INFORMATION REQUIREMENTS REGARDING TERRESTRIAL SOURCES AND SINKS OF ATMOSPHERIC CO2 INCLUDE BOTH SCIENCE AND POLICY APPLICATIONS (REFER TO SECTION 2). While the knowledge of spatial and temporal patterns of these carbon exchanges is important, the development of scientific understanding of the processes and the prediction of future behaviour of the sources and sinks also requires the identification and quantitative analysis of the mechanisms responsible.
In the terrestrial environment, carbon is present in three main pools: atmosphere, plants and soil. The primary pathways causing changes in the terrestrial pools are photosynthesis (gain), respiration (loss), burning (loss), and other disturbances or removals (harvest, etc.; loss). The observational challenge is to determine the resulting changes in terrestrial carbon distribution, and, so far, two main approaches have been used for this purpose. One, usually called 'bottom-up', starts with a specific parcel of land and aims to account for the various pathways of carbon exchange between the ecosystem and the atmosphere; the large-scale pattern then emerges after combining the exchanges involving individual land parcels. The other ('top-down') begins with measured changes in atmospheric gas concentrations and attempts to infer the spatial distribution and magnitude of the net exchanges.
The policy community requires information on spatial and temporal patterns of CO2 flux at high resolution over very large areas (Chapter 2). These requirements imply the use of models linked to satellite measurements that are available everywhere. Satellite data can also provide up-to-date information frequently, in relation to the rate of change of the variables of interest. Figure 1 gives an overview of some important variables and the data flow involved in the bottom-up approach. The process models can be developed and tested with local-scale field measurements from inventory data, eddy covariance flux towers, carbon enrichment (FACE) experiments, and long-term ecological monitoring sites. In this bottom-up strategy, local processes are thus scaled-up in space and time using satellite imagery and other spatial data.
Some advantages of such a strategy are:
fluxes estimated from spatial data and imagery using process models can be made available everywhere, all the time;
the flux estimates are produced at the apparent resolution of the input data, which may be quite high;
the estimates are based on mechanistic hypotheses about the processes that control the fluxes (e.g. climate fluctuations, land-use change, nitrogen deposition, etc.);
changes can be attributed to various mechanisms or compartments within the ecosystem.
Figure 1. Typical data flow for bottom-up approach
The primary disadvantage of the bottom-up integration of model estimates is the difficulty of determining the accuracy and reliability of the scaled-up estimates, which is further complicated by the problem of assessing the representativeness of the sites used to calibrate and validate the process models. There is no independent way to evaluate the fluxes computed by the models, except at the small spatial scales of field experiments. The eddy covariance methods can now measure net CO2 fluxes for areas as large as 1 km2 under favourable meteorological conditions and homogeneous, level terrain. These data are extremely valuable for model development and evaluation, but they are very expensive to collect and are presently available for only ~100 locations globally. Furthermore, eddy covariance measurements do not adequately constrain the many components of these fluxes related to the processes represented in the extrapolation models. Some of the most important processes thought to contribute to terrestrial CO2 sinks (e.g. recovery from past disturbance, changes in nutrient cycles due to land management, and climatic trends) are inadequately sampled by the current network of flux towers. The present eddy covariance measurement network must be expanded because of the critical function of these data for scaling up. However, the coverage and accuracy of the measurements will not likely be sufficient for obtaining confidence in the large-scale flux estimates derived through process models and satellite data.
In addition to ecosystem model deficiencies, the calculated fluxes may also be incorrect because of errors in the input data (model parameters and satellite-derived information). Model deficiencies are of at least two types, inadequate representation of the processes considered and the absence of important processes in the model. For example, a carbon source or sink which results from a mechanism that is not represented in the model will be completely undetectable by the bottom-up methods.
An alternative and complementary approach is to analyse the carbon budget of the atmosphere from a mass-balance point of view. Such an analysis is predicated on the availability of atmospheric concentration data, and it can be carried out in several spatial and temporal domains. This approach has been used at the global scale for decades, since C. D. Keeling first began measuring CO2 at Mauna Loa, and is in fact the primary line of evidence that originally suggested the existence of a terrestrial carbon sink. Today, there are nearly 100 flask sampling sites around the world from which air is analysed for several trace gases, including CO2.
Inverse methods have been developed to estimate the spatial and temporal variations in CO2 flux from atmospheric concentration data, and they are now being applied by about a dozen modelling groups world-wide (http://transcom.colostate.edu). These methods aim to deduce surface emissions or sinks responsible for the spatial and temporal variations in concentration by accounting for atmospheric transport using numerical models of winds, convection, and turbulence (Figure 2). At the coarsest spatial scales (global to hemispheric), these methods provide very robust estimates of the spatially integrated flux on time scales of seasons to years. More recent studies (Rayner et al., 1999; Bousquet et al., 2000; Peylin et al., 1999; Kaminski et al., 1999) have estimated monthly CO2 fluxes for as many as 25 regions (at sub-continental spatial scales), including interannual variability. At much finer spatial scales, pollutant emissions have been estimated from local time series data using mesoscale transport models and "back-trajectory" analysis (e.g. Morris et al., 1995; Pryor et al., 1995; Fast and Berkowitz, 1997), but this is only now being attempted for CO2 (Gerbig et al., Appendix 3)
Figure 2. Typical data flow for top-down approach
Some advantages of the top-down methods are:
a robust estimate of spatially-integrated carbon flux over very large areas is produced and is independent of process-based model estimates;
fluxes and their variations can be detected and quantified even if they result from unexpected or poorly understood processes;
some inverse methods allow concurrent estimation of both fluxes and uncertainty in the inferred fluxes;
spatial and temporal patterns can often be interpreted in terms of underlying mechanisms, facilitating further development and refinement of ecosystem process models.
