Uses of in situ data and implications
Specific data needs
THE VISION OF TCO IS AN OBSERVING SYSTEM WHICH CAN CONTRIBUTE TO THE INTEGRATED UNDERSTANDING AND HUMAN MANAGEMENT OF THE CARBON CYCLE. TO ACHIEVE THIS SYNTHESIZED INFORMATION SEVERAL TYPES OF MEASUREMENTS ARE NEEDED INCLUDING: THE CONCENTRATION OF ATMOSPHERIC CO2 AND OTHER GASES; SURFACE FLUX OBSERVATIONS AND OTHER IN SITU MEASUREMENTS, AND SATELLITE REMOTE SENSING.
An integrated observing strategy should include: the application of multiple constraints at various spatial scales; the use of process-based research; the application of satellite data to map heterogeneous properties of the surface features, and the averaging properties of the atmosphere to quantify CO2 fluxes over large areas.
Land, ocean, and atmospheric measurements are all system elements that make up such a programme (Figure 1). Based on local-scale measurements of the processes that control CO2 exchanges between the atmosphere and the Earth's surface, this system recognizes the crucial importance of models and other scaling algorithms to extrapolate to representative regional and global scales. The system includes multiple comparisons between process model predictions and large-scale observations. This allows the results of models that predict large-scale fluxes to be disproved.
This quality check is a necessary step in the prediction of future atmospheric CO2 levels. At the global scale, changes in atmospheric CO2 are the benchmark against which all process models must be tested. This constraint can improve quantitative process model estimates only if applied regionally.
Ultimately, an integrated observation strategy should provide a timely diagnosis of carbon sources and sinks at high resolution in both space and time that simultaneously satisfies all the observations/data constraints at multiple scales. Such a system will be more than a set of observations. Observations alone can characterize processes at local scales, and can constrain overall mass balances at the largest scales. Models using data from space-borne sensors must be used to extrapolate local understanding to regional scales.
Spatial and temporal requirements include the need to quantify patterns of terrestrial CO2 flux at high resolution and over large areas. These patterns imply the use of ecosystem process models linked to spatial measurements that are available everywhere, such as from satellites. Satellite data can also provide up-to-date information, frequently in comparison with the rate of change of the variables of interest.
In this 'bottom-up' strategy, local land processes are scaled-up in space and time using satellite imagery and other spatial data. The primary limitation of this approach is the difficulty of conclusively establishing the accuracy and reliability of the scaled-up estimates. In addition, the success of the approach depends on the availability of reliable models that represent all the important processes affecting CO2 exchange with the atmosphere, including the impact of various land use measures. Such models are not yet available for all processes, although inventory-based methods and conversion tables currently provide an acceptable approach (IPCC, 1996).
A complementary method is to analyse the carbon budget of the atmosphere from a mass balance viewpoint ('top-down'). Such an analysis is dependent on the availability of atmospheric concentration data and other inputs, and can be carried out in several spatial and temporal domains. Both approaches have been used in various studies.
Figure 1. Observing the Global Carbon Cycle
Used synergistically, the 'top-down' and 'bottom-up' approaches takes advantage of their strengths to compensate for their respective weaknesses (FAO, 2002b). Such synergy may be achieved through atmospheric inversion methods or through a multiple constraint approach. In the latter, the various data sources are employed to constrain the process parameters in a biosphere model, so that model predictions are consistent with all available observations (Figure 1).
The essence of this approach is to constrain the model parameters to optimal values using inversion theory, and thus to infer the complete space-time distribution of carbon stores and fluxes. In practice, the model predicts the observed variables at locations where measurements (surface- or satellite-based) are available, and then finds the parameter values that minimize the overall difference with the measurements. Although more complex, the multiple-constraint approach offers the possibility of employing multiple types of data with very different spatial, temporal, and process resolutions.
The predictions of a multiple-constraint approach are subject to verification by confronting the model with a wide range of data sources representing various spatial scales. Failure to accommodate all data streams simultaneously with a common parameter set enables finding model errors, thus preserving the scientific integrity of the approach. It should be noted that the availability of diverse observations is essential in preventing incorrect results being obtained which still meet the acceptance criteria.
After combining the bottom-up and top-down techniques, the spatial distribution of carbon sources and sinks can be produced with high spatial and temporal resolutions, and of the best quality possible with the current available observations and understanding of the carbon cycle. The spatial and temporal resolutions will then be constrained by the resolutions of the input satellite (and other) data. This approach is also fully compatible with reporting needs for small areas (such as those that may be required for the Kyoto Protocol), although additional data may also be required in certain cases.
