New approaches to inverse modeling are needed to apply highly resolved atmospheric data to constrain regional fluxes. Atmospheric transport across regional areas is sufficiently rapid that concentration changes will have to be resolved on the order of hours to a few days rather than months or years. Trace gas transport will need to be represented at much higher spatial and temporal resolutions than at present,possibly using "observed" meteorological fields from four-dimensional data assimilation systems,or by incorporating carbon fluxes into models used in operational weather forecasting. Significant improvements in the land-surface parameterizations used in numerical weather prediction would be required.

An objective of the global observational and inverse modeling system should be to provide meaningful integral constraints on spatially extrapolated estimates of carbon fluxes derived by "upscaling"local fluxes using process based models and remote sensing. These observing and modeling programs,described in the next section, would be extremely valuable in the context of top-down estimates of flux derived independently from the global observing program.

Terrestrial Observations, Experiments, and Models

Studies to refine understanding of terrestrial CO2 exchange confront fundamental questions. What are the fluxes of carbon into today's ecosystems? Which systems are taking up how much carbon? What factors influence changes in past and contemporary ecosystem carbon storage (e.g.,CO2 itself, nitrogen deposition,other pollutants, dimate, management practices)? How has the rate of carbon storage changed in the past centuries and decades? What systems and management practices cause net losses or gains of carbon? How will fluxes and storage of carbon in the terrestrial biosphere change with changes in climate and the chemical composition of the atmosphere?

Studies of terrestrial carbon cycling must focus on systematic sampling strategies designed to characterize quantitatively essential processes and to reject or confirm specific hypotheses concerning responses along gradients of principal controlling factors. There are several current hypotheses concerning the particular ecosystems or processes that take up CO2 in response to environmental change. For example, variations in temperature and soil moisture (Dai and Fung 1993), growing season length (Myneni et al. 1997), N availability (Holland et al. 1997), CO2 fertilization (Friedlingstein et al. 1995), and forest regrowth (Turner et al. 1995) have all been suggested to be involved (see,e.g., Goulden et al. 1996, Fan et al. 1998). Although measurements along gradients are a powerful technique for assessing ecosystem responses in systematically different conditions, in practice the factors that determine the changes along the gradient are confounded to

some degree. Thus, great care must be used in interpreting observations and experiments along gradients. In addition, the confounding of control variables, together with vari ability, means that significant replication must be obtained at least at some points along environmental gradients.

The role of land use must be a central subject in any plan for terrestrial carbon research. In the annual ter restrial CO2 budget summarized above,it is evident that a sig. nificant portion of the terrestrial fluxes is related to present and/or past land use. Additional evidence from atmospheric and oceanic measurements suggests that most of the -1.8 Gt C/yr land sink may be occurring in the Northern Hemisphere (Ciais et al.1995a). Some recent analyses suggest that an appreciable fraction of the total terrestrial sink may reside in North America (Fan et al. 1998, Rayner et al. 1998). Inventory information is also accumulating to suggest significant sinks of carbon in North America, although the inventoried sinks are typically smaller than those suggested by the evidence from atmospheric measurements. The vast majority of land in the United States and southern Canada was disturbed in the past and is managed,intensively or extensively for human use. North American carbon sinks, as well as uptake by European or Asian ecosystems,are strongly affected by human activities. Studies of the role of land use history in determining the fluxes are discussed below in this chapter in section "Goal 3: Land Use."

A deliberate sampling and experimental design is required,aimed at characterizing fluxes and processes controlling carbon storage in forests, grasslands, agricultural lands and soils. The design should emphasize not the identification of "typical"sites for a vegetation regime,but the identification of a network of sites within vegetation types that sample the principal axes of variation. These axes of variation would include not only climate and soils (Schimel et al.1997), but also disturbance type and time since disturbance. A high priority is the development of new methods for measuring carbon fluxes belowground.

