Stacking- and voting-based ensemble deep learning models (SEDL and VEDL) and active learning (AL) for mapping land subsidence

Land subsidence prediction in coal mining using machine learning models and optimization techniques

Land surface subsidence is an environmental hazard resulting from the extraction of underground resources. In underground mining, when mineral materials are extracted deep within the ground, the emptying or caving of the mined spaces leads to vertical displacement of the ground, known as subsidence. This subsidence can extend to the surface as trough subsidence, as the movement and deformation of the hanging-wall rocks of the mining stope propagate upwards. Accurately predicting subsidence is crucial for estimating damage and protecting surface buildings and structures in mining areas. Therefore, developing a model that considers all relevant parameters for subsidence estimation is essential. In this article, we discuss the prediction of land subsidence caused by the caving of a stop's roof, focusing on coal mining using the longwall method. The main aim of this research is to improve the accuracy of prediction models of land subsidence due to mining. For this purpose, we consider a total of 11 parameters related to coal mining, including mining thickness and depth (related to the deposit), as well as density, cohesion, internal friction angle, elasticity modulus, bulk modulus, shear modulus, Poisson's ratio, uniaxial compressive strength, and tensile strength (related to the overburden). We utilize information collected from 14 coal mines regarding mining and subsidence to achieve this. We then explore the prediction of subsidence caused by mining using the gene expression programming (GEP) algorithm, optimized through a combination of the artificial bee colony (ABC) and ant lion optimizer (ALO) algorithms. Modeling results demonstrate that combining the GEP algorithm with optimization based on the ABC algorithm yields the best subsidence prediction, achieving a correlation coefficient of 0.96. Furthermore, sensitivity analysis reveals that mining depth and density have the greatest and least effects, respectively, on land surface subsidence resulting from coal mining using the longwall method.

An interpretable deep learning model to map land subsidence hazard

The main goal of this research is the interpretability of deep learning (DL) model output (e.g., CNN and LSTM) used to map land susceptibility to subsidence hazard by means of different techniques. For this purpose, an inventory map of land subsidence (LS) is prepared based on fieldwork and a record of LS presence points, and 16 features controlling LS were mapped. Thereafter, 11 effective features controlling LS were identified by means of a particle swarm optimization (PSO) algorithm, which was then used as input in the CNN and LSTM predictive models. To address the inherent black box nature of DL models, six interpretation methods (feature interaction, permutation importance plot (PFIM), bar plot, SHapley Additive exPlanations (SHAP) main plot, heatmap plot, and waterfall plot) were used to interpret the predictive model outputs. Both models (CNN and LSTM) had AUC > 90 and therefore provided excellent accuracy for mapping LS hazard. According to the most accurate model-the CNN predictive model-the range from very low to very high hazard classes occupied 20%, 20%, 25%, 16.3%, and 18.7% of the study area, respectively. According to three plots (bar plot, SHAP main plot, and heatmap plot), which were constructed based on the SHAP value, distance from the well, GDR and DEM were identified as the three most important features with the highest impact on the DL model output. The results of the waterfall plot indicate two effective features consisting of distance from the well and coarse fragment, and two effective features comprising landuse and DEM, contributed negatively and positively to LS, respectively. Overall, these explanation techniques can address critical concerns with respect to the interpretability of sophisticated DL predictive models.

Projection of land susceptibility to subsidence hazard in China using an interpretable CNN deep learning model

Land subsidence is a worldwide geo-environmental hazard. Clarifying the influencing factors of land subsidence hazards susceptibility (LSHS) and their spatial distribution are critical to the prevention and control of subsidence disasters. In this study, we selected natural and anthropogenic features or variables on LSHS and used the interpretable convolutional neural network (CNN) method to successfully construct a LSHS model in China. The model performed well, with AUC and F1-score testing set accuracies reaching 0.9939 and 0.9566, respectively. The interpretable method of SHapley Additive exPlanations (SHAP) was use to elucidate the individual contribution of input features to the predictions of CNN model. The importance ranking of model variables showed that population, gross domestic product (GDP) and groundwater storage (GWS) change are the three major factors that affect China's land subsidence. During year 2004-2016, an area of 237.6 thousand km2 was classified as high and very high LSHS, mainly concentrated in the North China Plain, central Shanxi, southern Shaanxi, Shanghai and the junction of Jiangsu and Zhejiang. There will be 333.82-343.12 thousand km2 of areas located in the high and very high LSHS in the mid-21st century (2030-2059) and 361.9-385.92 thousand km2 of areas in the late-21st century (2070-2099). Future population exposure to high and very high LSHS will be 252.12-270.19 million people (mid-21st century) and 196.14-274.50 million people (late-21st century), respectively, compared with the historical exposure of 210.99 million people. The proportion of future railway and road exposure will reach 14.63 %-14.89 % and 11.51 %-11.82 % in the mid-21st century, and 15.46 %-17.12 % and 12.35 %-13.11 % in the late-21st century, respectively. Our findings provide an important information for creating regional adaptation policies and strategies to mitigate damage induced by subsidence.