In situ data have several roles to play in the development of TCO products and are relevant at all spatial scales, from site to global. However, such data are far from complete in space or time and there are also many gaps in thematic coverage.
In situ data required in TCO activities include:
1. for the development and validation of carbon flux models (mostly point/site data, Table 1);Areas 1 to 5 are necessary to generate quality TCO products and are consistent with the dual constraint concept adopted by TCO (FAO, 2002a). Data for areas 6 and 7 are needed to provide independent assessment of the final TCO products (product validation). Area 6 is identified separately since different data are used and instead of direct accuracy assessment, the emphasis is on 'confidence building'. These (typically gridded) data products provide the basis for higher confidence in model outputs, especially in regions or countries where few independent data sets are available for comparison, by demonstrating the performance of the models in more data-rich regions.
2. for the definition of model constants/parameters over the spatial domain of interest (point or gridded);
3. as input data (gridded) into models for TCO product generation (Table 1);
4. as inputs (point or gridded) to atmospheric inversion modelling (Table 1);
5. for preparing or validating model input data, especially data products derived from satellite measurements;
6. for qualitative assessment of the outputs, as confidence-building/constraining process;
7. for quantitative validation of TCO products by independent means for determining fluxes.
The main TCO end products are:
Derived or ancillary products
Other mechanisms for reducing long-term bias in net carbon fluxes are also available in TCO. For example the atmospheric inversion (top-down) method brings together the estimated fluxes with accurately observed atmospheric concentration changes. Another option is the use of ecosystem models which mimic the long-term changes in the major pools (e.g. above- and below-ground biomass compartments), and validating these models with the aid of detailed inventory data where available (note, however, that this is a model validation, not an output product validation).
For spatially distributed fluxes ('direct products' listed above), the main basis for validation is flux measurements obtained with eddy correlation techniques (e.g. www.bgc-jena.mpg.de/public/carboeur/). Such measurements are performed in the growing, but currently sparse FLUXNET network (http://daac.ornl.gov/FLUXNET/). Since FLUXNET measurements are typically also used in model development, the coverage of independent data for the validation of outputs is further reduced.
This reduction is compensated somewhat by the fact that comparisons may be made on a daily basis for most current models, thus offering a large volume of measurements from which to choose. Given that the direct validation of net CO2 flux estimates can now only be made against tower flux data, only point validation of the modelled fluxes is possible.
Since average carbon fluxes can be expressed as change in stocks over a time period of interest, data on stocks can in principle be also used to estimate fluxes (IPCC, 1996). This is potentially an attractive option because many inventory data sets are available in various countries. However, the existing data on total biomass (below and above ground) are for the most part insufficient. Information on the growth of woody biomass in forests is frequently missing, especially in tropical regions. Furthermore, the accuracy of carbon stock data is also inadequate for inferring fluxes from areas that are accumulating carbon over a period of interest (i.e. have not been subjected to disturbance).
This relatively low accuracy of the inventory data is a major obstacle in achieving an independent large-scale validation of net fluxes obtained through models (direct products listed above). The reason is that in such undisturbed areas, the net annual carbon uptake (sink/source) is usually far below 1% of the total ecosystem carbon stock and thus exceedingly difficult to quantify over shorter time periods.
This situation is further complicated by different factors responsible for uncertainties in stocks and fluxes. For example, Heath and Smith (2001) found that uncertainties in the soil and tree carbon were the most important causes of uncertainty in United States stock projections, while for flux projections growth and harvests ranked as the highest. Under some conditions (e.g. where the total carbon sink is relatively large and its below-ground component small) it may be possible to do satisfactory inventories of stock changes over longer time intervals at site level, for example 10-20 years. It is important to note that inventory data also captures parts of the regional carbon fluxes (e.g. crop and forest harvests) that may not be measured by flux towers.
In forests, the varying stand age within a forested area complicates estimation of fluxes by presenting a situation of different growth rates. In early growing forests, carbon accumulation can be relatively rapid, but in mature forests a more nearly steady state condition exists until harvest or stand replacement occurs. Variable age structure is thus an additional information requirement for this ecosystem.
