Thu, Aug 18, 2022: 5:00 PM-6:30 PM
ESA Exhibit Hall
Background/Question/Methods: Estimates of forest fragmentation and connectivity typically rely on maps of forest cover where pixels are classified as forest or nonforest based on whether they exceed a set threshold value of tree cover. Another approach of land cover maps is to define forest in terms of a relative confidence of labeling a pixel as forest versus some other land cover. In this study we examined the sensitivity of forest fragmentation estimates to those considerations of mapping confidence. We developed several versions of binary forest/nonforest maps from three Landsat-derived land cover maps: National Land Cover Dataset (NLCD), Land Change Monitoring, Assessment, and Projection (LCMAP), and Landscape Change Monitoring Systems (LCMS). The most conservative version was based solely on LCMAP primary land cover ‘forest’ classification, and the most liberal version was based on LCMS including all ‘tree’ land cover classifications (e.g. Barren & Trees Mix). For each version, we conducted a multi-scale analysis of forest fragmentation based on forest area density for 2019, and summarized the information to examine differences in overall fragmentation estimates and to understand the places/circumstances where the differences were most important. We also looked at estimates of change of forest fragmentation from 2001 to 2019.
Results/Conclusions: Forest cover based on LCMAP primary ‘forest’ land cover alone excluded lowland forests of the Mississippi Alluvial and Southeast Coastal Plains, and Mixed Wood Shield (aka Great Lakes) ecoregions, resulting in greater estimates of forest fragmentation in these regions. Using NLCD ‘forest’ and ‘woody wetland’ classifications to define forest cover, resulted in 15% greater overall forest cover, and decreased forest fragmentation, especially in lowland forests. Forest inventory and analysis (FIA) plot data was used to compare ground verified ‘forested’ locations between land cover datasets. For all FIA ‘forested’ datapoints in the Southeastern U.S., estimated median forest density in 2019 for the surrounding landscape (27 x 27 pixel window) was 57% using LCMAP, 73% using NLCD, and 88% LCMS. Forest density estimates surrounding urban centers also strongly differed between land cover classification approaches, with NLCD classifications of ‘low’ and ‘open space’ development reducing forest cover estimates, resulting in lower forest density compared to using LCMS. These results demonstrate the sensitivity of forest fragmentation estimates to confidence of forest cover mapping in the U.S. and caution special consideration when choosing forest fragmentation estimates to include in resource assessments and wildlife habitat connectivity analyses.
Results/Conclusions: Forest cover based on LCMAP primary ‘forest’ land cover alone excluded lowland forests of the Mississippi Alluvial and Southeast Coastal Plains, and Mixed Wood Shield (aka Great Lakes) ecoregions, resulting in greater estimates of forest fragmentation in these regions. Using NLCD ‘forest’ and ‘woody wetland’ classifications to define forest cover, resulted in 15% greater overall forest cover, and decreased forest fragmentation, especially in lowland forests. Forest inventory and analysis (FIA) plot data was used to compare ground verified ‘forested’ locations between land cover datasets. For all FIA ‘forested’ datapoints in the Southeastern U.S., estimated median forest density in 2019 for the surrounding landscape (27 x 27 pixel window) was 57% using LCMAP, 73% using NLCD, and 88% LCMS. Forest density estimates surrounding urban centers also strongly differed between land cover classification approaches, with NLCD classifications of ‘low’ and ‘open space’ development reducing forest cover estimates, resulting in lower forest density compared to using LCMS. These results demonstrate the sensitivity of forest fragmentation estimates to confidence of forest cover mapping in the U.S. and caution special consideration when choosing forest fragmentation estimates to include in resource assessments and wildlife habitat connectivity analyses.