The response of atmospheric CO sub 2 to changes in land use Page: 15 of 54

                                
14
a lesser extent, this is also true for the discrete-advection model using history A and the
box-diffusion model using history B (Figures 5a,b).
The improvement in coincidence between observation and simulation with the changes
from land-use emission history A to history C is due principally to a change in the amount
of CO2 released from past changes in land use rather than to the timing of that release.
Except for nearly constant or slightly declining CO2 releases from ca. 1920-1945 in histories
B and C compared with the increasing release for that period in history A, the pattern or
timing of CO2 release does not differ appreciably among the alternative emission histories
(Figure 3). Furthermore, the difference between histories B and C and the accompanying
improvement in correspondence between observation and simulation when history C is used
(Figure 5c) are due to the lower estimates of tropical biomass used in producing history
C.
Clearly, understanding of past global carbon cycle dynamics as summarized in the
models and emission histories we use here is incomplete. Somewhere, either singly or in
some combination, errors in the record of past CO2 concentrations, the reconstruction
of past industrial CO2 emissions, the understanding and modeling of ocean-atmosphere
carbon dynamics, or the understanding of terrestrial biospheric sources and sinks (includ-
ing the reconstruction of past CO2 release from changes in land use) are large enough to
produce the level of disagreement between observation and simulation seen here and else-
where {Peng et al., 1983; Enting and Pearman, 1986; Enting and Mans bridge, 1987; Keeling
et al., 1989]. For our present purpose we assume that errors in the record of past CO2
concentrations and the reconstruction of industrial CO2 emissions are negligible; indeed,
these are two of the better-known components of the global carbon cycle, especially during
the Mauna Loa period. Similarly, any error in modeling the net uptake of CO2 by the
ocean that might account for the disagreement between simulation and observation must
exceed the uncertainty represented by differences among the three carbon cycle models we
use because all three models fail to produce the desired agreement; we assume that error