PS 70-101 - Utilizing NASA Earth Observations to delineate riparian corridors and evaluate invasive species cover in the Verde River Watershed

Friday, August 11, 2017
Exhibit Hall, Oregon Convention Center
Sarah Carroll1, Amandeep Vashisht2, Chanin Tilakamonkul2, Leana Scwartz2, Paul Evangelista3, Brian Woodward3 and Amanda West3, (1)Fort Collins, NASA DEVELOP, Ft Collins, CO, (2)Fort Collins, NASA DEVELOP, (3)Natural Resource Ecology Laboratory, Colorado State University

Riparian corridors in the semiarid Colorado River Basin act as an interface between terrestrial and aquatic systems, play an important role in maintaining biodiversity and wildlife habitat, and contribute to controlling erosion and buffering pollutant and nutrient runoff. However, the proliferation of invasive species such as Tamarisk (Tamarix spp.) within these corridors disrupts biodiversity and essential ecological and hydrogeomorphic processes, including water balance and sediment and nutrient loads. This project utilized terrain data from SRTM, spectral and thermal indices derived from NASA’s Landsat 5, Landsat 7, and Landsat 8 to map the current maximum potential riparian corridor area and riparian vegetation in the Verde River watershed, which feeds major Colorado river tributaries in the lower Colorado River Basin. Potential riparian corridors were mapped for both 2015 and 2010 to enable partners at the Walton Family Foundation to prioritize future ecological restoration areas as well as to evaluate the efficacy of previous management efforts in the Verde watershed.


The predictive performance of the classification models was first evaluated on the basis of statistical metrics and then by visually inspecting the resulting prediction maps compared with NAIP imagery. Both the 2010 and 2015 continuous (multivariate regression) models performed reasonably well reporting an R2 of 0.50 and 0.54, respectively (Table 2). The Root Mean Square Error (RMSE) was slightly higher for the 2010 model than for the 2015 model, however in general both continuous models had less predictive power at low values of vegetation cover (i.e. 20 percent or less). The predictive performance of the 2010 and 2015 binary classification models were evaluated on the basis of the out-of-bag estimate of error and the classification error rate for presence and absence points . The out-of-bag error was just over 20 percent for both the 2010 and the 2015 models. NASA Earth Observation data were integrated successfully in our two-step Random Forest land cover classification model to successfully map percent riparian vegetation cover in the Verde River Watershed.