B13F-0687
Harmonization of Multiple Forest Disturbance Data to Create a 1986-2011 Database for the Conterminous United States
Monday, 14 December 2015
Poster Hall (Moscone South)
Christopher E Soulard, USGS California Water Science Center Sacramento, Sacramento, CA, United States, William Acevedo, US Geological Survey, Menlo Park, CA, United States, Zhiqiang Yang, Oregon State University, Corvallis, OR, United States, Warren B. Cohen, US Forest Service Corvallis, Corvallis, OR, United States, Stephen V. Stehman, SUNY College of Environmental Science and Forestry, Syracuse, NY, United States and Janis L Taylor, US Geological Survey, Sioux Falls, SD, United States
Abstract:
A wide range of spatial forest disturbance data exist for the conterminous United States, yet inconsistencies between map products arise because of differing programmatic objectives and methodologies. Researchers on the Land Change Research Project (LCRP) are working to assess spatial agreement, characterize uncertainties, and resolve discrepancies between these national level datasets, in regard to forest disturbance. Disturbance maps from the Global Forest Change (GFC), Landfire Vegetation Disturbance (LVD), National Land Cover Dataset (NLCD), Vegetation Change Tracker (VCT), Web-enabled Landsat Data (WELD), and Monitoring Trends in Burn Severity (MTBS) were harmonized using a pixel-based data fusion process. The harmonization process reconciled forest harvesting, forest fire, and remaining forest disturbance across four intervals (1986-1992, 1992-2001, 2001-2006, and 2006-2011) by relying on convergence of evidence across all datasets available for each interval. Pixels with high agreement across datasets were retained, while moderate-to-low agreement pixels were visually assessed and either manually edited using reference imagery or discarded from the final disturbance map(s). National results show that annual rates of forest harvest and overall fire have increased over the past 25 years. Overall, this study shows that leveraging the best elements of readily-available data improves forest loss monitoring relative to using a single dataset to monitor forest change, particularly by reducing commission errors.