A principal focus of studies within this network would be systematic observation of carbon exchange fluxes along environmental gradients. This is now possible to an unprecedented degree. The key gradients (e.g., climate, nutrient deposition, forest age,plant functional types,land use) can be defined. Rates of carbon exchange can be measured. Previous efforts to estimate net CO2 exchange have been hindered by pervasive small-scale heterogeneity in terrestrial carbon storage,and by difficulties in assessing changes in belowground carbon storage. Forest inventories represent a critical first step in quantifying storage, but they need to be upgraded to provide better information on carbon,coordinated to help scale results from flux stations and airborne regional measurements,and integrated to provide consistent, continental-scale estimates of net carbon exchange. Eddy flux data are providing high resolution on changes in carbon storage on the scale of

hectares (e.g., Goulden et al. 1996, Goulden et al, 1998). However, further technological developments are needed to bring down costs and improve the accessibility of the technique and the reliability of the method. (Commercial devel opment of the technology now appears to be underway.) Finally, hectare to kilometer-scale-resolution data can be extrapolated to regional (and ultimately global) domains using advanced remote-sensing techniques and verified through expanded atmospheric concentration measurements and models described earlier in this chapter.

A network of flux measurements along well-defined environmental gradients provides several valuable products. Fluxes can be directly extrapolated using area weighting from remote sensing and inventory information. Flux observations can provide information about responses to environmental forcing (such as temperature and soil moisture). Better understanding of the response to envi ronmental forcing can then be used in extrapolations and analysis. Flux measurements allow estimates of carbon sequestration from inventories to be compared to measured carbon uptake. Fluxes can also be extrapolated using models. Estimates of seasonal or interannual variations in fluxes can be compared to changes inferred from atmospheric measurements. Applying this approach to temporal variations is especially important. For example, while interannual variations in local climate and carbon fluxes may suggest hypotheses about large scale regulation, they provide limited insight without direct large-scale mass-balance constraints. Conversely, estimates of the large-scale atmos pheric CO2 seasonal cycle and sources and sinks provide a vital global constraint on models. However, these estimates provide limited information about the modeled mechanisms and sensitivity without greater spatial resolution and precision in estimated fluxes.

The ability to apply eddy-covariance flux measure ments to regions will be limited by knowledge of errors in both temporal and spatial scales. Because the technique accumulates data at high frequency, there is essentially litthe problem in the temporal resolution of the specific measurements. It is critical,however, that the variability of NEP over seasons and years be captured by continuous, high-quality operation of each of the sites. It is in the spatial domain that problems with regional measures of NEP using the eddy-covariance flux method may arise. A region (or biome) can be thought of as a unit of observation from which samples can be drawn to allow the region to be characterized quantitatively with explicit statements of error for NEP in space. This characterization can be achieved if eddy-covariance measurements are replicated within regions,not only across the axis of a particular gradient (as discussed above) but also normal to the gradient axis in order to quantify variance at each gradient level. The number of replications needed cannot be stated a priority as it is itself a research issue, Nevertheless, adequate replication is absolutely essential

to providing information that ultimately can be used to reduce uncertainty in estimates of current and future regional carbon flux and storage. Such replication may depend to a large degree on the development of low-cost, stable, reliable, semiautomated instrument packages that will greatly reduce the manpower and logistical costs associated with the measurements.

The envisioned approach would combine process studies and experiments (designed to increase basic knowl edge and improve predictive models) with flux measurements designed to allow models to be tested. A welldesigned network of study sites would serve as a focal point for many different types of research. Process studies on nutrient interactions,on feedbacks from species diver sity and changes to biogeochemistry, and on climate effects are all needed. Experimental manipulations can help untangle complex mechanisms and test hypotheses for ecosystem responses to conditions outside the current envelope. For example,studies using preindustrial CO2 levels can yield critical information on carbon storage from past changes. Studies manipulating CO2, nitrogen deposition, and climate at sites at a range of times since disturbance are crucial for quantifying interactions among these key global change drivers. It is essential that the research network use common approaches and methods with the highest degree of standardization of methods and instruments as possible. Particular attention must be given to quality assurance in the operation of monitoring equipment and conduct of manipulation experiments. Formal protocols will certainly be required at an early stage. Recognizing that new and better methods and instruments will be developed, the networks need to be able to accommodate innovations so that the innovations can be applied across the entire system,not piecemeal.

The United States and other nations already invest sig nificantly in natural resource inventories for management purposes. These inventories should be designed to provide better information on carbon. Only large-scale opera tional inventories-such as those maintained by govern ment agencies can provide data from the hundreds to thousands of sites needed for direct spatial integration. The inventory data need to be more effectively integrated with other sources of information,including eddy flux and remote sensing. In addition,it is critical to extend the inventory approach to cover the fate of carbon after it is harvested from forests. This carbon includes not only logging residues, but also manufactured products and waste streams.

