The study of land use/land cover (LULC) changes plays an important guiding role in regional ecological protection and sustainable development policy formulation. Especially, the simulation study of the future scenarios may provide a hypothetical prospect which could help to determine the rationality of current and future development policies. In order to support the ecological protection and high-quality development strategy of the Yellow River Basin proposed by the Chinese government, the Great Yellow River Region (GYRR) is taken as the research area. The multi-period land cover data are used to carry out the analysis of land cover changes. The MOLUSCE (Modules for Land Use Change Simulations) plugin of QGIS software is used to carry out a land cover simulation and prediction study for 2030 on a large regional scale. Finally, the land cover status in the mountainous areas of the GYRR is analyzed thoroughly. The results show a decrease in agricultural land and increase in forest land during the past 25 years from 1995 to 2020, and that this trend would continue to 2030. The landscape pattern index analysis indicates that the land cover in the GYRR has become more and more abundant, and the degree of fragmentation has become higher and higher, while landscape patches were more evenly distributed in the GYRR until 2020. On the other hand, the landscape pattern would tend to achieve a certain degree of stability in 2030. The decrease in farmland and the increase in forest land illustrate the efforts made by the GYRR residents and governments in improving the ecological environment under the policy of returning farmland to forests and grasslands. On the other hand, although the residential areas in the mountainous areas are far away from the mountain hazard historical points because of consideration during construction with the help of the development of disaster prevention and mitigation over the years, there could be problem of rapid and haphazard urbanization. It is worth mentioning here that the harmonious and sustainable development of people and land in the GYRR mountainous areas still requires a large amount of effort.

The eastern margin of the Qinghai-Tibet Plateau is an extreme topography transition zone, and characterized by significant vegetation zonation, in addition to geographic features (such as enormous topographic relief and active tectonics) that control the occurrence of debris flows, which are rapid, surging flows of water-charged clastic sediments moving along a steep channel and are one of the most dangerous mountain hazards in this region. There is thus an urgent need in this region to conduct a regional-scale debris flow susceptibility assessment to determine the spatial likelihood of a debris flow occurrence and guarantee the safety of people and property, in addition to the smooth operation of the Sichuan-Tibet transport corridor. It is, however, a challenging task to estimate the region’s debris flow susceptibility while taking into consideration the comprehensive impacts of vegetation on the occurrence of debris flows, such as the positive effect of root anchoring and the negative effect of vegetation weight loads. In this study, a novel regional-scale susceptibility assessment method was constructed by integrating state-of-the-art machine learning algorithms (such as support vector classification (SVC), random forest (RF), and eXtreme Gradient Boosting (XGB)) with the removing outliers (RO) algorithm and particle swarm optimization (PSO), allowing the impacts of vegetation on debris flow initiation to be integrated with the topographical conditions, hydrological conditions, and geotechnical conditions. This method is finally applied to assess the regional-scale susceptibility of debris flows in the Dadu River basin on the eastern margin of the Qinghai-Tibet Plateau. The study results show that (i) all hybrid machine learning techniques can effectively predict the occurrence of debris flows in the extreme topography transition zone; (ii) the hybrid machine learning technique RO-PSO-SVC has the best performance, and its accuracy (ACC) is 0.946 and the area under the ROC curve (AUC) is 0.981; (iii) the RO-PSO algorithm improves SVC, RF, and XGB performance (according to the ACC value) by 3.84%, 2.59%, and 5.94%, respectively; and (iv) the contribution rate of ecology-related variables is almost only one-tenth that of topography- and hydrology-related factors, according to the factor important analysis for RO-PSO-SVC. Furthermore, debris flow susceptibility maps for the Dadu River basin were created, which can be used to assess and mitigate debris flow hazards.

