Hodnotenie zosuvného hazardu pomocou multivariačnej a bivariačnej štatistickej analýzy v Košickej kotline (Západné Karpaty)

Abstract

Landslide hazard assessment bivariate multivariate statistical analysis in Košická kotlina basin (Western Carpathians) Landslides are one of the most frequently occurring natural global hazards which threatens human activities in the landscape. This paper employs a landslide hazard assessment in the Košická kotlina basin, Eastern Slovakia, where the State Geologi- cal Institute of Dionýz Štúr registered 1 233 landslides within an area of 328 km 2 . Lithological conditions, elevation, slope, aspect, curvature, land cover and registered landslide raster datasets were evaluated and statistically processed for landslide ha zard assessment. The bivariate and multivariate conditional analysis was applied and the spatial distribution of the slope movement hazard calculated by using the methods of multivariate and bivariate analysis, which was classified into five classes by geo- metrical interval function. This data showed that the majority of the studied area is associated with a medium, high or very high degree of landslide hazard. Receiver operating characteristic (ROC) analysis was performed in order to assess the accuracy of created models and validation pointed that the AUC (area under the curve) in case of multivariate analysis is equal to 0.886 and bivariate analysis to 0.846, which provided an accuracy of 88.6% conditions, elevation, slope, aspect, curvature, land cover and registered landslides) with a 10×10 meter pixel resolution. Two statistical analyses for land- slide hazard assessment were implemented: 1) bivariate analysis using the weights of input parameters and 2) multivariate conditional analysis. The results were reclassified into five classes of landslide hazard susceptibility (very low, low, moderate, high and very high) by geometrical interval function and expressed by two raster datasets which reflected the ha zard calculated by using the multivariate and bivariate classification. Furthermore, data showed that the majority of the evaluated area belongs to the medium, high or very high degree of landslide hazard, which was 45.23% of the total study area for multivariate analysis and 56.85% for bivariate analysis. Validation pointed to a landslide prediction accuracy to 92.81% for multivariate analysis and 91.71% for bivariate analysis. Receiver operating characteristic (ROC) curves and area under curve (AUC) calculated from ROC shows 88.6% accuracy of multivariate analysis and 84.6% of bivariate analysis. The final assessment of landslide hazard susceptibility was performed by using the raster difference calcu- lation between datasets generated by bivariate and multivariate methods. The results showed agreement in landslide hazard classification for 50.37% of study area.