Historical Reconstruction of Vegetation Change and a Carbon Budget for the Brazilian Cerrado Using Multiple Satellite Sensors and Historical Aerial Photography
Michael M. Keller USDA Forest Service (US-PI)
The Cerrado region of Brazil is covering 2 million square kilometers accounts for 9% of the global area of tropical savannas. The Cerrado has suffered a high rate of land use conversion, although it has been neglected in regional carbon budgets. Preliminary calculations indicate that the Cerrado could account for an annual release of carbon (~0.1 Pg per year) to the atmosphere 50% as large as the annual net deforestation flux from the Brazilian Amazon. We propose to estimate the vegetation carbon of the Cerrado for two time periods: prior to large scale conversion (~1964) and the present (2000 to 2008). The 1964 estimates will depend upon a random sample of aerial photography combined with a biophysical stratification of the Cerrado produced with a regression tree. The quantification of biomass from aerial photographs or high resolution images will depend upon measured biophysical properties in the images (e.g. tree stem frequency and canopy size frequency distributions) compared to ground based biomass distributions. Biometric data for both above- and below-ground biomass will be gathered from publications, databases, and limited field work. For 2000 to 2008, we will estimate total biomass carbon and inter-annual changes in Cerrado biomass carbon stocks using moderate resolution imagery (MODIS + MISR) in an artificial neural network model trained with biophysical properties derived from the analysis of high resolution satellite imagery (IKONOS, Quickbird). Current (2000 to 2008) biomass distributions will be checked against an independent estimate derived from calibrated GLAS lidar data. We will calculate changes in the Cerrado carbon content between 1964 and the present. Uncertainties in both biometry and image analyses will be propagated through all biomass carbon estimates. Explicit uncertainty estimates for all portions of the biomass carbon estimates will be calculated use k-fold cross validation and more traditional in- sample statistics.
Michael William Palace University of New Hampshire (US-PI)