QUANTITATIVE METHODS FOR DETECTING, MAPPING, AND MONITORING LANDSLIDE-SUSCEPTIBLE LANDFORMS AT LANDSCAPE SCALES USING AN AUTOMATED, OBJECT-BASED-IMAGE-ANALYSIS APPROACH

Landslide monitoring studies in steep, forested terrain of the Pacific Northwest (PNW) USA consistently have identified a subset of landforms that appear to be relatively more susceptible than others to slope instability. So far, however, geospatial relationships between landslides and these landforms have not been quantified and few studies have measured the spatio-temporal evolution of landslide-susceptible landform morphologies at the landscape scale. Landform identification and monitoring are especially critical in environments like the rural-urban interface, where humans can be exposed to elevated landslide risks associated with intensive resource-extraction practices. Accurate landform delineation and change detection analyses at landscape scales have been hindered by a lack of high-resolution, chronologic, geospatial datasets and by limited techniques for objectively, consistently, and reproducibly identifying and mapping landforms.

We are developing an integrated set of unique lidar- and spectral-imagery- based tools for monitoring landslide and landform spatial distributions over time, across broad landscapes with diverse terrain characteristics where ground monitoring is not feasible or cost-effective. These tools include: (1) a comprehensive landform classification system designed for PNW terrain; (2) an automated GEOBIA (Geographic Object-Based Image Analysis) landform detection and mapping model capable of objectively segmenting and classifying landforms at multiple spatial and temporal scales based on their geomorphometric attributes; (3) statistical algorithms for correlating spatial distributions of inventoried landslides and mapped landforms; and, (4) computational methods for detecting 3D changes in landform morphologies using successive lidar datasets. We are employing 2000 inventoried landslides mapped after a massive 2007 storm event in SW Washington, together with model-delineated landforms derived from 2006 and 2008 lidar DEM products, to analyze storm-induced changes in landform morphometry and to statistically correlate landslide and landform frequency distributions. These tools show strong promise for improving our understanding of landslide-landform spatial associations as a basis for better informing land-management decisions.