Representing landslides as polygon (areal) or points? How different data types influence the accuracy of landslide susceptibility maps

Abstract

In the literatures, discussions on the accuracy of different models for landslide analysis have been discussed widely. However, to date, arguments on the type of input data (landslides in the form of point or polygon) and how they affect the accuracy of these models can hardly be found. This study assesses how different types of data (point or polygon) applied to the same model influence the accuracy of the model in determining areas susceptible to landsliding. A total of 137 landslides was digitised as polygon (areal) units and then transformed into points; forming two separate datasets both representing the same landslides within the study area. These datasets were later separated into training and validation datasets. The polygon unit dataset uses the area density technique reported as percentage, while the point data uses the landslide density technique, as means of assigning weighting to landslide factor maps to generate the landslide susceptibility map that is based on the analytical hierarchy process (AHP) model. Both data groups show striking differences in terms of mapping accuracy for both training and validation datasets. The final landslide susceptibility map using area density (polygon) as input only has 48% (training) and 35% (validation) accuracy. The accuracy for the susceptibility map using the landslide density as input data achieved 89% and 82% for both training and validation datasets, respectively. This result showed that the selection of the type of data for landslide analysis can be critical in producing an acceptable level of accuracy for the landslide susceptibility map. The authors hope that the finding of this research will assist landslide investigators to determine the appropriateness of the type of landslide data because it will influence the accuracy of the final landslide potential map.