Abstract:Debris flow has always been a serious problem in mountainous areas. Accurate debris flow susceptibility (DFS) assessment and interpretable prediction results play an important role in the prevention and control of debris flow disasters. Some commonly used machine learning algorithms based on Boosting ensemble techniques were widely used in the study of geohazard susceptibility due to its excellent predictive ability. However, the Categorical Boosting (CatBoost) and Natural Gradient Boosting (NGBoost) have not yet been applied in the field of DFS assessment, and few geohazard studies systematically compare and research these boosting-based algorithms. Meanwhile, previous researches have mostly focused on comparing the predictive ability of algorithms, identifying the susceptibility zones of the entire study area, and ranking the importance of the indicators, but little thorough analysis of the relationship between the indicators and debris flow susceptibility on different types of construction land. The aims of this study were to explore the optimal boosting-based DFS model, and the distribution characteristics and change rules of DFS in the study area, so as to provide decision supports for debris flow disaster prevention and reduction. This was the first time that six boosting-based machine learning algorithms have been compared in the study of DFS assessment. After determining the optimal model, the change rules of indicators in the entire study area and two types of construction lands under different DFS levels were studied respectively. An eXplainable Artifcial Intelligence (XAI) method called SHapley Additive exPlantations (SHAP), combined with zonal statistics function in geographic information system (GIS) were adopted to explore how each indicator affects the occurrence of debris flows. The results showed that the CatBoost performed best and provided the most reasonable DFS result among six boosting-based models. We found that debris flows were more likely to occur along rivers and construction lands at low altitude. Rural areas faced more stronger pressure from rainfall and were featured by worse disaster-breeding environment than urban areas. This research enriches the application of machine learning in DFS assessment, explores the changing trends of indicators between different DFS levels, and provides suggestions for better debris flow disaster prevention and mitigation management.

Predictive Performances of Ensemble Machine Learning Algorithms in Landslide Susceptibility Mapping Using Random Forest, Extreme Gradient Boosting (XGBoost) and Natural Gradient Boosting (NGBoost)

Advanced hyperparameter optimization for improved spatial prediction of shallow landslides using extreme gradient boosting (XGBoost)

Application of Tree-Based Ensemble Models to Landslide Susceptibility Mapping: A Comparative Study

Improving the forecast performance of landslide susceptibility mapping by using ensemble gradient boosting algorithms

Comparison of tree-based ensemble learning algorithms for landslide susceptibility mapping in Murgul (Artvin), Turkey

Landslide Susceptibility Modeling Based on GIS and Novel Bagging-Based Kernel Logistic Regression

A Comprehensive Assessment of XGBoost Algorithm for Landslide Susceptibility Mapping in the Upper Basin of Ataturk Dam, Turkey

Improving the Landslide Susceptibility Prediction Accuracy by Using Genetic Algorithm Optimized Machine Learning Approach

A Novel Performance Assessment Approach using Photogrammetric Techniques for Landslide Susceptibility Mapping with Logistic Regression, ANN and Random Forest

Application of alternating decision tree with AdaBoost and bagging ensembles for landslide susceptibility mapping

Landslide Susceptibility Mapping using Machine Learning Algorithm

Performance Evaluation of the GIS-based Data Mining Techniques of Best-First Decision Tree, Random Forest, and Naïve Bayes Tree for Landslide Susceptibility Modeling

Integrating Machine Learning Ensembles for Landslide Susceptibility Mapping in Northern Pakistan

Earthquake-Induced Landslide Susceptibility Assessment Using a Novel Model Based on Gradient Boosting Machine Learning and Class Balancing Methods

Improved landslide assessment using support vector machine with bagging, boosting, and stacking ensemble machine learning framework in a mountainous watershed, Japan

Landslide Susceptibility Mapping and Driving Mechanisms in a Vulnerable Region Based on Multiple Machine Learning Models

Landslide susceptibility mapping and sensitivity analysis using various machine learning models: a case study of Beas valley, Indian Himalaya

Predicting and analyzing flood susceptibility using boosting-based ensemble machine learning algorithms with SHapley Additive exPlanations

Landslide susceptibility mapping using J48 Decision Tree with AdaBoost, Bagging and Rotation Forest ensembles in the Guangchang area (China)

Explainable AI Integrated Feature Selection for Landslide Susceptibility Mapping using TreeSHAP

Debris flow susceptibility assessment based on boosting ensemble learning techniques: a case study in the Tumen River basin, China