Finally, eddy-covariance flux time-series are needed to provide closure on carbon budgets at key sites to measure CO2 uptake or loss as a function of the location in the experimental design.

The proposed terrestrial carbon research network poses several significant challenges. For measurements

along gradients to have power in rejecting hypotheses or parameterizing models,large sample sizes are required. Today's networks of flux sites number in the tens of installations. Globally this approach will require significantly more measurement sites,an arrangement that requires significant prior investment in autonomous measurement technology and theory to make the technique simpler, more robust, and less expensive. The program must be sustained for a significant period,because the measurements become valuable only when reasonably long timeseries have been collected,and they will become more valuable over time.

Clearly, both the design of the initial network and its implementation will also require extraordinary care,statistical rigor, and investment in technology. Enormous resources could be expended on process studies, experiments, flux measurements,and inventories without materially reducing uncertainty about either today's CO2 budg et or simulations of future trends. For the nation's scientif ic resources to be efficiently deployed,an initial synthesis and analysis of existing in situ data (including soil and sediment surveys, forest inventories, and observations at existing Long-Term Ecological Research [LTER] and AmeriFlux sites (LTER maintained by the National Science Foundation and AmeriFlux maintained by the Department of Energy, National Aeronautics and Space Administration,and National Oceanic and Atmospheric Administration). remote sensing,and model results is needed to define patterns of variability. Then, the research community must be engaged in the effort to use this synthesis as a basis for site selection in a network designed to understand patterns at large scales.

Once an analysis of existing data and models is done, a critical set of measurements and experiments can be designed to efficiently sample the space identified. Different sets of measurements may be appropriate for dif ferent suites of sites. The design should take advantage of remote observations of land cover and land cover change, which are quite comprehensive for the United States. Similarly, inventory data are already available,and with sys tem and data management upgrades, management data may be made highly useful. With some effort in technology and theory development, the present network of roughly 20 flux measurement sites can be increased by a factor of 2 to 10.

Six other types of studies are proposed to complement the proposed flux measurements at network sites: 1. Experimental manipulations of CO2, temperature, nitrogen,and other key controlling factors. Manipulations are an ongoing line of research that must be enhanced. A number of specific questions about the nature.quantitative importance,and persistence of mechanisms driving the current terrestrial carbon sink can be best addressed through direct experimentation.

Manipulations including ecosystem-scale climate change.nitrogen additions,and elevated or decreased CO2 can provide critical insights on interactions, on responses to conditions in the past or the future,on ecosystem-to-ecosystem variation in responses,and on interactions with other anthropogenic impacts, includ ing harvesting, other land use change,biological invasions,and altered biological diversity.

The next generation of manipulative experiments designed to understand and quantify the current terrestrial sink should emphasize processes at the ecosystem scale, including responses of both biogeochemistry and ecological dynamics. Studies with preindustrial CO2 are critical, but will require new technologies,especially for experiments at the ecosystem scale. Experiments to quantify effects of multiple factors, alone and in combination,are also crucial. Currently, we have little idea of the extent to which the carbon sink in forest regrowth includes signals from increasing CO2, N deposition, or climate change. Similarly, we have no idea of the consequences for carbon storage of vegetation changes, for example,increasing shrub abundance in many of the world's grasslands (Archer et al. 1995). Experiments could be used to probe both the drivers and the conse quences of vegetation changes,especially experiments on ecosystems where the transitions can occur quickly.

The next generation of experiments should also include pilot studies to evaluate deliberate carbon sequestration strategies. Issues concerning the limited spatial and temporal scale of manipulative experiments should receive intensive attention,so that lessons from the experiments can be effectively interpreted and incorporated into global-scale models.

2. Long-term terrestrial observations. Expansion and enhancement of the LTER network will pay huge dividends by defining the ecological and current and historical land use factors that regulate sequestration and release of carbon from major ecosystems. The network expansion is envisioned to provide new sites, roughly equal in number to the roughly 20 currently existing strategically located to examine ecotones, or boundary zones between regions with different vegetations or biomes,likely to play a significant role in regulating global CO2. Enhancements are needed in two dimensions. Quantitative,ecosystem-level work on carbon stores and turnover should become a major component of each site, and LTERS should become key points for large-scale manipulations and for the expanded flux network. By doubling the budget and number of sites in the current network, and adding important new research tasks, there should be a strong synergy between the carbon focus and present ecological and process-oriented goals of the LTERS,

enhancing both.

