We have developed a
carbon flux inversion method for using a mesoscale meteorological model
(CSU-RAMS) within a Maximum Likelihood Ensemble Filter (MLEF, Zupanski 2005;
Zupanski and Zupanski 2005). The MLEF is a variant of the Ensemble Kalman
Filter, and is used to optimize model state variables and parameters based on
continuous observations of CO2 mixing ratio. The method does not
require the development of a model adjoint, but rather relies on transformation
of variables to efficiently obtain estimates of fluxes with uncertainties and
dynamical model error from an ensemble of forward model simulations. We
demonstrate this method using a mesoscale simulation of weather, transport, and
the surface carbon budget over the continental USA during the summer. The
estimation procedure decomposes the total surface flux into photosynthesis and
respiration (which are assumed to be modeled correctly to first order), plus an
unknown but time-invariant fractional error in each. These residuals are estimated for each model
grid cell over a moving window in time, allowing atmospheric observations to be
integrated over sufficient time to obtain constraint. Model error can also be
estimated by this procedure, and the method can be extended to larger domains
and longer integrations.
Author: A. S. Denning, Dusanka Zupanski, Marek Uliasz, et al (denning at atmos dot colostate dot edu)
Filesize: 32.10 Kb