Debris flow susceptibility mapping is considered to be useful for hazard prevention and mitigation. As a frequent debris flow area, many hazardous events have occurred annually and caused a lot of damage in the Sichuan Province, China. Therefore, this study attempted to evaluate and compare the performance of four state-of-the-art machine-learning methods, namely Logistic Regression (LR), Support Vector Machines (SVM), Random Forest (RF), and Boosted Regression Trees (BRT), for debris flow susceptibility mapping in this region. Four models were constructed based on the debris flow inventory and a range of causal factors. A variety of datasets was obtained through the combined application of remote sensing (RS) and geographic information system (GIS). The mean altitude, altitude difference, aridity index, and groove gradient played the most important role in the assessment. The performance of these modes was evaluated using predictive accuracy (ACC) and the area under the receiver operating characteristic curve (AUC). The results of this study showed that all four models were capable of producing accurate and robust debris flow susceptibility maps (ACC and AUC values were well above 0.75 and 0.80 separately). With an excellent spatial prediction capability and strong robustness, the BRT model (ACC = 0.781, AUC = 0.852) outperformed other models and was the ideal choice. Our results also exhibited the importance of selecting suitable mapping units and optimal predictors. Furthermore, the debris flow susceptibility maps of the Sichuan Province were produced, which can provide helpful data for assessing and mitigating debris flow hazards.

The need to protect forests and enhance the capacity of mountain ecosystems is highlighted in the U.N.’s Sustainable Development Goal (SDG) 15. The worst-hit areas of the 2008 Wenchuan Earthquake in southwest China were mountainous regions with high biodiversity and the impacted area is typical of other montane regions, with the need for detecting vegetation changes following the impacts of catastrophes. While the widely used remotely sensed vegetation indicator NDVI is available from various satellite data sources, these satellites are available for different monitoring periods and durations. Combining these datasets proved challenging to make a continuous characterization of vegetation change over an extended time period. In this study, compared with linear regression, multiple linear regression, and random forest, Convolutional Neural Networks (CNNs) performed best with an average R² of 0.819 (leave-one-out cross-validation). Thus, the CNNs model was selected to establish the map of the overlapping periods of two remote-sensing products: SPOT-VGT NDVI and PROBA-V NDVI, to reconstruct a SPOT-VGT NDVI for the period from June 2014 to December 2018 in the worst-hit areas of the Wenchuan earthquake. We analyzed the original and reconstructed SPOT-VGT NDVI in the hard-hit areas of the Wenchuan earthquake from 1999 to 2018, and we concluded that NDVI showed an overall upward trend throughout the study period, but experienced a sharp decline in 2008 and reached its lowest value a year later (2009). Vegetation recovery was rapid from 2009 until 2011 after which, it returned to a pattern of slower natural growth (2012–2018). The Longmenshan fault zone experienced the greatest vegetation damage and initiation of recovery there has caused the overall regional average recovery to lag by 1–2 years. In areas where the land was denuded of vegetation (i.e., effectively all vegetation was stripped from the surface) after the earthquake, the damage exceeded what was experienced anywhere else in the entire study area, and by 2018 it remained unrestored. In the 15 years since the earthquake, the areas that were denuded were expected to recover to the level of restoration equivalent with the NDVI of 2007, as was the case in other earthquake-damaged regions. In addition to the earthquake and the immediate loss of vegetation, the Chinese government’s Grain for Green Policy, the elevation ranges within the region, the forest’s phenological conditions, and human activities all had an impact on vegetation recovery and restoration. The reconstructed NDVI provides a long-term continuous record, which contributes to the identifying changes that are improving predictive forest recovery models and to better vegetation management following catastrophic disturbances, such as earthquakes.

Hydrological changes combined with earthquakes easily trigger secondary disasters, including geological hazards. The secondary hazard of precipitation is the main disaster type in the Longmenshan Area (China). The 2008 Wenchuan earthquake caused more than 60,000 landslides, severely affecting rural households. This study aimed to answer two questions: (1) How did households adapt to the landslide-prone post-earthquake environment? (2) How will the households’ adaptation strategies change if landslide frequency changes? Different post-disaster adaptation strategies of households in Longmenshan Town, Sichuan, China were identified through a questionnaire survey and then clustered into groups based on similarity using a K-means algorithm. Afterward, a gradient boosting decision tree (GBDT) was used to predict change in adaptation strategies if there was a change in the frequency of landslides. The results show that there are three types of landslide adaptation strategies in the study area: (1) autonomous adaptation; (2) policy-dependent adaptation; and (3) hybrid adaptation, which is a mixture of the first two types. If the frequency of landslides is increased, then around 5% of households previously under the autonomous adaptation type would be converted to policy-dependent and hybrid adaptation types. If the frequency of landslides is reduced, then around 5% of households with policy-dependent adaptation strategies would be converted to the autonomous adaptation type. This exploratory study provides a glimpse of how machine learning can be utilized to predict how adaptation strategies would be modified if hazard frequency changed. A follow-up long-term study in Longmenshan Town is needed to confirm whether the predictions are indeed correct.