Document: AND-3-40-48

Predictive mapping of forest structure and species composition in western Oregon using Landsat TM imagery and forest inventory data.

HUDAK, A.T.*, J.L.OHMANN, M.MOEUR, M.A.LEFSKY and W.B.COHEN

USDA Forest Service, Corvallis, OR 97331 USA ¹

Abstract:
Landsat Thematic Mapper (TM) imagery is sensitive to variation in forest stand and community structure of high biomass forests in the Pacific Northwest, yet detail is insufficient for mapping structural attributes at the species level. Detailed individual tree data have been compiled for over 2,000 forest inventory plots distributed at 1.7-5.5 km intervals. Thus, statistical methods of linking detailed yet spatially discontinuous plot-level data to spatially continuous TM data would enhance the accuracy of regional vegetation maps. We are comparing two multivariate statistical methods, canonical correlation analysis (CClA) and canonical correspondance analysis (CCsA), to relate stand structure and species composition measured at the plots to TM image layers, as well as to several mapped climatic and topographic variables. To predict structure and composition at unsampled locations, the weights derived from these relationships are used to select whichever plot is nearest the unsampled location in multivariate space, to act as a stand-in. We hypothesize that CCsA will perform better at larger geographic scales, where species typically exhibit a Gaussian distribution along environmental gradients, while CClA will perform better at smaller scales, where species should be distributed more linearly along sections of the same environmental gradient. We have thus far applied the models to a 25,000 km² area contained within one TM scene that spans west-central Oregon from the Cascade crest to the coast. Preliminary results using CClA indicate 56% accuracy in predicting the dominant tree species (19% for codominant) from TM data, compared to 60% accuracy (28% for codominant) from the climatic and topographic variables. We expect that at regional scales, climatic variables may be more important for predictive vegetation mapping while at local or landscape scales, the addition of TM data to the models may be more advantageous.

Keywords: forest inventory, Landsat TM, Pacific Northwest, species composition, stand structure, vegetation mapping

This abstract is being presented at: 10:30 AM in session:
Poster Session #5: Landscape Ecology.