Interpretability of simple RNN and GRU deep learning models used to map land susceptibility to gully erosion

Gully erosion possess a serious hazard to critical resources such as soil, water, and vegetation cover within watersheds. Therefore, spatial maps of gully erosion hazards can be instrumental in mitigating its negative consequences. Among the various methods used to explore and map gully erosion, advanced learning techniques, especially deep learning (DL) models, are highly capable of spatial mapping and can provide accurate predictions for generating spatial maps of gully erosion at different scales (e.g., local, regional, continental, and global). In this paper, we applied two DL models, namely a simple recurrent neural network (RNN) and a gated recurrent unit (GRU), to map land susceptibility to gully erosion in the Shamil-Minab plain, Hormozgan province, southern Iran. To address the inherent black box nature of DL models, we applied three novel interpretability methods consisting of SHaply Additive explanation (SHAP), ceteris paribus and partial dependence (CP-PD) profiles and permutation feature importance (PFI). Using the Boruta algorithm, we identified seven important features that control gully erosion: soil bulk density, clay content, elevation, land use type, vegetation cover, sand content, and silt content. These features, along with an inventory map of gully erosion (based on a 70 % training dataset and 30 % test dataset), were used to generate spatial maps of gully erosion using DL models. According to the Kolmogorov-Smirnov (KS) statistic performance assessment measure, the simple RNN model (with KS = 91.6) outperformed the GRU model (with KS = 66.6). Based on the results from the simple RNN model, 7.4 %, 14.5 %, 18.9 %, 31.2 % and 28 % of total area of the plain were classified as very-low, low, moderate, high and very-high hazard classes, respectively. According to SHAP plots, CP-PD profiles, and PFI measures, soil silt content, vegetation cover (NDVI) and land use type had the highest impact on the model's output. Overall, the DL modelling techniques and interpretation methods used in this study proved to be helpful in generating spatial maps of soil erosion hazard, especially gully erosion. Their interpretability can support watershed sustainable management.

Novel integrated modelling based on multiplicative long short-term memory (mLSTM) deep learning model and ensemble multi-criteria decision making (MCDM) models for mapping flood risk

Flood risk assessment is a key step in flood management and mitigation, and flood risk maps provide a quantitative measure of flood risk. Therefore, integration of deep learning - an updated version of machine learning techniques - and multi-criteria decision making (MCDM) models can generate high-resolution flood risk maps. In this study, a novel integrated approach has been developed based on multiplicative long short-term memory (mLSTM) deep learning models and an MCDM ensemble model to map flood risk in the Minab-Shamil plain, southern Iran. A flood hazard map generated by the mLSTM model is based on nine critical features selected by GrootCV (distance to the river, vegetation cover, variables extracted from DEM (digital elevation model) and river density) and a flood inventory map (70% and 30% data were randomly selected as training and test datasets, respectively). The values of all criteria used to assess model accuracy performance (except Cohens kappa for train dataset = 86, and for test dataset = 84) achieved values greater than 90, which indicates that the mLSTM model performed very well for the generation of a spatial flood hazard map. According to the spatial flood hazard map produced by mLSTM, the very low, low, moderate, high and very high classes cover 26%, 35.3%, 20.5%, 11.2% and 7% of the total area, respectively. Flood vulnerability maps were produced by the combinative distance-based assessment (CODAS), the evaluation based on distance from average solution (EDAS), and the multi-objective optimization on the basis of simple ratio analysis (MOOSRA), and then validated by Spearman's rank correlation coefficients (SRC). Based on the SRC, the three models CODAS, EDAS, and MOOSRA showed high-ranking correlations with each other, and all three models were then used in the ensemble process. According to the CODAS-EDAS-MOOSRA ensemble model, 21.5%, 34.2%, 23.7%, 13%, and 7.6% of the total area were classified as having a very low to very high flood vulnerability, respectively. Finally, a flood risk map was generated by the combination of flood hazard and vulnerability maps produced by the mLSTM and MCDM ensemble model. According to the flood risk map, 27.4%, 34.3%, 14.8%, 15.7%, and 7.8% of the total area were classified as having a very low, low, moderate, high, and very high flood risk, respectively. Overall, the integration of mLSTM and the MCDM ensemble is a promising tool for generating precise flood risk maps and provides a useful reference for flood risk management.

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.