For most ecosystems it is not feasible to validate flux estimates from stock measurements, particularly over seasonal to interannual periods, the use of stand biomass or other forest parameters permits direct testing of simulation model dynamics without the confusion caused of manipulating the observed data points to estimate fluxes. In this manner, inventory or biomass data may be used in a 'confidence-building process', for example by comparing spatial trends in model outputs with those in inventory data; by independently assessing input data sets; by evaluating ecosystem model performance over longer periods, or by employing inventory data to constrain model estimates.
As noted in Chapter 5, inventory data are used according to IPCC guidelines (IPCC, 1996) to estimate fluxes by determining change in area multiplied by a typical flux per unit area. This simplified approach is useful only in areas where major changes occur and should not be expected to produce accurate flux estimates, unless the 'typical fluxes' are carefully calibrated for the specific local and climatic conditions encountered.
There are limited options for validating spatially distributed fluxes products obtained through atmospheric inversion models with a spatial resolution of ~107km2 and temporal resolution of months to seasons (product 2 above). This is because both surface and atmospheric observations are used in its generation. The final product is thus (by definition) consistent with all the input data. One way to evaluate the robustness of the estimates is to use conceptually different models in deriving the product; similar results produced from different models provide a basis for higher user confidence.
The uncertainties in cumulative flux products (third category above) are due to the limitations of the validation approaches, the models employed, and the degree to which the models take into account the important environmental processes. Comparisons of model-generated carbon stocks with independent (e.g. inventory-based) estimates would be useful in assessing regional carbon stock and flux statistics produced by models.
National inventories of carbon stocks are currently under-utilized in the estimation of the global carbon budget and its spatial distribution. The major obstacle of a lack of spatial distribution of carbon stocks as inputs to models has not been overcome for most continents. For example, current biomass distributions are not available from in situ or satellite-based techniques, yet forest biomass affects autotrophic respiration and net primary productivity.
Biomass is also strongly related to forest age under given environmental conditions. This is also the case for soil carbon, which is calculated in models, but cannot be verified because of a lack of independent data. Obtaining quality biomass and soil carbon data sets (through new measurements or through compilation of existing data) would contribute significantly to improving the accuracy and confidence in TCO model outputs. These issues are addressed further in Chapter 5.
Conceptually, the main strategies for the use of in situ data in product validation at different scales are:
1. At site level (<1 km2): LAI, biomass, tree diameter, tree basal area, coarse woody debris, soil organic matter, NPP and net ecosystem productivity (NEP) can all be used for testing models of carbon dynamics. LAI (input), NPP (output), and NEP (output) should be validated at flux sites situated in key biomes (such as forests, sites with soils high in carbon content). Understanding of soil processes (e.g. soil carbon decomposition) is necessary for ecosystem model validation. At the site level spatial heterogeneities should also be considered in site-level validation.
2. At the landscape level (100-103 km2): independent estimates of net biome productivity are needed, including spatially variable effects of disturbance and land-use change. Forest biomes are a priority for these assessments but coverage of other biomes (agroecosystems, grassland, and tundra) is also required. Development of robust scaling techniques that combine in situ data, fine- and coarse-scale earth observation data and models is essential.
3. At the national level, there is a need to make better use of spatially distributed national inventories of carbon pools (e.g. forest inventory, agricultural inventory). Development of gridded products and a sound assessment of the errors attached to these products is required, together with the development of methodologies to reduce the errors and to use appropriate comparison methods. This should include an assessment of errors in the inventories themselves (e.g. Heath and Smith, 2001) and the uncertainties of conversions to the carbon content (e.g. Goodale et al., 2001).
4. At regional to global levels, the atmospheric inversion methods require improved data on industrial sources, horizontal transport of carbon (fluxes from marine boundaries, exports by rivers, transport through trade) and other input data.
5. At all scales there is a need for baseline estimates (with errors) of carbon in the main pools, both above and below ground. The errors should be sufficiently small to enable a ~0.1-1.0 % change in carbon pools to be specified in 10, 20, 50, and 100 years. The pool data are required to assess cumulative changes in carbon pools (estimated from the annual TCO flux measurements).
At the TCO synthesis workshop (FAO, 2002b), specific observation requirements were identified (Table 1). The in situ data may be point/site specific or spatial (polygons or gridded) and have various uses (see page 13). Furthermore, the data may be acquired by different organizations at various spatial scales: within countries, at the regional level, or globally (Appendix 3 and 5).
In view of the various origins and uses of in situ data, separate discussions were undertaken for point/site data and spatial data (at national/sub-national and supranational levels). Chapters 4 and 5 of this report present a summary of the relevant findings in these two domains, with further details provided in tables in the Appendix.