In the western United States, aspen (Populus tremuloides Michx.) is a species of particular importance, contributing disproportionally to biodiversity in the region. In some areas of the American West, aspen has been in decline in recent years from Sudden Aspen Decline (SAD). Likely reasons for aspen decline include warming temperatures, pathogens, browsing, and a lack of disturbance, which allows long-lived conifers to slowly overtake aspen and dominate the overstory. Accurate and up-to-date land cover maps are essential for good land management, planning, and decision making. Remotely sensed data such as Landsat have been publicly available since 2008 and are invaluable for mapping land cover because they provide continuous spatial coverage. We used random forest (RF) modelling to integrate spectral and terrain variables and map the percentage of aspen in mountainous regions of Utah. Our goal was to: 1) provide an aspen baseline map for managers of National Forests who are planning landscape-scale aspen restoration projects, and 2) determine whether we could discern changes over time in aspen distribution or dominance, based on yearly Landsat images.
Results/Conclusions
Results of the random forest regression model have identified summer near infrared (NIR) reflectance, summer normalized differenced vegetation index (NDVI) values, autumn NIR, and elevation to be the most important variables in predicting aspen density in Utah forests. Accuracy assessment, using a separate dataset than the training data, indicated that our map is accurate in predicting aspen locations and canopy dominance. We were able to identify locations where aspen has declined over time using yearly Landsat images. However, more work is needed to verify these temporal changes. We plan to use our map for additional analysis in the future, including an assessment of aspen’s effectiveness in reducing wildfire severity.