2.11. Inverse Modeling
The success of future policies to limit atmospheric CO2 emissions depends critically on the ability to understand and quantify the budget of CO2, in particular the distributions of CO2 sources and sinks. Over the past decade inverse modeling techniques, which have a long tradition of application to geophysical problems such as seismology, satellite data retrieval, and acoustic tomography, have been applied to the problem of determining the source and sink distribution of atmospheric CO2. Early attempts were aimed at obtaining the interhemispheric gradient of the CO2 sources and sinks [e.g., Tans et al., 1989, 1990a; Enting et al., 1991; Ciais et al., 1995a,b; Bousquet et al., 1996; Law et al., 1996]. With the expansion of the CO2 observational network, and especially after Globalview became available to the community, attempts have been made over recent years to obtain source and sink distributions over continental scales. A number of inverse techniques are currently being tested by various groups, and two common problems have emerged: (1) the calculation can be quite unstable due to the relative sparseness of the observational network, and (2) the transport models likely do not represent transport processes as realistically as desired.
Over the past few years we have developed an inverse technique and studied its performance using the TM2 model [Heimann, 1995]. In general, the inversion uses data from the CO2 observational network and forward simulations of atmospheric CO2 from 14 continental and oceanic regions, each with a standardized source strength. The linear combination of the calculated CO2 pattern from each of the regions that provides the best fit to observations is calculated using the singular value decomposition technique.
The technique was first tested by performing forward calculations of CO2 using source and sink distributions that were as realistic as possible [Ramonet, 1994]. These model calculations then provided a set of "pseudodata" that were used to test whether or not the inversion technique was capable of recovering the source and sink distributions used in the forward calculation. It was found that with a network comprised of as few as 144 regularly-spaced stations, the initial source and sink distributions could be recovered fairly accurately. The errors increased significantly when the "observational network" was reduced to 50 stations. The use of modeled CO2 abundances at the same locations as CMDL observing sites resulted in additional error due to lack of data over large continental regions, especially Africa and South America. Preliminary results show that the error can be significantly reduced by the addition of a surface site in central Brazil.
We are participating in the multi-laboratory Carbon Modeling Consortium that aims to make more rapid progress in deciphering and predicting the global carbon cycle by bringing together atmospheric and oceanic data in a global modeling environment.
Our future work will be focused primarily on two issues: (1) interannual variability of CO2 sources and sinks and (2) design of an optimized atmospheric observational network.