Paper No. 248-2
Presentation Time: 1:20 PM
PIXEL-BASED AND OBJECT-BASED LANDSLIDE MAPPING: A METHODOLOGICAL COMPARISON
Reliable and accurate mapping of event landslides is important for supporting disaster mitigation and management. Earth observation (EO) data provides a valuable basis for gaining information about the spatial distribution and location of landslides. Traditionally, landslides are delineated manually, mainly based on aerial photographs. Though, this is time-consuming, especially for larger areas, and the quality of the results depends on the skills of the interpreter. The increasing availability of very high resolution (VHR) optical satellite imagery and advances in computer technologies led to a clear trend towards semi-automated landslide mapping approaches based on EO data. Present methods can be divided into pixel-based and object-based approaches. So far, pixel-based techniques have been predominately used for information extraction from remote sensing data, but during the past decade object-based image analysis (OBIA) has gained more and more prominence and has recently been recognised as new and evolving paradigm in the field of remote sensing and Geographic Information Science (GIScience). In this study, two workflows for semi-automated landslide mapping, i.e. a pixel-based and an object-based one, are presented and compared. Image analyses are carried out for the Tseng-Wen Reservoir catchment in Taiwan, a region that is highly susceptible to typhoon-triggered landslides due to its fragile geology and steep slopes. The input data consist of a Formosat-2 image with 2 m spatial resolution and four spectral bands, and a 5 m digital elevation model (DEM). Both landslide mapping methods rely on equal thresholds (e.g. slope values) to allow for an objective comparison of classification results. This methodological comparison shows to which extent the parameters of a pixel-based approach can be transferred to an object-based approach. The results of the two methods are quantitatively validated against a manually derived landslide inventory map. Since the demand for consistent landslide mapping and monitoring approaches is steadily increasing, findings of this study will be relevant to the landslide research community.