hierarchy process: fuzzy method (case study: Damaneh-Daran Plain in the west of Isfahan Province, Iran).

Land subsidence susceptibility mapping is a necessary prerequisite for land subsidence management. Proper assessment of land subsidence requires determining parameters affecting subsidence to discover the spatial relationships with subsidence. For this purpose, two frameworks were prepared, according to the land site subsidence, eight parameters affecting the subsidence in the Damaneh-Daran plain. In the first framework, eight parameters including annual groundwater level drawdown, aquifer medium, land use, discharge, aquifer saturation thickness, net recharge, fault proximity and topography called PWRSTFAL based on weighting parameters and the second model based on optimization and compatibility of eight active layers on subsidence, were prepared by AHP-fuzzy method and the final map of subsidence vulnerability obtained. Radar images and the InSAR method validated the methods. The results show that both maps show a good correlation with radar data. Still, the map prepared by the AHP-fuzzy method offers the highest correlation with radar data and subsidence in the plain. It separates subsidence vulnerability in more detail on the whole surface of the plain. According to this model, most plain areas, especially the eastern part, are subject to subsidence, and management programs should control subsidence. The results ROC diagram obtained from the performance of the PWRSTFAL framework based on the AHP-fuzzy method was more promising than the ordinary PWRSTFAL framework.

Performance Evaluation of GIS-Based Novel Ensemble Approaches for Land Subsidence Susceptibility Mapping

The optimal prediction of land subsidence (LS) is very much difficult because of limitations in proper monitoring techniques, field-base surveys and knowledge related to functioning and behavior of LS. Thus, due to the lack of LS susceptibility maps it is almost impossible to identify LS prone areas and as a result it influences severe economic and human losses. Hence, preparation of LS susceptibility mapping (LSSM) can help to prevent natural and human catastrophes and reduce the economic damages significantly. Machine learning (ML) techniques are becoming increasingly proficient in modeling purpose of such kinds of occurrences and they are increasing used for LSSM. This study compares the performances of single and hybrid ML models to preparation of LSSM for future prediction of performance analysis. In this study, the spatial prediction of LS was assessed using four ML models of maximum entropy (MaxEnt), general linear model (GLM), artificial neural network (ANN) and support vector machine (SVM). Alongside, the possible numbers of novel ensemble models were integrated through the aforementioned four ML models for optimal analysis of LSSM. An inventory LS map was prepared based on the previous occurrences of LS points and the dataset were divvied into 70:30 ratios for training and validating of the modeling process. To identify the robust and best LSSMs, receiver operating characteristic-area under curve (ROC-AUC) curve was employed. The ROC-AUC result indicated that ANN model gives the highest ROC-AUC (0.924) in training accuracy. The highest AUC (0.823) of the LSSMs was determined based on validation datasets identified by SVM followed by ANN-SVM (0.812).

Land Subsidence Susceptibility Mapping Using Persistent Scatterer SAR Interferometry Technique and Optimized Hybrid Machine Learning Algorithms

In this paper, land subsidence susceptibility was assessed for Shahryar County in Iran using the adaptive neuro-fuzzy inference system (ANFIS) machine learning algorithm. Another aim of the present paper was to assess if ensembles of ANFIS with two meta-heuristic algorithms (imperialist competitive algorithm (ICA) and gray wolf optimization (GWO)) would yield a better prediction performance. A remote sensing synthetic aperture radar (SAR) dataset from 2019 to 2020 and the persistent-scatterer SAR interferometry (PS-InSAR) technique were used to obtain a land subsidence inventory of the study area and use it for training and testing models. Resulting PS points were divided into two parts of 70% and 30% for training and testing the models, respectively. For susceptibility analysis, eleven conditioning factors were taken into account: the altitude, slope, aspect, plan curvature, profile curvature, topographic wetness index (TWI), distance to stream, distance to road, stream density, groundwater drawdown, and land use/land cover (LULC). A frequency ratio (FR) was applied to assess the correlation of factors to subsidence occurrence. The prediction power of the models and their generated land subsidence susceptibility maps (LSSMs) were validated using the root mean square error (RMSE) value and area under curve of receiver operating characteristic (AUC-ROC) analysis. The ROC results showed that ANFIS-ICA had the best accuracy (0.932) among the models (ANFIS-GWO (0.926), ANFIS (0.908)). The results of this work showed that optimizing ANFIS with meta-heuristics considerably improves LSSM accuracy although ANFIS alone had an acceptable result.

