HR: 1330h
AN: A52B-0786    [PDF]
TI: Using Virtual Tall Tower [CO$_{2}$] Data in Inverse Models to Reduce Uncertainty in Global and Regional Estimates of Carbon Flux
AU: * Skidmore, J
EM: joanne@atmos.colostate.edu
AF: Department of Atmospheric Science, Colorado State University, Fort Collins, CO 80523 United States
AU: Denning, A
EM: denning@atmos.colostate.edu
AF: Department of Atmospheric Science, Colorado State University, Fort Collins, CO 80523 United States
AU: Gurney, K R
EM: keving@atmos.colostate.edu
AF: Department of Atmospheric Science, Colorado State University, Fort Collins, CO 80523 United States
AU: Davis, K J
EM: davis@essc.psu.edu
AF: Department of Meteorology, Pennsylvannia State University, University Park, PA 16802 United States
AU: Rayner, P J
EM: peter.rayner@csiro.au
AF: CSIRO Atmospheric Research, PMB 1, Aspendale, Victoria, 3195 Australia
AB: Previous studies in optimization of carbon observing networks considered any global grid cell as eligible for selection as a measurement site. In reality, measurement of mean CO$_{2}$ over highly variable terrestrial regions is very impractical and expensive. A global network of eddy covariance flux towers already exists where continuous measurements of CO$_{2}$ are taken, as well as measurements of sensible heat (H), latent heat (LE), and net ecosystem exchange (NEE). If the CO$_{2}$ measurements were calibrated, these surface layer values could be extrapolated to the mixed-layer using similarity theory providing a means to sample the continental mixed-layer for use in global inversions to further constrain the carbon budget. We show that with this method for estimating the mid-day continental boundary layer [CO$_{2}$] from calibrated [CO$_{2}$] and eddy covariance measurements at flux towers, a network of "virtual tall towers" can be readily implemented using existing infrastructure and minimal additional instrumentation. Using the TransCom3 experimental protocol, sites where high-frequency timeseries were saved from the forward runs of 12 transport models were used as eligible sites for possible global networks. The fluxes were optimized to fit the monthly mean of mid-day values by sub-sampling the global fields during the afternoon only, when this method works best. A genetic algorithm was used to determine which existing measurement sites should be grouped together in a network that can most reduce the root mean square uncertainty on estimates of carbon flux. The algorithm evolves the most "fit" population of flux tower sites by comparing and prioritizing them according to their performance in the inversions. A global and North American experiment were performed to determine which ten or five towers should be implemented first as virtual tall towers. In the global selection of ten towers, the algorithm selected four sites in North, Central, and South America in the strong flux areas of the tropics and eastern US. Two towers were selected in southern Europe, even though Europe has a dense network, to constrain the regions of North and South Africa. Two towers were selected in Thailand and two in Japan to constrain Temperate and Boreal Asia, respectively. In the North American experiment, the optimal network selected five sites located in North Carolina, Missouri, Illinois, Tennessee, and Maryland. Overall, the strategies of bracketing the main flux areas and making observations through a gradient of fluxes didn't work well in the TransCom3 inversions. The best virtual tall tower networks emphasized placement of measuring sites in and just downwind of strong fluxes.
UR: http://biocycle.atmos.colostate.edu
DE: 0315 Biosphere/atmosphere interactions
DE: 1610 Atmosphere (0315, 0325)
SC: Atmospheric Sciences [A]
MN: 2003 Fall Meeting