Southern Wind River Landscape Feature Extraction for Risk Assessment

Figure 4. Example of a well pad & road network w/a high-pass filter over ETM imagery. The Southern Wind River Landscape (SWRL) is a three million acre area of Intermountain Semidesert in west central Wyoming. Within the three million acres lie both public and private land holdings. The Nature Conservancy (TNC) has significant conservation interests in this Landscape as well and has created a network of conservation sites that include Red Canyon Ranch, Sweetwater River Preserve, and three conservation easements, over approximately 60,000 acres. The purpose of this project was to compare and contrast two methods of terrestrial classification of feature types to aid the Wyoming Chapter of The Nature Conservancy in developing landscape-level conservation plan. ERDAS Imagine models were applied to the feature extraction process and compared to an ArcView extension, Feature Analyst, produced by Visual Learning Systems, Inc. ERDAS Imagine provides classification models in addition to texture algorithms and convolution filtering that assist in detecting various feature types. ESRI's ArcView and ArcGIS Feature Analyst extension provides are readily accessable and user-friendly method to detect features. With the addition of texture and high-pass filtering, many roads were extracted using standard image processing techniques. The best results were derived using simple edge enhancement and edge detection algorithms. However, accurate results were limited to the more open landscapes with less cover as well as more defined edges. Image processing can be time consuming and require numerous iterations and subsets due to multiple soil types and vegetation cover. Individual features detection such as a paved road or an oil pads using a clustering approach is difficult, but does provide the analyst with a simpler visual interpretation of the image that can assist in identifying these features through regular photo interpretation techniques and heads-up digitizing. Training the Feature Analyst to identify roads took as many, if not more iterations to complete as were required using standard image processing techniques. The value of Feature Analyst is that the operator does not have to completely understand the classification process. The ETM image's coarse spectral resolution proved to be an asset in the more open areas within the study area. However, as was the case for both the IKONOS and ETM data, dense conifers obscured roads and bright soils were confused with roads. Computation time is an issue for both procedures. The time an operators spends on training (digitizing in the case of Feature Analyst) is …