A comparison of landslide inventories produced by manual and automated methods on timberlands in the Pacific Northwest
Abstract
Landslides can threaten the potential value of total production on timberlands either by destroying a present crop or by inhibiting future growth. To mitigate or prevent such loss of value due to landslides, it is essential to identify past and existing mass movement events. This is accomplished through the creation and maintenance of a landslide inventory map. Through this project, I partnered with a timber company to create a landslide inventory of their existing properties in Washington and Oregon. I used two methods to create the inventory and compared the results. This inventory will facilitate and guide the decision making process when considering properties for harvest, sale, or development.
Traditional methods for identifying landslides area all conducted manually. These range from field investigations to interpretation of aerial imagery and elevation derived maps. Traditional methods suffer from a lack of a standardized method and because they are inherently subjective. Newer landslide identification methods can be performed automatically. These are typically based on roughness calculations or classifications of spectral imagery. Automated methods continue to be developed and still need refinement before being considered as a replacement for manual mapping.
Using high resolution LiDAR imagery, I created a manually identified landslide inventory through visual interpretation of a bare earth surface. I used the same LiDAR imagery and a roughness calculation based on a continuous wavelet transform method to create an automatically identified landslide inventory. The results of these two methods were compared based on total areal coverage, spatial agreement, and overall geographic patterns.
I identified and delineated a total of 470 individual landslides through manual review of all properties. Landslide terrain covered approximately 2.4 percent of the total area evaluated. The automated review did not identify individual landslides, but grouped any adjacent slides together, preventing any evaluation of individual events. In the results of the automated review, slide terrain accounted for 11.5 percent of the total area evaluated. The degree of matching between the two methods equaled 11, 17, 20, and 27 percent in the four subsets of the study area. Most of the mismatch is accounted for areas that were identified by the automated method but not the manual method.
Landslide inventories are inherently incomplete due to the dynamic nature of surface morphology and the potential for unidentified slides. The two methods used in this evaluation present two unique interpretations of the landslides on the partner organizationís properties. Further review, both manual and automated, may result in a higher degree of agreement between the methods. Currently, the manual inventory is a reliable identifier of existing slides. The automated inventory worked poorly for identifying individual slides but could be used to guide further investigations into areas that have a higher potential for slides.