The Hengduan Mountains region (HMR) is vulnerable to flash flood disasters, which account for the largest proportion of flood-related fatalities in China. Flash flood regionalization, which divides a region into homogeneous subdivisions based on flash flood-inducing factors, provides insights for the spatial distribution patterns of flash flood risk, especially in ungauged areas. However, existing methods for flash flood regionalization have not fully reflected the spatial topology structure of the inputted geographical data. To address this issue, this study proposed a novel framework combining a state-of-the-art unsupervised Graph Neural Network (GNN) method, Dink-Net, and Shapley Additive exPlanations (SHAP) for flash flood regionalization in the HMR. A comprehensive dataset of flash flood inducing factors was first established, covering geomorphology, climate, meteorology, hydrology, and surface conditions. The performances of two classic machine learning methods (K-means and Self-organizing feature map) and three GNN methods (Deep Graph Infomax (DGI), Deep Modularity Networks (DMoN), and Dilation shrink Network (Dink-Net)) were compared for flash-flood regionalization, and the Dink-Net model outperformed the others. The SHAP model was then applied to quantify the impact of all the inducing factors on the regionalization results by Dink-Net. The newly developed framework captured the spatial interactions of the inducing factors and characterized the spatial distribution patterns of the factors. The unsupervised Dink-Net model allowed the framework to be independent from historical flash flood data, which would facilitate its application in ungauged mountainous areas. The impact analysis highlights the significant positive influence of extreme rainfall on flash floods across the entire HMR. The pronounced positive impact of soil moisture and saturated hydraulic conductivity in the areas with a concentration of historical flash flood events, together with the positive impact of topography (elevation) in the transition zone from the Qinghai–Tibet Plateau to the Sichuan Basin, have also been revealed. The results of this study provide technical support and a scientific basis for flood control and disaster reduction measures in mountain areas according to local inducing conditions.

Remote sensing change detection (CD) involves identifying differences between two satellite images of the same geographic area taken at different times. It plays a critical role in applications such as urban planning and disaster management. Traditional CD methods rely on manually extracted features, which often lack robustness and accuracy in capturing the details of objects. Recently, deep learning-based methods have expanded the applications of CD in high-resolution remote sensing images, yet they struggle to fully utilize the multi-level features extracted by backbone networks, limiting their performance. To address this challenge, we propose a Semantic-Guided Cross-Attention Network (SCANet). It introduces a Hierarchical Semantic-Guided Fusion (HSF) module, which leverages high-level semantic information to guide low-level spatial details through an attention mechanism. Additionally, we design a Cross-Attention Feature Fusion (CAFF) module to establish global correlations between bitemporal images, thereby improving feature interaction. Extensive experiments on the IWHR-data and LEVIR-CD datasets demonstrate that SCANet significantly outperforms existing State-of-the-Art (SOTA) methods. Specifically, the F1-score and the Intersection over Union (IoU) score are improved by 2.002% and 3.297% on the IWHR-data dataset and by 0.761% and 1.276% on the LEVIR-CD dataset, respectively. These results validate the effectiveness of semantic-guided fusion and cross-attention feature interaction, providing new insights for advancing change detection research in high-resolution remote sensing imagery.

In recent years, mountainous areas in China have faced frequent geological hazards, including landslides, debris flows, and collapses. Effective simulation of these events requires a solver for shallow water equations (SWEs). Traditional numerical methods, such as finite difference and finite volume, face challenges in discretizing convection flux terms, while theory-based models need to account for various factors such as shock wave capturing and wave propagation direction, demanding a high-level understanding of the underlying physics. Previous deep learning (DL)-based SWE solvers primarily focused on constructing direct input–output mappings, leading to weak generalization properties when terrain data or stress constitutive relations change. To overcome these limitations, this study introduces a novel SWE solver that combines theory and data-driven methodologies. The core idea is to use artificial neural networks to compute convection flux terms, and to reduce modeling complexity. Theory-based modeling is used to tackle complex terrain and friction terms for the purpose of ensuring generalization. Our method surpasses challenges faced by previous DL-based solvers in capturing terrain and stress variations. We validated our solver’s capabilities by comparing simulation results with analytical solutions, real-world disaster cases, and the widely used Massflow software-generated simulations. This comprehensive comparison confirms our solver’s ability to accurately simulate hazard scenarios and showcases strong generalization on varying terrain and land surface friction. Our proposed method effectively addresses DL-based solver limitations while simplifying the complexities of theory-driven numerical methods, offering a promising approach for hazard dynamics simulation.