Land Subsidence Susceptibility Mapping in Jakarta Using Functional and Meta-Ensemble Machine Learning Algorithm Based on Time-Series InSAR Data

Areas at risk of land subsidence in Jakarta can be identified using a land subsidence susceptibility map. This study evaluates the quality of a susceptibility map made using functional (logistic regression and multilayer perceptron) and meta-ensemble (AdaBoost and LogitBoost) machine learning algorithms based on a land subsidence inventory map generated using the Sentinel-1 synthetic aperture radar (SAR) dataset from 2017 to 2020. The land subsidence locations were assessed using the time-series interferometry synthetic aperture radar (InSAR) method based on the Stanford Method for Persistent Scatterers (StaMPS) algorithm. The mean vertical deformation maps from ascending and descending tracks were compared and showed a good correlation between displacement patterns. Persistent scatterer points with mean vertical deformation value were randomly divided into two datasets: 50% for training the susceptibility model and 50% for validating the model in terms of accuracy and reliability. Additionally, 14 land subsidence conditioning factors correlated with subsidence occurrence were used to generate land subsidence susceptibility maps from the four algorithms. The receiver operating characteristic (ROC) curve analysis showed that the AdaBoost algorithm has higher subsidence susceptibility prediction accuracy (81.1%) than the multilayer perceptron (80%), logistic regression (79.4%), and LogitBoost (79.1%) algorithms. The land subsidence susceptibility map can be used to mitigate disasters caused by land subsidence in Jakarta, and our method can be applied to other study areas.

Integration of InSAR Time-Series Data and GIS to Assess Land Subsidence along Subway Lines in the Seoul Metropolitan Area, South Korea

The aims of this research were to map and analyze the risk of land subsidence in the Seoul Metropolitan Area, South Korea using satellite interferometric synthetic aperture radar (InSAR) time-series data, and three ensemble machine-learning models, Bagging, LogitBoost, and Multiclass Classifier. Of the types of infrastructure present in the Seoul Metropolitan Area, subway lines may be vulnerable to land subsidence. In this study, we analyzed Persistent Scatterer InSAR time-series data using the Stanford Method for Persistent Scatterers (StaMPS) algorithm to generate a deformation time-series map. Subsidence occurred at four locations, with a deformation rate that ranged from 6–12 mm/year. Subsidence inventory maps were prepared using deformation time-series data from Sentinel-1. Additionally, 10 potential subsidence-related factors were selected and subjected to Geographic Information System analysis. The relationship between each factor and subsidence occurrence was analyzed by using the frequency ratio. Land subsidence susceptibility maps were generated using Bagging, Multiclass Classifier, and LogitBoost models, and map validation was carried out using the area under the curve (AUC) method. Of the three models, Bagging produced the largest AUC (0.883), with LogitBoost and Multiclass Classifier producing AUCs of 0.871 and 0.856, respectively.

Land-subsidence susceptibility zonation using remote sensing, GIS, and probability models in a Google Earth Engine platform

Land subsidence causes many problems, including faulting of Earth's surface, damage irrigation systems, reduced soil fertility, misalignment and failure of the walls of wells, changes of river courses and roadways, and even damages in residential areas and agricultural lands. This study assesses groundwater changes using the Google Earth Engine (GEE) and land-subsidence susceptibility (LSS) on the Gharabolagh Plain in Fars Province, Iran, using two probabilistic models, evidential belief function (EBF) and Bayesian theory (BT). In extensive field surveys, the land-subsidence points identified in the study area were recorded by handheld GPS. A land-subsidence location map (LSIM) was prepared in ArcGIS 10.6.1. The sites were divided into two sets: 70% of the subsidence points locations to be used for modeling (calibration) and 30% to be used for testing (validation) of the models. Effective factors that promote land-subsidence occurrence (LSO) were selected and maps for each were prepared from several sources—topographical and geological maps, and satellite images. Changing groundwater level was determined from the GEE platform using GRACE satellite images from a 14-year period. Three methods—CSR (Center for Space Research at the University of Texas), JPL (Jet Propulsion Laboratory), and GFZ (German Research Centre for Geosciences)—were employed and show that the greatest changes in water depth in the study area are equal to − 16.55, − 17.77, and − 15.09 cm from CSR, GFZ, and JPL, respectively. The spatial relationships between land subsidence and the ten effective factors were assessed using the EBF algorithm and BT. Also, analysis of variables importance was done using the random forest model. Prioritizing these factors, it was revealed that excessive withdrawal of groundwater, elevation, and distance from rivers were the most important variables promoting subsidence. Finally, the LSS maps were prepared in GIS. The validation analysis of the two probability models by ROC curve revealed that EBF had 93.2% accuracy and BT was 99.8% accurate. The maps generated with both models can effectively indicate LSS for managers of groundwater and can be used to prevent fatalities, financial losses, and damage to property caused by this phenomenon.

