PARENT SESSION
Poster Session #66: Landscape Ecology II.
Friday, August 9. Presentation from 8:00 AM to 9:30 AM. Exhibit Hall B & C, TCC
61
From points to landscapes: regression tree analysis and estimates of species richness.
Chong, Geneva*,1,2,3, Kalkhan, Mohammed2,3, Reich, Robin3, Stohlgren, Thomas1,2,3, 1 US Geological Survey- Biological Discipline, Fort Collins, CO2 Natural Resource Ecology Laboratory, Fort Collins, CO3 Colorado State University, Fort Collins, CO
ABSTRACT- We compared two approaches to estimating several vegetation characteristics across a 35,000 ha area of Rocky Mountain National Park, CO, USA. Our sample consisted of 147 1000-m2 plots spread over the 35,000 ha area through random allocation to one of 20 vegetation types identified on 1:15840 color air photos. Because these points were not spatially autocorrelated, kriging could not be used to develop spatial models for estimating variables in areas that were not sampled. Ordinary Least Squares (OLS) regression models had very low explanatory ability (R2 = 0.02 to 0.43). Regression tree analysis models greatly improved our ability to develop spatial models with significantly higher R2 values (0.42 to 0.59). Cross-validation of the regression tree models resulted in estimates very close to the observed values for the number and cover of all plants and exotic plants, and the cover of native plants (differences between mean observed and expected <1). Estimates for the number of native plants were less accurate (difference between mean observed and expected = 21). Regression tree analyses usually used more Landsat TM bands in the models, with the exception of the model for the number of exotic plants, which only used Band 6 (10.4-12.5 mm, thermal infrared- useful for vegetation classification, vegetation stress analysis, soil moisture studies, and capturing unique information on differences in aspect in mountainous areas- Jensen 1996).
KEY WORDS: regression tree analysis, exotic plant species, Rocky Mountain National Park, plant diversity