30 Years of Forest Change in the Eastern United States Highlands

Friday, 19 December 2014: 8:15 AM
Michael Joseph Hughes1, Steven Douglas Kaylor1 and Daniel J Hayes2, (1)University of Tennessee, Knoxville, TN, United States, (2)Oak Ridge National Laboratory, Oak Ridge, TN, United States
Forest disturbances drive successional changes, impact management, redistribute carbon among ecosystem pools and the atmosphere, and alter nutrient cycling. Detecting these important ecological dynamics using remotely sensed methods allows for a consistent and comprehensive census of vegetation change over the study area, unlike in traditional plot-based methods. However, in the species-rich and structurally complex forests of the eastern United States, disturbance events caused by low-intensity fires or species-specific insects and disease are often partial, and therefore difficult to detect using satellite-based methods that rely only on total vegetation amount. We developed a set of new algorithms, collectively called VeRDET (Vegetation Regeneration and Disturbance Estimates through Time), which use a novel patch-based approach to incorporate spatial information from Landsat TM to detect disturbance, stable, and regeneration periods in a time-series of imagery. VeRDET uses SPARCS to identify clear-sky probabilities for each pixel, generates a yearly clear-sky composite of those images, calculates a vegetation index using that composite, spatially segments the vegetation index into patches using total variation regularized denoising, and then temporally segments the time-series of each pixel into a piecewise linear function. For each pixel, the slopes of the segments in the piecewise linear function are interpreted as disturbed, stable, or regenerating. We present a map of cumulative forest disturbance and regeneration over the Landsat 5 record in the eastern United States. We detect higher total rates of forest disturbance than previous studies, likely because we include stress and non-mortality declines in vegetation cover. Additionally, ecoregions predict major differences in forest change, with the Piedmont and Southeastern Plains having upwards of 7% of total forest lands affected by disturbance each year on the high end and the Blue Ridge Mountains having a low of 1.9% annual area affected. Across all ecoregions, the predominant trend is for dynamic equilibrium or afforestation. Overall, VeRDET is a promising new tool to answer spatially and temporally explicit questions about vegetation change, and can open new avenues of inquiry into forest monitoring and assessment.