The primary disadvantages of these methods are that (1) they provide no direct information about the mechanisms responsible for the fluxes, and thus have no predictive power; and (2) the current configuration of atmospheric observing stations is so sparse and the stations are generally so far from major landmasses that terrestrial fluxes can only be inferred at extremely coarse spatial resolutions. Although a number of attempts to recover monthly fluxes at sub-continental scales from flask data have now been published (Rayner et al., 1999; Kaminski et al., 1999; Peylin et al., 1999; Bousquet et al., 2000), they disagree dramatically about the spatial structure of the sources and sinks. Atmospheric transport is rapid in the mid-latitude westerlies, with a parcel of air requiring a few weeks to circumnavigate the globe. Inter-hemispheric transport is much slower, with a mixing time in the order of one year. The atmospheric signal is therefore relatively easy to resolve in terms of latitude zones, but the more dynamic longitudinal structure is not well determined by the current observing network. Significant uncertainty in the estimation of regional fluxes by inverse methods arises from errors in the model transport which are difficult to evaluate. Inter-comparison experiments (http://transcom.colostate.edu) have shown that the leading models can to reproduce the available surface data (mostly marine, distant from local terrestrial sources or sinks), but they disagree over the continents and aloft, where there is no data constraint (Law et al., 1996; Denning et al., 1999).
An integrated global carbon observing strategy would include elements of both the top-down and bottom-up approaches because significant synergy can be achieved by applying both types of constraints simultaneously (Figure 3). Such a strategy would seek to maximize the information extracted from the observing network in terms of both distributions of sources and sinks in space and time and the mechanisms responsible for the distributions. In addition to CO2 fluxes, this strategy would include estimation of the uncertainty associated with the fluxes as a critical part of a credible information product. The strategy would also make possible direct testing of quantitative hypotheses about the function of the current carbon cycle, thus facilitating the development of improved process models and of predictive models.
Figure 3. Data flow for dual constraint approach
A 'nested design' with multiple observational components is envisioned that allow meaningful comparisons of flux estimates obtained by independent methods at multiple spatial and temporal scales. Such a system should include a set of ongoing and associated research activities:
Ongoing:
Routine regional atmospheric sampling for multiple trace gases over continental areas from in situ and airborne platforms to establish spatial and temporal patterns.
Routine collection of spatial data and imagery needed to apply process models at large scales and to unsampled locations.
Estimation of local to global daily carbon fluxes from gridded climate, vegetation, soils, land-use, emissions, and other spatial data using process-based models and extrapolation/scaling algorithms.
Estimation of regional to global sources and sinks by mesoscale inverse modelling using atmospheric data to establish integral mass-balance constraints on the operational model/satellite products.
Associated research:
Field experiments (flux towers, FACE rings, LTER sites, etc.) deployed globally across gradients in climate, vegetation, soils, nutrient deposition, disturbance, and land use history to improve our understanding of the processes of carbon exchange.
Evaluation of spatial and temporal variability of ecosystem properties and fluxes in the field, to allow upscaling to patches (1 km2 or less) observed by satellite sensors such as MODIS and similar instruments (Table 4).
Atmospheric observing campaigns and associated local- to mesoscale transport modelling to allow direct estimation of area-mean carbon fluxes and flux uncertainties over field sites for an evaluation of models and scaling algorithms (e.g. using convective boundary layer budgets).
Model development and testing associated with the field research sites, with an attempt to capture the source/sink mechanisms across a range of different conditions.
Estimation of sub-continental scale sources and sinks by global inverse modelling using an augmented set of surface and tropospheric measurements of multiple tracers.
The goals of an integrated observing strategy should include meaningful overlaps between independent methods of flux estimation at each transition in spatial scales, from site to pixel to region to continent to globe. These overlaps can be built through a combination of ongoing/research and observation/modelling tools listed above. An important consideration in optimising such a strategy is to incorporate the scale integration through modelling efforts as well as through data collection, both in terms of ecosystem processes and atmospheric processes.
Ultimately, an integrated global strategy seeks near-real-time diagnosis of carbon sources and sinks at high resolution on both space and time that simultaneously satisfies all the data constraints (in situ, remotely sensed, and atmospheric) at multiple scales. Such an observing system is more than a set of observations. Rather, it constitutes a carbon cycle data assimilation system, analogous to the current observing systems used for operational weather prediction.
Many of the elements of an integrated terrestrial carbon observing strategy outlined above are in place now or already under development. The challenges of this strategy are to ensure the presence of the appropriate overlaps and leverage among the disparate data sets, to ensure that important data gaps are filled and that the necessary modelling work is carried out to link the data sets of various types at different scales. The next sections discuss possibilities for a phased implementation of an integrated observing system.
This report focuses on the terrestrial system, but information on air-sea fluxes of carbon provides a valuable top-down constraint through the global atmospheric mass-balance. A fully integrated global strategy would thus also include estimation of the carbon balance of the oceans at multiple spatial scales from site studies, process modelling, ocean-colour imagery, flux surveys, gridded climate and sea-surface temperature data, tracer studies, and ocean models.