Simulated sources and sinks of atmospheric CO2, calculated by a threedimensional tracer inversion model of the global carbon cycle, have been subjected to tests to determine their sensitivity to uncertain model specifications and input data. The model, described in a companion article [Piper et al., 2001a], employs regional CO2 source components as boundary conditions to a three-dimensional transport model, TM2, driven by observed winds and a vertical convection scheme based on both observational data and dynamic theory. The model, by an inverse calculation, adjusts the strengths of 7 source components, pertaining to boreal, temperate, and tropical zones, to achieve an optimum fit to measurements of atmospheric CO2 concentration and 13C/12C isotopic ratio from 1981 through 1999 at an array of 9 stations. A standard reference fit was made using initial choices of model parameters and input data for wind, convection, temperature, solar irradiance, remotely sensed plant activity, and related factors, expressed as averages over as many years of recent data as were available for each, but with respect to winds and convection just for the year 1986. This standard case is challenged by means of sensitivity tests in which model specifications and input data have been varied. Although temporal variability in simulated oceanic and terrestrial biospheric sources and sinks deduced in this standard case mainly reflects variability in observations of atmospheric CO2, as much as a fourth is found to be owing to variable wind and convection, and some is likely to be due to temporally varying errors in the specification of CO2 emissions from fossil fuel which lead to nearly compensating departures in inferred biospheric fluxes. Averages of deduced fluxes and the patterns of variability are found to be only slightly sensitive to model specifications, except for the distribution of terrestrial fluxes near the La Jolla sampling site, set a priori, which strongly affects the deduced strength of the northern temperate biospheric sink. As a preferred model solution for further study of relationships to environmental factors and for comparison with other investigations, we adopted the specifications of the standard case in all respects except the distribution of sources and sinks in the vicinity of our observing station at La Jolla, California which is close to an urban area. Both the standard and preferred cases, supported by sensitivity tests, indicate that the terrestrial biosphere has been a sink of atmospheric CO2 in the temperate latitudes of both hemispheres over the past two decades, a source in the tropics and boreal zone. The oceans have been sinks everywhere except in the tropics.

A three-dimensional tracer inversion model is described that couples atmospheric CO2 transport with prescribed and adjustable source/sink components of the global car- bon cycle to predict atmospheric CO2 concentration and 13C/12C isotopic ratio taking account of exchange fluxes of atmospheric CO2 with the terrestrial biosphere and the oceans. Industrial CO2 emissions are prescribed from fuel production data. Transport of CO2 is prescribed by a model, TM2, that employs 9 vertical levels from the earth’s surface to 10 mb, a numerical time step of 4 hours, and a grid spacing of approxi- mately 8° of latitude and 10° of longitude. Horizontal advection is specified from analyzed observations of wind. Vertical advection is consistent with mass conservation of wind within each grid box. Convective mixing and vertical diffusion are determined at each time step from meteorological data. The source/sink components represent various CO2 exchanges, some sources to the atmosphere, others sinks. The study focuses on establishing interannual variability in net terrestrial biospheric and net oceanic fluxes with the atmosphere revealed by variability in atmospheric CO2, taking account of possible stimulation of land plant growth ("CO2 fertilization") and oceanic CO2 uptake, as well as industrial CO2 emissions. Net primary production of land plants (NPP) and heterotrophic respiration are specified to vary only seasonally, on the basis of data averaged from 1982-1990, inclusive. NPP is determined from a vegetative index, NDVI, derived from remotely sensed radiometric data from satellites. Heterotrophic respiration is a function of surface air temperature. Oceanic exchange of CO2 varies seasonally as specified by a coefficient of CO2 gas exchange. Spatial varia- bility of all source/sink components is specified for each 8° x 10° grid box of TM2, a priori, for 5 terrestrial biospheric and 5 oceanic source/sink components, and with respect to emissions of industrial CO2. Spatial variations of terrestrial exchange are made proportional to NPP. Heterotrophic respiration similarly varies by setting its annual average for each grid box equal to NPP. Spatial variations in oceanic CO2 exchange take account of gas exchange dependence on wind speed and temperature and, in the tropics, on a time-invariant spatially variable specification of the partial pressure of CO2 of surface sea water, based on direct observations. Carbon-isotopic fractionation is taken into account for all chemical processes modeled. To produce an optimal fit to observations of atmospheric CO2, the inversion model adjusts the magnitude of 7 additional source/sink components divided with respect to tropical, temperate, and polar geographic zones. There are 4 terrestrial zones, excluding a southern polar zone of negligible importance. There are 3 oceanic zones: one tropical, and one combined temperate/polar zone in each hemisphere. Calculations are carried out in a quasi-stationary mode that repeats a single annual cycle 4 times, and saves the results for the final year. Alternatively, the model has been run in an extended response mode that takes account of a 4-year history of atmospheric CO2 response to a pulse introduced during the first year of this history. Interannual variations in exchange are established by adjusting the model to predict atmospheric CO2 concentration and 13C/12C ratio averaged for annual periods at overlapping 6-month intervals. Net CO2 exchange fluxes, seasonally adjusted, were determined from 1981-1999, inclusive, using atmospheric CO2 data reported by Keeling et al. [2001].