Observations of the carbon cycle are conducted at many scales. Eddy covariance flux measurements and other local studies of terrestrial ecosystems at plot scales or of air-sea gas exchange at moorings help to define basic processes that control carbon dynamics. Repeated visits to sites over decades in forest inventory analysis or sampling of the ocean interior on global surveys help establish the long-term mass balance of land and ocean stores. Changes in concentrations of CO2 and other trace gases have helped establish temporal and gross spatial variability of sources and sinks. Analysis of ice cores and marine sediments have provided a sketch of the huge changes in the carbon cycle in the deep past. These data have provided insight into the functioning of the carbon cycle, but many have also presented fresh puzzles. Sea-surface pCO2 data in the Southern Ocean suggest a sink of nearly twice the magnitude implied by meridional gradients in atmospheric CO2, for example, and forest inventory data support smaller rates of uptake in the northern midlatitudes than have been implied by some atmospheric inversion studies.
Satellite imagery allows process-based models of both land and ocean carbon fluxes that are consistent with in-situ observations at local scales to be extrapolated to regional and global scales. To the extent that models extrapolated with satellite data can be evaluated against independent observations at larger scales, they gain credibility for prediction of the future behavior of the carbon cycle. Trace gas concentrations in the atmosphere have been observed for decades by a set of growing networks of about 100 flask sampling stations and a handful of continuous analyzers. Given a knowledge of atmospheric transport (from yet another set of numerical models), these data can provide a degree of constraint to process-based flux estimates extrapolated using satellite data. Fluxes estimated in this manner are only reliable at very coarse spatial scales because the concentration data are sparse and the model transport is prone to serious errors (Gurney et al, 2001). Unfortunately, the huge gap in spatial scales between local flux footprints (about 1 km2) and reliable fluxes derived from atmospheric or oceanic mass balance (about 107 km2) complicates the evaluation of upscaled models.
Technological developments and a growing realization by both the science community and its sponsors have raised the possibility of more rigorous evaluations of spatially extrapolated process-based models. After decades of comparative neglect, ocean color is being observed at high resolution in time and space, allowing estimation of sea-surface chlorophyll and NPP. Spaceborne lidar and imaging radar instruments may provide spatially dense observations of vegetation canopy structure and terrestrial biomass. Fires and other disturbances can be observed over much of the globe. Hyperspectral instruments provide unprecedented detail about vegetation properties and the partition of terrestrial carbon into loving biomass and dead litter.
In the next few years, observations of atmospheric trace gases by in-situ sampling will be augmented by global mapping of CO2 concentrations by spaceborne instruments. These products will be of much lower quality than laboratory analyses of flask samples, but will increase data volume by many orders of magnitude. Rayner and O’Brien (2001) have shown that even low-precision data at relatively coarse resolution (8 x 10 degrees) can add information to the flask sampling networks through inversion if spatially-coherent biases in concentration and transport can be eliminated. Early global measurements by thermal emission spectroscopy and interferometry (AIRS, IASI) will be weighted primarily to the mid-troposphere where spatial gradients are weak. Later instruments (e.g., SCHIAMACHY) will estimate column concentrations weighted toward the lower troposphere using near-infrared spectrometry of reflected sunlight, but will need sophisticated models to correct for diurnal bias. Eventually, active lidar instruments will likely be flown that can provide better precision and 24-hour coverage, and perhaps even some information about vertical structure.
Major programs of coordinated, multiscale field measurements are being conducted or planned in Europe, Russia, North America, Japan, and Australia. Some of these programs include oceanographic, atmospheric, and terrestrial observations and targeted efforts to evaluate the ability of numerical models to bridge the gaps in scale. Within 10 years, the carbon cycle will likely be observed systematically by coordinated measurements from space, in the atmosphere, and at the surface. Atmospheric concentration variations at what we would call mesoscale will be commonplace. Current inverse models struggle to interpret subtle differences between Atlantic and Pacific sectors, but in the near future we will be interpreting the changes in CO2 of an airmass as a baroclinic wave progresses across Europe. Low-precision, high-resolution satellite CO2 maps will need to be “anchored” by routine in-situ sampling calibrated directly to primary standards, and perhaps by upward-looking solar spectroscopy at many locations.
