The knowledge expression on debris flow potential analysis through PCA + LDA and rough sets theory: a case study of Chen-Yu-Lan watershed, Nantou, Taiwan

Debris flow is often performed through identifying and analyzing the soil condition, hydraulic, geomorphological factors and vegetation conditions. In the present study, a spatial information analysis system is combined with a linear statistical method (principle components analysis with linear discriminant analysis, PCA + LDA) and an advanced data mining technique (discrete rough sets, DRS) to investigate the debris flow occurrence based on geomorphological and vegetation conditions factors. The analyzed data sources include (1) digital elevation model: to investigate the variation in the landscape, and (2) remote sensing data: to analyze the vegetation and plant conditions on the ground surface. The objective of this research is to define a method with the ability to forecast the level of debris flow susceptibility through the parallel study of statistical outcomes (PCA + LDA) and data mining results (DRS). The outcomes from PCA + LDA are inadequate due to the thresholds of the influenced variables not being examined. In this study, the DRS approach not only showed satisfactory results for the thresholds of influenced variables in the study area, but also the occurrence rules of debris flow are generated. Finally, the results show superior classification accuracy (70.8% for debris flow occurrence) for the DRS method over those of PCA + LDA analysis (54.2% for debris flow occurrence) for the analysis of debris flow occurrence. Therefore, this is an encouraging preliminary approach in the hazard assessment of debris flow.