Characterization and scaling of vegetation structure in Amazonia and Cerrado regions using remotely sensed imagery

Michael Palace, ESRC-University of New Hampshire, michael.palace@unh.edu (Presenter)
Bobby Braswell, Applied GeoSolutions LLC, rbraswell@appliedgeosolutions.com
Stephen Hagen, Applied GeoSolutions LLC, shagen@appliedgeosolutions.com
Laerte Ferreria, ESRC-University of New Hampshire, lapig.ufg@gmail.com
Mercedes Bustamante, Universidade de Brasilia, mercedesmcb@gmail.com
Michael Keller, USDA Forest Service, mkeller.co2@gmail.com
Sabrina Miranda, Universidade Estadual de Goias, sabrina_miranda@yahoo.com.br
Julia Shimbo, Universidade de Brasilia, juliazashi@gmail.com
Christina Herrick, ESRC-University of New Hampshire, c.herrick@unh.edu
Franklin Sullivan, ESRC-University of New Hampshire, fsulliva@gmail.com

The tropical vegetation that composes the Amazonia and Cerrado regions in South America is structurally complex, with a range of environmental and historical factors influencing this structure. Comprehensive analysis of forest structural properties across these regions is lacking. Forest structural components include biomass, tree size distribution (diameter of trunks and width of crowns), and areal density of trees, among a multitude of other structural parameters, ranging from the mass of dead wood to species diversity. Accurate quantification of these forest parameters, specifically in tropical regions provides insight into carbon and nutrient cycling, hydrological processes, and historical influences.

We present characterization of tropical vegetation using high resolution remotely sensed optical data, scaled using a maximum entropy modeling approach. Our area of study spans the Cerrado and Amazon regions with focus on areas with field-based data and availability of image data. We developed five different sets of automated techniques for analyzing high resolution optical imagery. We then developed three machine learning regressions (boosted tree) to predict plot level biomass, basal area, and areal density. Using MODIS MCD44A4 (ndair BRDF-adjusted reflectance) monthly averages of derived indices and other geospatial datasets we developed probability maps of binned forest structural estimates. We discuss the combination of techniques that prove most appropriate in predictability of these vegetation characteristics.

Presentation Type:  Poster

Session:  General Contributions   (Tue 4:35 PM)

Associated Project(s): 

• Keller, Michael: A Historical Reconstruction of Vegetation Change and a Carbon Budget for the Brazilian Cerrado Using Multiple Satellite Sensors and Historical Aerial Photography ...details
• Palace, Michael: Scaling Forest Biometric Properties Derived from High Resolution Imagery to the Amazon Basin using Moderate Resolution Spectral Reflectance Data ...details

Poster Location ID: 212