PARENT SESSION
Contributed Oral Session 64: GIS / Remote Sensing and Landscape Ecology
Tuesday, August 9, 1:30 PM - 5:00 PM, Meeting Room 519 B, Level 5, Palais des congrès de Montréal
Creating spatial probability distributions for longleaf pine ecosystem across east Mississippi, Alabama, the panhandle of Florida, and west Georgia.
Hogland, John*,1, MacKenzie, Mark1, 1 School of Forestry and Wildlife Sciences, Auburn University, Auburn, Alabama
ABSTRACT- Longleaf pine ecosystems have seen a dramatic increase in attention over the past several decades. Recent studies documenting regional trends and conditions of longleaf pine ecosystems have increased awareness of these "critically" endangered ecosystems and provide individuals and organizations with the justification for intensive management, monitoring, and restoration (MMR). Unfortunately, due to the coarse nature of these studies, fine scale longleaf ecosystem MMR can be problematic. To better aid MMR we have created spatially explicit ecosystem data sets (grain size 30 meters) using polytomous logistic regression (PLR), multi-temporal Landsat enhanced thematic mapper plus (ETM+) imagery, and ancillary data sets (QCH,85 = 2747.01, p-value < 0.001; max rescaled R2 = 0.9604). These data sets identify separate probability distributions for hardwood, mixed pine/hardwood, loblolly pine, slash pine, mountain longleaf pine, coastal longleaf pine, and other forested ecosystems across the Southeast. Using a maximum likelihood allocation rule (MLAR) in conjunction with our ecosystem probability distributions and an independent data set, we determined that mean kappa values and overall accuracies of our "most probable" ecosystem map are 0.543 (0.491, 0.596 lower and upper 95% kappa confidence intervals) and 62.2% respectively. While this "most probable" ecosystem map represents one potential output of our data sets, users are not limited to this map alone. Due to the PLR modeling technique users can determine ecosystems locations and identify potential restoration sites by setting probability thresholds for each ecosystem distribution. To calculate the amount of area for a particular ecosystem, users can weigh pixel area by the probability of that particular ecosystem, on a pixel-by-pixel base, and then sum the weighted pixel areas within a predefined area of interest. Furthermore, estimates of model error can be directly applied to each pixel to produce ecosystem probability confidence intervals. Results of this research are supporting the land cover mapping efforts of the Alabama gap analysis project (AL-GAP).
Key words: longleaf pine, polytomous, probabilities, pinus palustris