PARENT SESSION
Oral Session #16: Landscape Ecology: Fragmentation, ecotones, and models.
Presiding: X. Wu
Monday, August 5. 1:00 PM to 3:45 PM. Gila Meeting Room, TCC.

Interpolating sparse data over complex landscapes: a common problem for ecologists.

Reich, Robin^*,1, Lundquist, John^1,2, Bravo, Vanessa¹, ¹ Colorado State University, Fort Collins, CO² USDA Forest Service- Rocky Mountain Research Station, Fort Collins, CO

ABSTRACT- Ecologists often wish to make landscape-scale inferences with extremely sparse data at very fine scales. We describe a particularly difficult, yet typical sampling dilemma involving 151 30m x 30m plots in about 530,000 ha, a sampling intensity of only <0.000017%. Yet, various models using the independent variables of Landsat TM bands, forest class, elevation, slope, and aspect explain 69% and 67% of the variability in small (<1.2cm) and large (>1.2cm) woody fuels throughout the Black Hills National Forest in South Dakota. We show how the statistical properties of the models include trend surface analysis and binary regression trees that could be used to model virtually any natural resource from forest fuel loadings to the diversity of migratory birds. Cross validation techniques and evaluations of estimation uncertainty can be used to improve the utility of spatially explicit model results. The combination of purposive field sampling, nested plots, and predictive spatial models has broad application for ecologists and land managers.

KEY WORDS: woody fuels, regression tree analysis, natural resource modeling, Black Hills National Forest