Land subsidence susceptibility mapping using Analytical Hierarchy Process (AHP) and Certain Factor (CF) models at Neyshabur plain, Iran

Some areas in Iran are affected by the consequences of subsidence. This study aimed to compare the performance of the Analytical Hierarchy Process and Certain Factor models in subsidence susceptibility mapping in the Neyshabur Aquifer. At first, 60% of subsidence areas were randomly selected for building subsidence susceptibility and the remaining 40% were used to validate the models. Maps of effective parameters on subsidence event such as hydraulic conductivity, specific yield, groundwater level drop, alluvium thickness, saturation thickness, the thickness of compressible clay layer, and recharge groundwater were prepared in the GIS environment. Maps were zoned into low, moderate, high, and very high susceptibility. The highest levels of subsidence occurred in the northern and northeastern regions. Results confirmed by validation methods, R-index and area under the curve methods. The CF model generated high accuracy and prediction rated (91.7 and 90.2%) than that of the AHP model (88.4 and 85%), respectively.

A comparative study of land subsidence susceptibility mapping of Tasuj plane, Iran, using boosted regression tree, random forest and classification and regression tree methods

Land subsidence occurrence in the Tasuj plane is becoming more frequent and hazardous in the near future due to the water crisis. To mitigate damage caused by land subsidence events, it is necessary to determine the susceptible or prone areas. This study focuses on producing and comparing land subsidence susceptibility map (LSSM) using boosted regression tree (BRT), random forest (RF), and classification and regression tree (CART) approaches with twelve influencing variables, namely altitude, slope angle, aspect, groundwater level, groundwater level change, land cover, lithology, distance to fault, distance to stream, stream power index, topographic wetness index, and plan curvature. Moreover, by implementing the Relief-F feature selection method, the most important variables in LSSM procedure were identified. The performance of the adopted methods was assessed using the area under the receiver operating characteristics curve (AUROC) and statistical evaluation indexes. The results showed that all the employed methods performed well; in particular, the BRT model (AUROC = 0.819) yielded higher prediction accuracy than RF (AUROC = 0.798) and CART (AUROC = 0.764). Findings of this study can assist in characterizing and mitigating the related hazard of land subsidence events.

A novel ensemble computational intelligence approach for the spatial prediction of land subsidence susceptibility

Land subsidence (LS) is a significant problem that can cause loss of life, damage property, and disrupt local economies. The Semnan Plain is an important part of Iran, where LS is a major problem for sustainable development and management. The plain represents the changes occurring in 40% of the country. We introduce a novel-ensemble intelligence approach (called ANN-bagging) that uses bagging as a meta- or ensemble-classifier of an artificial neural network (ANN) to predict LS spatially on the Semnan Plain in Semnan Province, Iran. The ensemble model's goodness-of-fit (to training data) and prediction accuracy (of the validation data) are compared to benchmarks set by ANN-bagging. A total of 96 locations of LS and 12 LS conditioning factors (LSCFs) were collected. Each feature in the LS inventory map (LSIM) was randomly assigned to one of four groups or folds, each comprising 25% of cases. The novel ensemble model was trained using 75% (3 folds) and validated with the remaining 25% (1 fold) in a four-fold cross-validation (CV) system, which is used to control for the effects of the random selection of the training and validation datasets. LSCFs for LS prediction were selected using the information-gain ratio and multi-collinearity test methods. Factor significance was evaluated using a random forest (RF) model. Groundwater drawdown, land use and land cover, elevation, and lithology were the most important LSCFs. Using the k-fold CV approaches, twelve LS susceptibility maps (LSSMs) were prepared as each fold employed all three models (ANN-bagging, ANN, and bagging). The LS susceptibility mapping showed that between 5.7% and 12.6% of the plain had very high LS susceptibility. All three models produced LS susceptibility maps with acceptable prediction accuracies and goodness-of-fits, but the best maps were produced by the ANN-bagging ensemble method. Overall, LS risk was highest in agricultural areas with high groundwater drawdown in the flat lowlands on quaternary sediments (Qcf). Groundwater extraction rates should be monitored and potentially limited in regions of severe or high LS susceptibility. This investigation details a novel methodology that can help environmental planners and policy makers to mitigate LS to help achieve sustainability.