Piping is a major factor contributing to river embankment breaches, particularly during flood season in small and medium rivers. To reduce the costs of earth rock embankment inspections, avoid the need for human inspectors and enable the quick and widespread detection of piping hazards, a UAV image-acquisition function was introduced in this study. Through the collection and analysis of thermal infrared and visible (TIR & V) images from several piping field simulation experiments, temperature increases, and diffusion centered on the piping point were discovered, so an automatic algorithm for piping identification was developed to capture this phenomenon. To verify the identification capabilities, the automatic identification algorithm was applied to detect potential piping hazards during the 2022 flooding of the Dingjialiu River, Liaoning, China. The algorithm successfully identified all five piping hazard locations, demonstrating its potential for detecting embankment piping.

Investigating aeolian saltation over typical desert surfaces is crucial for understanding aeolian patterns and geomorphic development in arid and semi-arid regions. However, ground-based synchronous measurements on various land surfaces are scarce, and usually in limited measurement duration. The Alxa Plateau is characterized by strong wind, extensive wind erosion, and is one of the main dust source areas in China. In this study, we conducted year-long synchronous sand saltation measurements on five typical desert surfaces (gravel Gobi, fine-gravel flat, sand sheet, mud-flat nebkha, and salt-flat nebkha), in the transitional zone between the downstream plain of Heihe River and Badain Jaran Sand Desert. The measurements revealed that, under the local arid climate, the average annual 2 m-high wind speed on gravel Gobi was similar to that on sand sheet, but 1.26 to 1.43 times higher than that on the other three surface types. Sand saltating number on sand sheet was 1 order of magnitude greater than gravel Gobi and 2-3 orders greater than the other surfaces. The majority of sand saltation activities occurred within 10-30℃ air temperature and 10-30% relative humidity. Surface roughness and sand availability might be important factors affecting wind speed and aeolian saltation activity. The majority of sand saltation was chiefly caused by high wind speed and long-duration sand-transporting events. We detected the average particle kinetic energy decreased with increasing wind speed and concentration of saltating sand, implying that enhanced collisions among numerous sand particles at higher wind speeds might substantially reduce the abrasive efficiency of individual sand particles.

Against the backdrop of global climate change and accelerating urbanization, flood disasters have become increasingly frequent and severe, posing a significant threat to regional sustainable development. Traditional flood susceptibility assessments have predominantly focused on natural factors, often lacking systematic exploration of model interpretability and human adaptive behavior. To address these gaps, this study proposes an integrated flood disaster driving factor analysis and spatial prediction framework. This framework synthesizes factor preprocessing, machine learning modeling, interpretability analysis, and spatial prediction to systematically reveal the mechanisms underlying the spatial distribution of flood disasters and achieve high-precision forecasting. Three gradient boosting models—CatBoost, LightGBM, and XGBoost—were employed for training and comparison, with model performance evaluated using multiple metrics. Furthermore, the SHAP method was applied to interpret and rank feature importance. For spatial prediction, the classification probability outputs from the optimal model were used to generate flood susceptibility maps. Susceptibility levels were classified using the natural breaks method, highlighting the spatial distribution characteristics of high-susceptibility zones. Finally, by analyzing land use and settlement changes within these high-risk areas from 2000 to 2020, the study explores the adaptation and coping strategies of human activities in response to flood risk. By integrating machine learning with interpretability analysis, this research provides a systematic and interpretable spatial prediction tool for understanding flood mechanisms and supporting risk prevention and control.