Monitoring the surface subsidence in mining areas is conducive to the prevention and control of geological disasters, and the prediction and early warning of accidents. Hunan Province is located in South China. The mineral resource reserves are abundant; however, large and medium-sized mines account for a low proportion of the total, and the concentration of mineral resource distribution is low, meaning that traditional mining monitoring struggles to meet the needs of large-scale monitoring of mining areas in the province. The advantages of Interferometric Synthetic Aperture Radar (InSAR) technology in large-scale deformation monitoring were applied to identify and monitor the surface subsidence of coal mining fields in Hunan Province based on a Sentinel-1A dataset of 86 images taken from 2018 to 2020, and the process of developing surface subsidence was inverted by selecting typical mining areas. The results show that there are 14 places of surface subsidence in the study area, and accidents have occurred in 2 mining areas. In addition, the railway passing through the mining area of Zhouyuan Mountain is affected by the surface subsidence, presenting a potential safety hazard.

The prevailing research on forecasting surface deformations within mining territories predominantly hinges on parameter-centric numerical models, which manifest constraints concerning applicability and parameter reliability. Although Multi-Temporal InSAR (MT-InSAR) technology furnishes an abundance of data, the underlying information within these data has yet to be fully unearthed. Consequently, this paper advocates a novel methodology for prognosticating mining area surface deformation by integrating ensemble learning with MT-InSAR technology. Initially predicated upon the MT-InSAR monitoring outcomes, the target variables for the ensemble learning dataset were procured by melding distance-based features with spatial autocorrelation theory. In the ensuing phase, spatial stratified sampling alongside mutual information methodologies were deployed to select the features of the dataset. Utilizing the MT-InSAR monitoring data from the Zixing coal mine in Hunan, China, the relationship between fault slippage and coal extraction in the study area was rigorously analyzed using Granger causality tests and Johansen cointegration assays, thereby acquiring the dataset requisite for training the Bagging model. Subsequently, leveraging the Bagging technique, ensemble models were constructed employing Decision Trees, Support Vector Regression, and Multi-layer Perceptron as foundational estimators. Furthermore, the Tree-structured Parzen Estimator (TPE) optimization algorithm was applied to the Bagging model, resulting in an optimal model for predicting fault slip in mining areas. In comparison with the baseline model, the performance increased by 25.88%, confirming the effectiveness of the data preprocessing method outlined in this study. This result also demonstrates the innovation and feasibility of combining ensemble learning with MT-InSAR technology for predicting mining area surface deformation. This investigation is the first to integrate TPE-optimized ensemble models with MT-InSAR technology, offering a new perspective for predicting surface deformation in mining territories and providing valuable insights for further uncovering the hidden information in MT-InSAR monitoring data.

Most of the coal mines in Southwest China are located in mountainous areas with high vegetation coverage, and most activities are carried out under the mountains. The deformation monitoring and mechanical behavior analysis of the mining area helps reveal the typical mountain deformation and failure mechanism caused by underground mining activities and reduce the risk of mountain collapse in the mining area. In this manuscript, a research method for mountain stability in mining areas is proposed, which combines InSAR deformation monitoring with numerical analysis. Based on the high-precision deformation information obtained by DS-InSAR and the landslide range, a three-dimensional explicit finite difference numerical analysis method was used to reconstruct the landslide model. According to the layout of the coal mining working face, the variation mechanism of overlying stratum stress and the mountain slip in the coal mining process is inverted, and the mechanism of mountain failure and instability in the mining area is analysed. Based on the sentinel data, the experiment performed time series monitoring and inversion analysis of the mountain collapse in Nayong, Guizhou, China. The results show that mining activities a certain distance from the mountain will affect mountain stability, and there are specific mechanisms. From 2015 to 2017, the stress redistribution of overlying strata above the goaf area resulted in dense longitudinal cracks in the landslide body due to coal mining. The mountain is in a continuous damage state, and the supporting force to prevent collapse continues to decrease, resulting in a gradual decrease in landslide stability. Both the time series DS-InSAR monitoring results and numerical simulation results verify the actual occurrence and development of the on-site subsidence.

Surface deformation induced by the extraction of natural resources constitutes a non-stationary spatiotemporal process. Modeling surface deformation time series obtained through Interferometric Synthetic Aperture Radar (InSAR) technology using deep learning methods is crucial for disaster prevention and mitigation. However, the complexity of model hyperparameter configuration and the lack of interpretability in the resulting predictions constrain its engineering applications. To enhance the reliability of model outputs and their decision-making value for engineering applications, this study presents a workflow that combines a Tree-structured Parzen Estimator (TPE)-based Bayesian optimization approach with ensemble inference. Using the Rhineland coalfield in Germany as a case study, we systematically evaluated six deep learning architectures in conjunction with various spatiotemporal coding strategies. Pairwise comparisons were conducted using a Welch t-test to evaluate the performance differences across each architecture under two parameter-tuning approaches. The Benjamini–Hochberg method was applied to control the false discovery rate (FDR) at 0.05 for multiple comparisons. The results indicate that TPE-optimized models demonstrate significantly improved performance compared to their manually tuned counterparts, with the ResNet+Transformer architecture yielding the most favorable outcomes. A comprehensive analysis of the spatial residuals further revealed that TPE optimization not only enhances average accuracy, but also mitigates the model’s prediction bias in fault zones and mineralize areas by improving the spatial distribution structure of errors. Based on this optimal architecture, we combined the ten highest-performing models from the optimization stage to generate a quantile-based susceptibility map, using the ensemble median as the central predictor. Uncertainty was quantified from three complementary perspectives: ensemble spread, class ambiguity, and classification confidence. Our analysis revealed spatial collinearity between physical uncertainty and absolute residuals. This suggests that uncertainty is more closely related to the physical complexity of geological discontinuities and human-disturbed zones, rather than statistical noise. In the analysis of super-threshold probability, the threshold sensitivity exhibited by the mining area reflects the widespread yet moderate impact of mining activities. By contrast, the fault zone continues to exhibit distinct high-probability zones, even under extreme thresholds. It suggests that fault-controlled deformation is more physically intense and poses a greater risk of disaster than mining activities. Finally, we propose an engineering decision strategy that combines uncertainty and residual spatial patterns. This approach transforms statistical diagnostics into actionable, tiered control measures, thereby increasing the practical value of susceptibility mapping in the planning of natural resource extraction.

In the context of anomalous global climate change and the frequent occurrence of droughts and floods, studying trends in the conversion rate between precipitable water vapor (PWV) and actual precipitation in a certain region can help in analyzing the causes of these natural disasters. This paper examines the variation trend in the conversion rate between PWV and actual precipitation on a monthly scale in Hubei from 1960 to 2020. To estimate historical PWV data, we propose a new method for estimating PWV using water vapor pressure based on the RF algorithm. The new method was evaluated by radiosonde data and improved the accuracy by 1 mm over the traditional method in Hubei. Based on this method, we extrapolate the monthly average PWV in Hubei from 1960 to 2020 and analyze the conversion rate between PWV and precipitation during this period. Our results showed that there was no obvious cyclical pattern in the conversion rate in either the longitude or latitude directions. In Hubei, where the topography varies significantly in the longitude direction, the conversion rate is influenced by topography, with the smallest conversion rate being in the transition zone between the mountainous region of western Hubei and the Jianghan Plain. In the latitudinal direction, the conversion rate decreases with increasing latitude.