Structure and Characteristics of Landslide Input Data and Consequences on Landslide Susceptibility Assessment and Prediction Capability

Abstract

The main objective of this study is to assess the influence of landslide representation format (i.e. landslide represented as points or areas) in landslide susceptibility results, especially at scales that can directly interfere with spatial planning. For the study area of Rio Grande da Pipa basin, Arruda dos Vinhos, Portugal, the Information Value method is used to statistically integrate two rotational slides groups (deep and shallow) and a dataset of independent predisposing geo-environmental factors. For both landslide groups, landslides were represented by: (i) the landslide area; (ii) the landslide depletion area; (iii) the centroid of landslide area; and (iv) the centroid of landslide depletion area. Additionally each group was randomly partitioned in two equivalent landslide sub-groups (50–50%), one for modeling and the other for independent validation of the landslide susceptibility maps. The evaluation of the landslide representation format on the prediction capacity of each landslide susceptibility model was based on Receiving Operating Characteristic curves and in the calculation of Area Under the Curve. As main results this work points out the sensitivity of landslide susceptibility models prediction capability to the landslide representation format. Consistently, for both landslide groups, the better predictive results were achieved by modeling with the landslide depletion area and validating with landslide depletion area and the worst results by modeling with landslide centroid and validating with the landslides area. Furthermore the same hierarchy of landslide representation formats regarding the prediction capability of the landslide susceptibility models was recorded independently of being deep or shallow rotational slide types.