Integrating phylogenetic and hyperspectral information to map forest communities from space.
J. Antonio Guzmán Q., University of Minnesota, guzman@umn.edu (Presenter)
Jonathan Knott, US Forest Service, jonathan.knott@usda.gov
Jesús N. Pinto-Ledezma, University of Minnesota, jpintole@umn.edu
Philip Townsend, University of Wisconsin, ptownsend@wisc.edu
Jeannine Cavender-Bares, University of Minnesota, cavender@umn.edu
Spectroscopy imagery from spaceborne sensors has long been recognized as a key tool for the future mapping of tree species and forest communities. However, the feasibility of mapping forest communities at large spatial extents from spaceborne data is not yet clear. Unlike species mapping using airborne spectroscopy, spaceborne spectroscopy faces ecological and technological challenges: large numbers of species at broad scales and imagery with fewer bands, coarse pixels (e.g., 30m), and potential atmospheric effects. Here we propose a method that integrates phylogenetics with spaceborne spectroscopy to predict forest community composition at the pixel level. We hypothesize that forest communities composed of closely related species are more likely to be spectrally similar than communities composed of phylogenetically distant species. As such, integrating phylogenetic information with spaceborne spectroscopy can help assess tree communities. We used phylogenetic and forest inventory information collected by the US Forest Service (Forest Inventory and Analysis Program) to ordinate forest communities in three dimensions. This ordination was performed using Non-metric Multi-dimensional Scaling based on phylogenetic-abundance matrices of beta diversity. Coordinates of the forest inventories were used to extract pixels captured by the DLR Earth Sensing Imaging Spectrometer (DESIS). The spectroscopic information extracted was then used to predict the community ordination by performing spatial-iterative Partial Least Square Regressions. Our results revealed that bands between 650 and 800 nm appear informative in predicting the community ordination. Predictions showed congruent performance (R2 ~ 0.55) that appears to decay when increasing the number of axes. We applied the resulting models to DESIS scenes to showcase maps of forest communities' composition. Our study demonstrated a novel method for mapping forest composition using spaceborne spectroscopic data that will be useful for the forthcoming Surface Biology and Geology mission.
Poster: Poster_Guzmn_Q_2-45_136_35.pdf
Associated Project(s):
Poster Location ID: 2-45
Presentation Type: Poster
Session: Poster Session 2
Session Date: Wed (May 10) 5:15-7:15 PM
CCE Program: BDEC