A data assimilation approach to quantify uncertainty for estimates of biomass stocks and changes in Amazon forests

Michael Keller, USDA Forest Service, mkeller.co2@gmail.com (Presenter)
Paul Duffy, Neptune and Company, Inc., paul.duffy@neptuneinc.org
Douglas Morton, NASA GSFC, douglas.morton@nasa.gov
David Schimel, Jet Propulsion Laboratory, david.schimel@jpl.nasa.gov
Maiza Nara dos Santos, EMBRAPA-CNPM, maizanara@gmail.com
Ekena Rangel Pinagé, EMBRAPA-CNPM, ekenapinage@hotmail.com

Carbon accounting for REDD+ requires knowledge of deforestation, degradation, and associated changes in forest carbon stocks. Degradation is more difficult to detect than deforestation so SilvaCarbon, an US inter-agency effort, has set a priority to better characterize forest degradation effects on carbon loss.

We have initiated efforts to quantify carbon stocks and changes and associated uncertainties in Paragominas, a jurisdiction in the eastern Brazilian Amazon with a high proportion of logged and burned degraded forests where political change has opened the way for REDD+. We build on a long history of research including our extensive studies of logging damage. Our study currently uses recent forest inventories and airborne lidar and will be extended in the future to include Landsat remote sensing that will incorporate a novel use of time series data to estimate the structural properties of degraded forests using bidirectional reflectance information. We identify two objectives for forest carbon accounting at the jurisdictional level:



* Quantify spatially explicit above-ground carbon stocks and the changes in carbon stocks;

* Quantify spatially explicit uncertainties in above-ground carbon stocks and changes in carbon stocks

We are working to meet these objectives by employing innovative data assimilation methods. Our approach employs a hierarchical Bayesian modeling (HBM) framework where the assimilation of information from multiple sources is accomplished using a change of support (COS) technique. The COS problem formulation allows data from several spatial resolutions to be assimilated into an intermediate resolution. This approach provides a mechanism to assimilate information from multiple sources to produce spatially-explicit maps of carbon stocks and changes with corresponding spatially explicit maps of uncertainty. Importantly, this approach also provides a mechanism that can be used to assess the value of information from specific data products. Hence future data collection can be optimized in the context of the reduction of uncertainty. The spatially explicit quantification of uncertainties naturally provides insights into effective sampling designs.

Associated Project(s): 

• Keller (CMS 2013)