Identifying the causes of landscape change in the North Coast and Cascades Network Parks

Antonova, N; Thompson, C; Wallin, D; Bryson, R; Clary, S; Hayes, A; Oehler, I; Pickard, B; Ponce-McDermott, M; Schaller, N

Final Report for Task Agreement J8W07110003 to PNW CESU Agreement H8W07110001. National Parks contain scenic and dynamic landscapes where natural disturbance processes dominate. Examples of disturbance agents are wildland fires, landslides, avalanches, floods, insect damage and windthrow. Monitoring the location, severity and types of disturbances occurring in the parks, including any variation in those factors associated with climate change, is a priority for park managers. Therefore landscape dynamics has been selected as one of seven core ?Vital Signs? tracked by the NPS North Coast and Cascades Network (NCCN) Inventory and Monitoring Program. In cooperation with academic partners from Oregon State University, NCCN has developed a landscape monitoring method using new remote sensing change detection method called LandTrendr (Thompson et al. in review). This method generates geographic information system (GIS) maps of landscape change from 1986 to the present for areas inside and outside NCCN park boundaries. After the location and extent of disturbances were mapped, a change agent needed to be determined for each disturbance. Following a remote sensing workshop in 2002, a set of eight agents were selected as of interest to the NCCN (Table 1). NCCN developed a Random Forest statistical model to classify disturbances into one of the categories shown in Table 1 based on spectral characteristics of the disturbance and its location on landscape. A tenth agent called "No Visible Change" was added to the model to represent areas that had undergone change that was not of interest to the network. These "uninteresting" changes included annual variations in cloud cover, snow level, phenology, topographic shadow, and soil moisture. The collaborative project between NPS and Western Washington University (WWU) funded by this cooperative agreement was initiated in order to provide student interns the opportunity to conduct lab- and field-based accuracy assessment of these labeled landscape change maps. The objectives of the project were threefold: 1) provide students with an opportunity to gain educational experience with both computer-based remote sensing methods and field work, 2) provide NPS with validation of the statistical agent labeling model using aerial photo interpretation method and field work, 3) determine the viability of using students in any future aerial photo and/or field-based map assessments. We wanted to achieve disturbance labeling accuracies in the 80th percentile that would be similar to those generated by the validation techniques provided by the Random Forest statistical package. The expectation was that validation using aerial photography and spectral characteristics of disturbances would be as accurate as, or more accurate, than validation using field visits. We also hoped that the students would be able to learn to identify disturbance agents correctly with no prior experience with aerial photo interpretation and only using the training provided by park staff.

Identifying the causes of landscape change in the North Coast and Cascades Network Parks

Collections