National Wildlife Refuge System, Refuge Planning
Land cover and vegetation data are critical to the planning process for National Wildlife Refuges. These data help inform planning and management activities and are included in Comprehensive Conservation Plans that guide refuge actions in the coming years. As part of the planning process, vegetation and land cover data were mapped at high resolution for Guadalupe-Nipomo Dunes National Wildlife Refuge on the central California coast. A combination of open source tools and commercial software was utilized to complete the mapping. WorldView 2 satellite imagery, digital elevation data, and several derived products were included in a large dataset that was analyzed using the RandomForest algorithm found in the R statistical software package. Random forest is a machine learning/classification method that uses decision trees. The output was then overlaid onto image objects that were created through an image segmentation algorithm in Trimble eCognition software. Finally, the output was filtered to remove small patches by applying the majority value for each image object, resulting in a clean and accurate land cover product, which was evaluated using field-collected data in the spring of 2014.
Overview of the vegetation mapping process for Guadalupe-Nipomo Dunes National Wildlife Refuge.