Landslide susceptibility from topography in Guatemala

Abstract

Listen

Translate article

Landslide susceptibility can be mapped using a number of different methods depending on the data available (e.g., Soeters & van Westen 1996, Guzzetti et al. 1999, Savage et al., this volume). Numerous recent studies use various statistical techniques and incorporate many (5 or more) parameters such as topography, geology, hydrology, and land-use in Geographic Information Systems (GIS) to derive landslide susceptibility (e.g., Chung & Fabbri 1999, Lineback et al. 2001, Santacana et al. 2003). Physically-based methods rely on physical properties of hillslope materials and topographic information from a digital elevation model (DEM) in slope-stability models (e.g., Montgomery & Dietrich 1994, Jibson et al. 2000, Savage et al. 2003). In many parts of the world however, the abundance of data such as that described above is not available, but the need for landslide susceptibility maps is great (e.g., Pallas et al., in press). A question asked is: Can reliable susceptibility maps be produced from limited data? Fabbri et al. (2003) suggested that this is not only possible, but, in at least one example, possibly more accurate. They outlined seven “myths” associated with GIS modeling of landslide hazard. Two of these myths were: 1) the more data layers we have, the better the prediction, and 2) the only thing we have is a DEM from satellite or aerial imagery or a topographic base-map; therefore we cannot make a prediction map. They evaluated the effectiveness of a landslide susceptibility map made using 6 data layers (including geology, surficial materials, land use, slope, elevation, and aspect) vs. one made using 3 data layers (slope, elevation, and aspect) and found that the 3 data layers (derived exclusively from a DEM) provided better results, seemingly indicating that topography was the dominant control in determining landslide location. Some of our work at the U.S. Geological Survey (USGS) involves responding to landslide disasters. One of these disasters was caused by Hurricane Mitch in Guatemala in 1998 (Bucknam et al. 2001). As part of our work following Hurricane Mitch we produced a landslide susceptibility map using two types of data, a landslide inventory map, and a DEM. Because of a lack of data, we were forced to confront the two “myths” given above. This paper grew out of this confrontation. In what follows, we describe our effort to map relative landslide susceptibility using inventory and DEM data. We estimate relative landslide susceptibility based on a comparison (a ratio) of topographic parameters at landslide DEM cells to the same parameters at a random sampling of the entire population of DEM cells. Although ratio methods are not new in landslide susceptibility mapping, to our knowledge, our approach differs from previous work in the way we combine parameters using a moving-count circle approach (Savage et al. 2001) to produce a susceptibility map for a 980 km study area in east-central Guatemala that was impacted by Hurricane Mitch.