AN INTEGRATING FRAMEWORK
The carbon cycle information content of the data-rich world envisioned above has the potential to allow more direct testing of process-based hypotheses, but poses an enormous challenge for synthesis and interpretation. Current inverse modeling studies optimize monthly mean concentration fields simulated by global atmospheric transport models over grid cells of more than 105 km2 through depths of hundreds of meters. Data used in these inversions are typically monthly mean concentrations calculated from discrete samples. Although many transport models used in the recent literature are driven by “analyzed winds,” they have typically calculated subgrid-scale mass transports by unresolved processes using their own algorithms which are inconsistent with the “parent” NWP model that produced the analysis. Other examples of deficiencies of some recent inversion studies include (1) lack of interannual variability in some transport models; (2) lack of diurnal or interannual variability in the biosphere models that are used to prescribe terrestrial fluxes; (3) incorrect specification of sub-regional variations in prescribed fluxes due to combustion and vegetation activity. As global atmospheric data products become available, transport errors are likely to become magnified, meaning substantial improvements will be required in the models themselves before we are likely to make much of enhanced data coverage.
Major improvements are within reach for synthesis inversion calculations simply by using highly resolved meteorological analyses produced by the NWP centers, including subgrid-scale mass fluxes produced by the on-line forecast model. This will require careful coordination with one or more operational NWP centers, and will probably require dedicated reanalysis to resolve PBL processes at high time resolution. Ideally, such a reanalysis would cover the period back to 1980 on a 0.5 degree or finer grid, and include parameterized mass fluxes as well as resolved transport. Synthesis inversion using such a model could be performed for the period from 1980-present using the historical flask observation record. Given a complete archive of both resolved and unresolved transport, the generation of the adjoint of a transport model is straightforward. This approach will allow flux estimates to be made at the native resolution of the gridded meteorological analysis, and then aggregated up to coarser scales according to the degree of data constraint available. The wealth of new observations envisioned (regional mass-balances field campaigns, estimates of air-sea fluxes from shipboard sampling, enhanced satellite vegetation imagery, an expanded in-situ network) would allow much tighter constraints on priors in such a calculation than have been used previously. The transport properties of the meteorological analyses could be evaluated in detail using data from campaigns. Systematic regionally-coherent offsets between forward calculations and inverse results could be interpreted in terms of hypotheses about slowly varying ecological processes like forest regrowth and woody encroachment.
A more ambitious goal is to develop formal data assimilation methods that make use of a more comprehensive suite of carbon-relevant observations. These include atmospheric composition measured by flask collection, tall towers with continuous samplers, instrumented aircraft, upward-looking spectrometers, and satellites; surface fluxes measured at instrumented towers by eddy covariance; inventories of biomass and soil carbon; agricultural productivity; buoys, moorings and ships at sea; and satellite imagery of vegetation and oceans. Simple process-based descriptions of photosynthesis, ecosystem respiration, growth, and air-sea gas exchange would be coupled to the atmospheric transport model, and an adjoint of the coupled model developed. A generalized cost function could then be minimized, allowing estimation of key parameters in the carbon process models rather than simply area-averaged surface fluxes as for synthesis inversion. These procedures are directly analogous to data assimilation procedures used in weather forecasting. This approach would optimize parameters in underlying biophysical models that describe the processes responsible for the fluxes. Such parameters might include the temperature sensitivity of soil respiration, the wind-speed dependence of the air-sea gas exchange coefficient, and the photosynthetic capacity of forest canopies. Assimilation into global coupled models will not only provide time-resolved maps of surface carbon exchange, but also lead to progressive improvements in the predictive capability of the process models over time. This kind of work has already begun for hydrological models, and has been demonstrated with simple atmosphere-land biosphere models.
An even more ambitious goal of the program is to facilitate the development of fully coupled variational data assimilation systems that combine meteorological analyses with carbon cycle process models, simultaneously constrained by meteorological as well as carbon data. This is envisioned as an effort that will require full participation by one or more operational NWP centers, because of the huge computational, data handling, and human resources required. It would probably not be desirable to perform the analysis in real time, because some of the key observations (e.g., flask sample analyses) would take weeks to obtain. Such a system would improve upon the carbon-only assimilation system described above by recognizing the coupled nature of the physical, biological, and chemical aspects of the Earth system. Surface temperature and humidity are related to photosynthesis rates over land by canopy physiology, PBL cloud fields are related to transpiration, and so forth. Optimizing a cost function containing both physical and biogeochemical observations would produce better carbon cycle analyses. It would also quite probably produce better weather forecasts (Engelen et al, 2001) and climate models.
The rapid development of observational capabilities for studying the carbon cycle requires a parallel effort to integrate these observations into coupled Earth system models. This development represents the maturation of the field and the recognition of the importance of biogeochemistry to understanding and prediction of physical climate. Just as simulation of climate change benefits from decades of daily experience with analyzing and predicting the weather, so too will coupled modeling of the carbon-climate system benefit from frequent confrontation between prediction and realization.
