Dependence of debris flow susceptibility maps on sampling strategy with data-driven grid-based model

• DOI: 10.1016/j.ecolind.2024.112534
• Publication Date: 2024-08-27T15:30:19Z
• Authors: Ning Jiang; Fenghuan Su; Ruilong Wei; Yu Huang; Wen Jin; Peng Huang; Qing Zeng
• Keywords: Centroid; Sample (material); Polygon (computer graphics); Debris flow

Different sampling strategies produce varying sample data, serve as the primary input data and directly affect accuracy of predictions in data-driven grid-based susceptibility models. This study analyzes variation debris flow maps (DFSMs) generated by various strategies. The area is Yingxiu region China, where six were applied, including three locations (deposition area, runout source area) two types (centroid polygon) for inventory. effectiveness 10 conditioning factors used to build model was assessed using Pearson correlation coefficient, variance inflation factor, information gain ratio (IGR) techniques. We then Weight Evidence (WofE), Logistic Regression (LR), Deep Neural Network (DNN) models DFSMs quantify their performance receiver operating characteristic curve (ROC), Accuracy (ACC), Precision, F1 score, Recall. results show that WofE (AUC: 0.754–0.960), LR 0.761–0.965), DNN 0.786–0.976) all perform well, but dominant depend strongly on strategies, especially location. If areas are excessively large span across different factor class labels, or if there a concentration either small within specific region, centroid polygon may differ even be contradictory.We recommend: (1) determining based research objectives provide more accurate evaluation results; (2) selecting type first considering size. aforementioned conditions not present, quicker convenient strategy can chosen; (3) an appropriate ensuring initial samples paramount before producing DFSMs.