Debris flows have been always a serious problem in the mountain areas. Research on the assessment of debris flows susceptibility (DFS) is useful for preventing and mitigating debris flow risks. The main purpose of this work is to study the DFS in the Shigatse area of Tibet, by using machine learning methods, after assessing the main triggering factors of debris flows. Remote sensing and geographic information system (GIS) are used to obtain datasets of topography, vegetation, human activities and soil factors for local debris flows. The problem of debris flow susceptibility level imbalances in datasets is addressed by the Borderline-SMOTE method. Five machine learning methods, i.e., back propagation neural network (BPNN), one-dimensional convolutional neural network (1D-CNN), decision tree (DT), random forest (RF), and extreme gradient boosting (XGBoost) have been used to analyze and fit the relationship between debris flow triggering factors and occurrence, and to evaluate the weight of each triggering factor. The ANOVA and Tukey HSD tests have revealed that the XGBoost model exhibited the best mean accuracy (0.924) on ten-fold cross-validation and the performance was significantly better than that of the BPNN (0.871), DT (0.816), and RF (0.901). However, the performance of the XGBoost did not significantly differ from that of the 1D-CNN (0.914). This is also the first comparison experiment between XGBoost and 1D-CNN methods in the DFS study. The DFS maps have been verified by five evaluation methods: Precision, Recall, F1 score, Accuracy and area under the curve (AUC). Experiments show that the XGBoost has the best score, and the factors that have a greater impact on debris flows are aspect, annual average rainfall, profile curvature, and elevation.

Accurate high spatial resolution snow depth mapping in arid and semi-arid regions is of great importance for snow disaster assessment and hydrological modeling. However, due to the complex topography and low spatial-resolution microwave remote-sensing data, the existing snow depth datasets have large errors and uncertainty, and actual spatiotemporal heterogeneity of snow depth cannot be effectively detected. This paper proposed a deep learning approach based on downscaling snow depth retrieval by fusion of satellite remote-sensing data with multiple spatial scales and diverse characteristics. The (Fengyun-3 Microwave Radiation Imager) FY-3 MWRI data were downscaled to 500 m resolution to match Moderate-resolution Imaging Spectroradiometer (MODIS) snow cover, meteorological and geographic data. A deep neural network was constructed to capture detailed spectral and radiation signals and trained to retrieve the higher spatial resolution snow depth from the aforementioned input data and ground observation. Verified by in situ measurements, downscaled snow depth has the lowest root mean square error (RMSE) and mean absolute error (MAE) (8.16 cm, 4.73 cm respectively) among Environmental and Ecological Science Data Center for West China Snow Depth (WESTDC_SD, 9.38 cm and 5.36 cm), the Microwave Radiation Imager (MWRI) Ascend Snow Depth (MWRI_A_SD, 9.45 cm and 5.49 cm) and MWRI Descend Snow Depth (MWRI_D_SD, 10.55 cm and 6.13 cm) in the study area. Meanwhile, downscaled snow depth could provide more detailed information in spatial distribution, which has been used to analyze the decrease of retrieval accuracy by various topography factors.

Drought is one of the major global natural disasters, and appropriate monitoring systems are essential to reveal drought trends. In this regard, deep learning is a very promising approach for characterizing the non-linear nature of drought factors. We used multi-source remote sensing data such as the Moderate Resolution Imaging Spectroradiometer (MODIS) and Climate Hazards Group Infrared Precipitation with Station (CHIRPS) data to integrate drought impact factors such as precipitation, vegetation, temperature, and soil moisture. The application of convolutional long short-term memory (ConvLSTM) to construct an integrated drought monitoring model was proposed and tested, using the Xinjiang Uygur Autonomous Region as an example. To better compare the monitoring performance of ConvLSTM models, three other classical deep learning models and three classical machine learning models were also used for comparison. The results show that the composite drought index (CDI) output by the ConvLSTM model had a consistent high correlation with the drought rating of the multi-scale standardized precipitation evapotranspiration index (SPEI). The correlation coefficients between the CDI and the multi-scale standardized precipitation index (SPI) were all above 0.5 (p < 0.01), which was highly significant, and the correlation coefficient between CDI-1 and the monthly soil relative humidity at a 10 cm depth was above 0.45 (p < 0.01), which was well correlated. In addition, the spatial distribution of the CDI-6 simulated by the model was highly correlated with the degree of drought expressed by the SPEI-6 observations at the stations. This study provides a new approach for integrated regional drought monitoring.

Water bodies extraction is important in water resource utilization and flood prevention and mitigation. Remote sensing images contain rich information, but due to the complex spatial background features and noise interference, problems such as inaccurate tributary extraction and inaccurate segmentation occur when extracting water bodies. Recently, using a convolutional neural network (CNN) to extract water bodies is gradually becoming popular. However, the local property of CNN limits the extraction of global information, while Transformer, using a self-attention mechanism, has great potential in modeling global information. This paper proposes the MU-Net, a hybrid MixFormer architecture, as a novel method for automatically extracting water bodies. First, the MixFormer block is embedded into Unet. The combination of CNN and MixFormer is used to model the local spatial detail information and global contextual information of the image to improve the ability of the network to capture semantic features of the water body. Then, the features generated by the encoder are refined by the attention mechanism module to suppress the interference of image background noise and non-water body features, which further improves the accuracy of water body extraction. The experiments show that our method has higher segmentation accuracy and robust performance compared with the mainstream CNN- and Transformer-based semantic segmentation networks. The proposed MU-Net achieves 90.25% and 76.52% IoU on the GID and LoveDA datasets, respectively. The experimental results also validate the potential of MixFormer in water extraction studies.