3. Intensified flux measurements at sites where detailed process studies are coordinated with eddy-covariance measurements. Intensive observa tional studies can be conducted where carbon uptake and many of its controls (N availability, soll moisture, microclimate,light) and mediating variables (Rubisco content of leaves, conductance,stem flow, belowground processes) are measured. In essence,such studies allow natural spatial and temporal variability to per form the experiments that test hypotheses. Hypothesized control processes can be evaluated if a suitable suite of associated measurements (climatic, atmospheric,and biological) is made. These studies share some of the advantages and disadvantages of deliberate manipulations. The major advantage is that the perturbations are "natural,” including all time scales. The disadvantage is that the conditions are not under control of the experimenter. This "natural approach"is that followed by the present AmeriFlux network.

4. Flux scaling studies in which tall towers, boundary layer measurements (Convective boundary layer budgets) and aircraft profiles address the scaling of land surface fluxes to their signatures in the atmosphere. Opportunistic use should be made of tall transmission towers (e.g.,Bakwin et al. 1995), which allow micrometeorological fluxes to be determined over much larger footprints than typical canopy towers. Additionally, tall towers allow the verti. cal profile of the eddy flux to be measured,testing scaling strategies by varying the footprint of the measurement. With appropriate inclusion of meteorological data collection (radar wind profilers or balloon sondes), tall towers also allow the direct measurement of the local forcing of the atmospheric CO2 rectifier effect, which will facilitate larger scale atmospheric modeling. Aircraft observing campaigns should use eddy-covariance sites as anchor points. These programs can be designed to provide flux estimates over a much larger footprint than that of even a tail flux tower, through measurements of continental boundary layer budgets for scales of several tens of kilometers (Chou 1999. Desjardins et al. 1997, Lloyd et al.1996) to transects measured by flux aircraft for scales of tens to hundreds of kilometers (ABLE-2/3 experiments, BOREAS,FIFE2). These data allow a quantitative evaluation of the relationship among intensive flux measure ments, land surface cover, and ancillary process data; and would facilitate the development of scaling strategies by providing spatially extensive snapshots as context.

5. Remote sensing of terrestrial properties. Remote sensing must be a critical component of any plan for terrestrial carbon research. Efforts to characterize

terrestrial carbon cycling should exploit interaction with programs such as the World Climate Research Program's Global Energy and Water Cycle Experiment Continental Scale International Project (GEWEX/GCP). This program has already demonstrated success in integrating remotely sensed and in situ measurements of energy and water fluxes. To the extent possible,carbon research plans should be structured so that they can take advantage of improved satellite data products expected in the near future. These products will include new 30-meter Landsat Thematic Mapperderived land cover data,high-resolution data sets expected in association with the ETA Mesoscale Model, and data products anticipated from the Mission to Planet Earth. Of particular interest is the possible appli cation of advanced algorithms to the upcoming Earth Observing System (EOS) sensors to derive plant canopy functional properties.

6. Integration of observations with model development. Understanding terrestrial processes requires that ongoing observations be linked to the continued development and testing of models. It is extremely difficult to develop terrestrial carbon models that include state-of-the-art process representations for all of the needed processes on multiple temporal and spatial scales. Often the limitations of models serve as signposts in formulating and testing new hypotheses. Examples of current model frontiers are the effects of CO2 on a full suite of plant processes (including alloca tion of carbon to different parts of a plant), dynamic interactions between carbon and nitrogen budgets, hydrologic changes (such as drying or thawing of bore al peat), and vegetation dynamics such as successional changes over long time scales. Models must also be improved to systematically incorporate information about human and natural disturbance of the land sur face. Current models emphasize physiological and bio geochemical processes and largely neglect the carbon storage dynamics induced by cultivation, forest harvest, fire, fire suppression,and other intensive disturbances as well as biological invasions, changes in biological diversity, and other ecological processes

The chief requirement for the progress of modeling. particularly in hypothesis testing—is better integration of models and data. Model development is currently sup ported by a variety of programs,including NOAA Carbon Modeling Consortium (CMC), Terrestrial Ecology and Global Change program (TECO), Vegetation/Ecosystem Modeling and Analysis project (VEMAP),NSF's Methods and Models for Integrated Assessment (MMIA), and numer ous disciplinary programs. Although this diverse range of support encourages innovations,more effort is needed in integration. The assembly of observational data into