HR: 1340h
AN: B53B-1184 [Abstracts]
TI: Assimilation of AmeriFlux Data in Two Terrestrial Carbon Cycle Models: Are Ecosystem- Scale Parameters Temporally and Spatially Coherent?
AU: * Ricciuto, D M
EM: ricciutodm@ornl.gov
AF: Oak Ridge National Laboratory, Building 1509, PO box 2008, Oak Ridge, TN 37831, United
States
AU: King, A W
EM: kingaw@ornl.gov
AF: Oak Ridge National Laboratory, Building 1509, PO box 2008, Oak Ridge, TN 37831, United
States
AU: Post, W M
EM: postwmiii@ornl.gov
AF: Oak Ridge National Laboratory, Building 1509, PO box 2008, Oak Ridge, TN 37831, United
States
AU: Davis, K J
EM: davis@meteo.psu.edu
AF: The Pennsylvania State University, 512 Walker Building, University Park, PA 16802, United
States
AU: Keller, K
EM: kkeller@geosc.psu.edu
AF: The Pennsylvania State University, 208 Deike Building, University Park, PA 16802, United
States
AB:
The growing volume of eddy covariance observations presents an opportunity to test and calibrate large-scale
terrestrial biosphere models with ecosystem-scale carbon and energy flux observations. Such models inherently
assume that many parameters are coherent across space and time for a given plant functional type. We test
these assumptions by assimilating CO2 and H2O fluxes to optimize parameters at several broadleaf deciduous
forest eddy covariance sites in the Ameriflux network each with multiyear records. Two models are used to
examine if the results are model dependent: Top-down Representation of Flora and Fauna Including Dynamics
(TRIFFID) and the Local Terrestrial Ecosystem Carbon model (LoTEC). A genetic algorithm (global optimization
method) is used to avoid misconvergence to local minima in the objective function. Models are forced with gap-
filled meterological data from the AmeriFlux sites. Observed hourly net ecosystem exchange (NEE), gross
primary productivity (GPP) and latent heat flux are used as data constraints. Optimized parameters are subsets of
model parameters that most strongly control photosynthesis, transpiration, heterotrophic and autotrophic
respiration, and phenology. Initial model carbon pools are also estimated as parameters to avoid long
equilibrium runs. Separate optimizations are performed for each site-year to obtain estimates of the variance for
each model parameter. We then calculate the variance of each model parameter at each site and across all site-
years. The spatial autocorrelation of each parameter is also calculated, but is highly uncertain due to the limited
number of sites. A joint optimization is performed using all site-years to obtain a single set of parameters for
each model to predict carbon fluxes across all site-years. We evaluate the ability of the models for out-of-sample
prediction using cross-validation.
DE: 0428 Carbon cycling (4806)
DE: 0439 Ecosystems, structure and dynamics (4815)
DE: 0466 Modeling
DE: 1631 Land/atmosphere interactions (1218, 1843, 3322)
SC: Biogeosciences [B]
MN: 2007 Fall Meeting