Within the carbon cycle research community two quite distinct modelling
techniques are used to estimate net land-atmosphere CO2
In inverse studies, atmospheric CO2 measurements are used in
conjunction with atmospheric transport modelling. These studies produce
useful estimates of the large-scale geographical distribution of sources
and sinks (Bousquet at al., 1999), but they are unconstrained by what is
already understood about physiological and ecological processes, and
cannot currently provide good estimates at the policy-relevant scale of
countries. In forward modelling, models of land carbon uptake and
release are developed based on an understanding of the relevant
processes (e.g. the response of photosynthesis to CO2, the
response of microbial respiration to climate), and then these models are
integrated forward in time to produce predictions of the temporal and
spatial variability of land-carbon sinks (Cramer et al., 2001).
Estimates of the land carbon balance produced by forward modelling to
the current day from the past are constrained by the understanding of
the system embodied in the model (e.g. conservation of carbon and
nitrogen), but not constrained by any direct observations of the carbon
cycle (e.g. flux measurements, forest inventories, CO2 flask
To date, the inverse and forward modelling approaches have not been
intelligently combined into a carbon data assimilation system, in which
the data used in inverse modelling and the understanding embodied in
forward modelling can be used simultaneously to produce the best
estimates of current sources and sinks. The development of such a data
assimilation for carbon is one of the main goals of CAMELS. The
nowcasting system will adopt data assimilation techniques as used in
numerical weather prediction (NWP). In NWP, data assimilation is used to
produce a best estimate of the initial state of the atmosphere based on
recent observations, and on information contained in the last numerical
forecast. The numerical forecast is consistent with our understanding of
the atmosphere (as embodied in the underlying model), but may differ
from the real weather because of inadequacies in the model or the
initial conditions. By contrast, direct observations provide a more
reliable estimate of the true state of the system but tend to be local.
Data assimilation for NWP acts by nudging the modelled state to get a
good fit to local observations. In this way, the model provides a tool
to interpolate local and often sparse observations to produce a
large-scale picture which is consistent with constraints provided by
both the observations and the model itself. A major innovation from
CAMELS will to develop an equivalent system, i.e. a carbon cycle data
assimilation system (CCDAS).
The combination of mechanistic forward modelling and observational
constraints will provide unique high-resolution maps of the European
land carbon sink, which can be broken down into the relative
contributions arising from land management (as covered under the Kyoto
protocol) and other environmental factors.
Summary of Innovative
||Current State of
distributions of internal TEM parameters.
normally calibrated to produce 'Central Estimate' parameters only.
| Error bars on
estimates of the land carbon sink.
forward modelling (e.g. of carbon uptake using TEMs ) usually
gives a single central estimate only.
| Development of a
carbon data assimilation system combining constraints from
mechanistic forward modelling and large-scale observational
techniques either use pure forward modelling (without observational
constraints) or pure inverse modelling (without physiological
| Separation of the
contemporary land carbon sink into contributions due to land management
and other accidental environmental effects.
models can only produce estimates of the net CO2 sources and
sinks, and not the contributions due to different processes.