The validation and critical assessment of comprehensive global biogeochemical models represents a crucial step in their development. While validation may proceed for certain individual components on the local scale, there is a clear need to assess overall model performance on the regional and global scale. For example, it has been shown recently that the spatio-temporal distribution of atmospheric CO2 and its isotopic forms provide powerful constraints on the location and magnitude of surface sources and sinks. During the next decade a wealth of new atmospheric and oceanic data will become available, both from the observational components of IGAC and JGOFS (as well as from WCRP), and from remote sensing platforms, which will enhance the power of this approach. In order to assess systematically this potential, a strategy is needed to link component models into a comprehnsive, global carbon model, exploiting detailed, contemporary, spatio-temporal patterns. This modelling framework should ensure that comparable simulations are performed with simliar iniital conditions, meteorological and/or climatic driving variables, and use other common data sets.
The GAIM project "The Coupled Carbon Model Linkage Project" is directed at the intercomparison and evaluation of state-of-the-art high resolution models of the global carbon cycle. Because of the relatively loose coupling of the terrestrial and oceanic carbon systems, which are essentially linked only through the globally averaged atmospheric CO2 concentration, it is advantageous in a first step to assess models of the oceanic and terrestrial biospheric components separately. To this end, the ocean carbon models are being intercompared in a special project within GAIM (OCMIP), while the present project is directed primarily at the model intercomparison and evaluation of terrestrial biogeochemical models (TBMs) that describe the exchanges of carbon between the land surfaces and the atmosphere.
Six different TBMs took part in a preliminary model intercomparison (See "Personnel," below). Each model describes the cycling of carbon on a global grid with a horizontal resolution of 0.5deg by 0.5deg as functions of environmental parameters (e.g. temperature, precipitation, insolation, soil, and vegetation). Internally, however, the models describe the various transformation processes (e.g. phenology, net primary production, litter production, soil depletion by heterotrophic organisms etc.) using very different approaches. For example, the SDBM model [Knorr and Heimann, 1995] includes remotely sensed data as driver data in addition to climatic variables, and hence represents a diagnostic model as compared to the other, fully prognostic TBMs.
In order to conduct the model intercomparison experiments most effectively, it is essential to define a specific simulation experiment protocol which describes the initialization, forcing boundary values, target model output variables and observational data against which the simulation results are to be compared. In the first phase of the project, the adopted experiment protocol includes simulation of the seasonal cycle and of longer-term simulations using climatic and other factors (e.g. increasing atmospheric CO2 concentration, changes in land-use) as prescribed boundary forcing variables.
Simulation runs will now be performed based on the following two experiments:
Seasonal cycle of atmospheric CO2 concentration
The seasonal cycle of the atmospheric CO2 concentration as recorded at the various networks of monitoring stations documents the large-scale, net CO2 exchange fluxes between the terrestrial biosphere and the atmosphere. The extent to which current TBMs are able to reproduce these features will be explored in this part of the project. In the first stage, each TBM will be run to equilibrium using a constant atmospheric CO2 concentration of 350ppm, a prescribed potential vegetation distribution (no prescribed land-use patterns) and a common climatology [Leemans and Cramer, 1991; Cramer and Leemans, 1995, unpubl.). The resulting seasonal fields of surface fluxes (net primary productivity and heterotrophic respiration) will be then input into an atmospheric transport model (TM2). The resulting spatio-temporal atmospheric CO2 concentration calculated by the transport model will then be compared to the observations from the NOAA monitoring station network. Since the ocean and the fossil fuel CO2 sources also contribute to the atmospheric seasonal cycle of CO2, estimates of these source fields will also be included in the simulation.
Clearly, not all of the models are able to faithfully reproduce the observations at all stations. The extent to which any mismatch represents a significant deviation and hence a model defect will be investigated in a series of sensitivity experiments. Two major features are apparent from preliminary analyses [Heimann et al., 1996] and appear robust: (1) In the northern hemisphere and in particular in the Arctic regions most TBMs tend to underestimate the full amplitude of the seasonal cycle. This indicates an overly weak seasonality of the net exchange of CO2 in higher latitudes. (2) In tropical regions, and in particular at the southern tropical station at Ascencion island, most prognostic models tend to overpredict the seasonality. It is very likely that this behavior reflects an overly strong seasonality of the TBMs, possibly related to too strong a sensitivity to the hydrologic cycle in the tropics. The results for the deep southern hemisphere (e.g. South Pole) are difficult to interpret since here the signals are the result of several factors of comparable importance: northern and southern hemisphere biosphere, oceanic contribution, and seasonal transport between the two hemispheres. In the planned study, additional information including variations of isotopic ratios and other tracers will be used to discriminate among the different factors.
Response to changes in climate variables (temperature and precipitation)
The global carbon cycle exhibits considerable variability caused by climate fluctuations, in addition to anthropogenic perturbations. The extent to which the present carbon cycle models are able to reproduce these is explored in this part of the proposed project. In order to isolate the contributions of temperature and precipitation explicitly, simulation runs will be performed with each model with variations of temperature and precipitation, respectively. In each case the TBM will be run to equilibrium under a prescribed 19th century climatology. Then the models will be run from this initial state through 1993 in which observed anomalies of temperature fluctuations [Jones et al., 1986] (Jones, pers. comm.), and precipitation [Hulme, 1994; Hulme and Jones, 1993] are superimposed on the standard climatology. The generated patterns of anomalous CO2 fluxes will then be integrated over latitudinal belts over the entire globe and compared to anomalous fluxes derived from atmospheric deconvolution studies [Keeling et al., 1995] over the time period of the Mauna Loa record (1959-1993).
A preliminary analysis demonstrates that the two process-based TBMs (FBM and SILVAN) are indeed able to reproduce a substantial fraction of the observed variability on the ENSO timescale (2-5 years). A third model (HRBM) which is based on empirical regression relationships between NPP, heterotrophic respiration and climate parameters, exhibits a substantially smaller sensitivity to the temperature fluctuations. Preliminary TBM simulations failed to reproduce recorded variations on decadal timescales, and did not realistically simulate the reduced atmospheric CO2 growth rate observed in the early 1990's, possibly associated with climate anomalies induced by the Pinatubo volcano eruption. We plan to further investigate this behavior in the proposed project. If the results based on additional models and parameters with common input data sets corroborates the preliminary result, it will indicate that the temperature sensitivity of the mechanistic TBMs is adequately modelled in the tropical regions which are primarily subject to the ENSO perturbations, but there is too weak a response in higher latitudes where the cooling in the early 1990's was most prominent. These anomalies may be partially explained by the fact that the only factor investigated in the preliminary analysis was air temperature. Changes in water availability, insolation, and air sea CO2 fluxes probably also play a significant role in determining the climate driven internal fluctuations of the carbon cycle. The sensitivity of the models to each of these parameters will be evaluated.
In a preliminary analysis, the following six different TBMs took part in a preliminary model intercomparison:
FBM: University of Frankfurt (Gundolf Kohlmaier)
HRBM: University of Giessen (Gerd Esser)
TEM: UNH and MBL-Woods Hole (B. Moore, A. Schloss, J. Melillo, &D. Kicklighter)
BIOME2: University of Lund (Colin Prentice, Alex Haxeltine)
SILVAN: MPI für Meteorologie, Hamburg (Joerg Kaduk, Martin Heimann)
SDBM: MPI für Meteorologie, Hamburg (Wolfgang Knorr, Martin Heimann)