B31F-0088:
Characterization of subarctic vegetation using ground based remote sensing methods

Wednesday, 17 December 2014
Daniel Finnell1, AJ Garnello2, Michael W Palace3, Franklin Sullivan3, Christina Herrick3, Samantha Marie Anderson3, Patrick M Crill4 and Ruth K Varner3, (1)Virginia Commonwealth University, Center for Environmental Studies, Richmond, VA, United States, (2)University of Arizona, Tucson, AZ, United States, (3)University of New Hampshire (UNH), Institute for the Study of Earth, Oceans, and Space (EOS), Durham, NH, United States, (4)Stockholm University, Stockholm, Sweden
Abstract:
Stordalen mire is located at 68°21’N and 19°02’E in the Swedish subarctic. Climate monitoring has revealed a warming trend spanning the past 150 years affecting the mires ability to hold stable palsa/hummock mounds. The micro-topography of the landscape has begun to degrade into thaw ponds changing the vegetation cover from ombrothrophic to minerotrophic. Hummocks are ecologically important due to their ability to act as a carbon sinks. Thaw ponds and sphagnum rich transitional zones have been documented as sources of atmospheric CH4. An objective of this project is to determine if a high resolution three band camera (RGB) and a RGNIR camera could detect differences in vegetation over five different site types.

Species composition was collected for 50 plots with ten repetitions for each site type: palsa/hummock, tall shrub, semi-wet, tall graminoid, and wet. Sites were differentiated based on dominating species and features consisting of open water presence, sphagnum spp. cover, graminoid spp. cover, or the presence of dry raised plateaus/mounds. A pole based camera mount was used to collect images at a height of ~2.44m from the ground. The images were cropped in post-processing to fit a one-square meter quadrat. Texture analysis was performed on all images, including entropy, lacunarity, and angular second momentum.

Preliminary results suggested that site type influences the number of species present. The p-values for the ability to predict site type using a t-test range from <0.0001 to 0.0461. A stepwise discriminant analysis on site type vs. texture yielded a 10% misclassification rate. Through the use of a stepwise regression of texture variables, actual vs. predicted percent of vegetation coverage provided R squared values of 0.73, 0.71, 0.67, and 0.89 for C. bigelowii, R. chamaemorus, Sphagnum spp., and open water respectively. These data have provided some support to the notion that texture analyses can be used for classification of mire site types.

Future work will involve scaling up from the 50 plots through the use of data collected from two unmanned aerial systems (UAS), as well as WorldView-2 satellite imagery collected during the years 2012-2014. Identification of methane flux regions will later be analyzed based on vegetation coverage to aid classification of increased emission zones within the mire.