The main goal of this study is to produce a landslide susceptibility map in the Wanzhou section of the Three Gorges reservoir area (China) with a weighted gradient boosting decision tree (weighted GBDT) model. According to the current research on landslide susceptibility mapping (LSM), the GBDT method is rarely used in LSM. Furthermore, previous studies have rarely considered the imbalance of landslide samples and simply regarded the LSM problem as a binary classification problem. In this paper, we considered LSM as an imbalanced learning problem and obtained a better predictive model using the weighted GBDT method. The innovations of the article mainly include the following two points: introducing the GBDT model into the evaluation of landslide susceptibility; using the weighted GBDT method to deal with the problem of landslide sample imbalance. The logistic regression (LR) model and gradient boosting decision tree (GBDT) model were also used in the study to compare with the weighted GBDT model. Five kinds of data from different data source were used in the study: geology, topography, hydrology, land cover, and triggered factors (rainfall, earthquake, land use, etc.). Twenty nine environmental parameters and 233 landslides were used as input data. The receiver operating characteristic (ROC) curve, the area under the ROC curve (AUC) value, and the recall value were used to estimate the quality of the weighted GBDT model, the GBDT model, and the LR model. The results showed that the GBDT model and the weighted GBDT model had a higher AUC value (0.977, 0.976) than the LR model (0.845); the weighted GBDT model had a little higher AUC value (0.977) than the GBDT model (0.976); and the weighted GBDT model had a higher recall value (0.823) than the GBDT model (0.426) and the LR model (0.004). The weighted GBDT method could be considered to have the best performance considering the AUC value and the recall value in landslide susceptibility mapping dealing with imbalanced landslide data.

The early identification of potential hazards is crucial for landslide early warning and prevention and is a key focus and challenging issue in landslide disaster research. The challenges of traditional investigation and identification methods include identifying potential hazards of landslides triggered by heavy rainfall and mapping areas susceptible to landslides based on rainfall conditions. This article focuses on the problem of early identification of rainfall-induced accumulation landslide hazards and an early identification method is proposed, which is “first identifying the accumulation that is prone to landslides and then determining the associated rainfall conditions”. This method is based on identifying the distribution and thickness of accumulation, analyzing the rainfall conditions that trigger landslides with varying characteristics, and establishing rainfall thresholds for landslides with different accumulation characteristics, ultimately aiming to achieve early identification of accumulation landslide hazards. In this study, we focus on the Zigui section of the Three Gorges Reservoir as study the area, and eight main factors that influence the distribution and thickness of accumulation are extracted from multi-source data, then the relative thickness information extraction model of accumulation is established by using the BP neural network method. The accumulation distribution and relative thickness map of the study area are generated, and the study area is divided into rocky area (less than 1 m), thin (1 to 5 m), medium (5 to 10 m), and thick area (thicker than 10 m) according to accumulation thickness. Rainfall is a significant trigger for landslide hazards. It increases the weight of the sliding mass and decreases the shear strength of soil and rock layers, thus contributing to landslide events. Data on 101 rainfall-induced accumulation landslides in the Three Gorges Reservoir area and rainfall data for the 10 days prior to each landslide event were collected. The critical rainfall thresholds corresponding to a 90% probability of landslide occurrence with different characteristics were determined using the I-D threshold curve method. Prediction maps of accumulation landslide hazards under various rainfall conditions were generated by analyzing the rainfall threshold for landslides in the Three Gorges Reservoir area, serving as a basis for early identification of rainfall-induced accumulation landslides in the region. The research provides a method for the early identification of landslides caused by heavy rainfall, delineating landslide hazards under different rainfall conditions, and providing a basis for scientific responses, work arrangements, and disaster prevention and mitigation of landslides caused by heavy rainfall.

Landslide susceptibility evaluation is an indispensable part of disaster prevention and mitigation work. Selecting effective evaluation methods and models for landslide susceptibility assessment is of significant importance. This study focuses on selected areas in Yunyang County, Chongqing City. By interpreting high-resolution satellite remote sensing images from before and after heavy rainfall on 31 August 2014, the distribution of rainfall-induced accumulation landslides was obtained. To evaluate the susceptibility of accumulation landslides, we have equated evaluation factors to accumulation distribution prediction factors. Eight evaluation factors were extracted using multi-source data, including lithology, elevation, slope, remote sensing image texture features, and the normalized difference vegetation index (NDVI). Various machine learning models, such as Random Forest (RF), Support Vector Machine (SVM), and BP Neural Network models, were employed to assess the susceptibility of rainfall-induced accumulation landslides in the study area. Subsequently, the accuracy of the evaluation models was compared and verified using the Receiver Operating Characteristic (ROC) curve, and the evaluation results were analyzed. Finally, the developed Random Forest model was applied to Gongping Town in Fengjie County to verify its applicability in other regions. The findings indicate that the complex geological conditions and the unique tectonic erosion landform patterns in the northeastern region of Chongqing not only make this area a center of heavy rainfall but also lead to frequent and recurrent rainfall-induced landslides. The Random Forest model effectively reflects the development characteristics of accumulation landslides in the study area. High and very high susceptibility zones are concentrated in the northern and central regions of the study area, while low and moderate susceptibility zones predominantly occupy the mountainous and riverside areas. Landslide susceptibility mapping in the study area shows that the Random Forest model yields reasonably graded results. Elevation, remote sensing image texture features, and lithology are highly significant factors in the evaluation system, indicating that the development factors of slope geological disasters in the study area are mainly related to topography, geomorphology, and lithology. The landslide susceptibility evaluation results in Gongping Town, Fengjie County, validate the applicability of the Random Forest model developed in this study to other regions.

Background: Following the impoundment of the Three Gorges Reservoir, deposit-layer landslides in the reservoir basin have experienced cyclic deformation under the combined influence of seasonal precipitation and periodic water-level operations. The deformation amplitude accumulates annually, continuously bringing uncertainty to the safety of the reservoir basin. Purpose: This study takes the Tanjiahe landslide (a typical feature in the Three Gorges Reservoir region) as the research object. By integrating its recent daily displacement measurements, daily rainfall data, and reservoir water-level observations, this research aims to establish a deep learning-based prediction model to forecast the landslide’s long-term deformation trends. Specifically, the aggregated displacement time series is segmented into a trend term and a cyclic term, so as to separately characterize the landslide’s long-term evolution and periodic dynamic behavior. Methods: First, Grey Relational Analysis (GRA) and Pearson’s correlation analysis are used to identify the dominant influencing factors of the Tanjiahe landslide (including antecedent rainfall, reservoir water-level variation characteristics, and past displacement increments) as input features. Then, considering the timeseries nature of monitoring data and limited sample size constraints, a Long Short-Term Memory (LSTM) network is developed to predict the landslide’s cyclic displacement, while the trend component is modeled via Support Vector Machine (SVM) regression. Results: The results indicate that the LSTM method significantly outperforms in depicting the complex nonlinear time-dependent features of landslide deformation; the SVM method stably captures the aggregate deformation trajectory and shows strong performance in low-frequency trend regression. Additionally, Rescaled Range (R/S) analysis of observed versus predicted sequences reveals significant long-term persistence in the Tanjiahe landslide’s deformation evolution at its current phase. Conclusion: This study demonstrates that the combined LSTM-SVM model can effectively predict the Tanjiahe landslide’s long-term deformation. The identified long-term persistence of deformation provides a basis for long-term risk assessment and early warning of this landslide.