2008 Joint Meeting of The Geological Society of America, Soil Science Society of America, American Society of Agronomy, Crop Science Society of America, Gulf Coast Association of Geological Societies with the Gulf Coast Section of SEPM

Paper No. 5
Presentation Time: 2:30 PM

The Effect of Pixel Size on Cartographic Representation of Shallow and Deep-Seated Landslide, and Its Collateral Effects on the Forecasting of Landslides by SINMAP and Multiple Logistic Regression Landslide Models

LEGORRETA-PAULIN, Gabriel1, BURSIK, Marcus2, LUGO HUBP, José3 and ZAMORANO OROZCO, José Juan3, (1)Department of Natural Resources, Forest Practices Division, Washington State, 1111 Washington St SE, PO Box 47012. Olympia, WA 98504-7012, Olympia, WA 98504-7012, (2)SUNY at Buffalo, 876 Natural Science Complex, Buffalo, NY 14260-3050, (3)Geografía Física, Instituto de Geografía, UNAM, Circuito Exterior, Ciudad Universitaria, 04510 Coyoacán, México, D. F, México, D.F, 04510, Gabriel.Legorreta-Paulin@dnr.wa.gov

It is important to evaluate the influence of pixel resolution on cartographic representations of landslides to assess landslide models for unstable hillslope detection. However, little work has been done on the effect of DEM resolution on landslide modeling. This paper evaluates how pixel size affects the cartographic representation of shallow and deep-seated landslides by using artificial landslides. The artificial landslides are created by using published landslide sizes, erosion filters, and manual modification of contour lines to produce landslide topography. The evaluation of landslide initiation models on this synthetic topography builds our understanding of plausible behavior at different grid pixel sizes with other factors held constant. Two landslide models are used to evaluate the effect of pixel size in the prediction of landslides: SINMAP (Stability Index MAPping) and Multiple Logistic Regression (MLR). The two models are embedded in a GIS system, LOGISNET (an acronym for Multiple Logistic Regression, Geographic Information System, and Neural Network) to facilitate the analysis. The evaluation is conducted at 1m, 5m, 10m, and 30m pixel resolutions. Pixel size is increased systematically to show how changing pixel size modifies geometry and shape of the landslides. The result shows that as the pixel size increases, the landslide loses cartographic representation. The result is a biased model prediction. In tests on real topography in landslide terrain, MLR predictions match existing landslides better than SINMAP predictions, if the MLR model has enough pixels to obtain reliable statistics. SINMAP more consistently produces a similar susceptibility map with different pixel resolutions. In general, MLR over-predicts while SINMAP under-predicts landslides as pixels coarsen.