5. PRESENT STATUS OF OBSERVATIONS

Atmospheric and Meteorological Observations

Atmospheric Gas Concentrations

PRESENT STATUS:

TWO MAJOR OBSERVATION NETWORKS EXIST AT PRESENT, OPERATED BY THE US NATIONAL OCEANIC AND ATMOSPHERIC ADMINISTRATION (NOAA) AND THE AUSTRALIAN COMMONWEALTH SCIENTIFIC AND INDUSTRIAL RESEARCH ORGANIZATION'S (CSIRO) ATMOSPHERIC RESEARCH. They involve high-precision, continuously operated baseline stations for measuring the concentrations of over 100 atmospheric constituents in the marine boundary-layer air, together with analyses of flask samples routinely gathered at scores of locations world-wide. The sampling interval is typically about two weeks. Data from these networks and other sites are compiled under the GLOBALVIEW-CO₂ project (http://www.cmdl.noaa.gov/ccgg/globalview/co2/default.html).

Gaps and proposed solutions

Significant issues in extending the gas concentration network for the purposes of terrestrial carbon observation include the following:

(a) Range of gases:

While the primary emphasis is on CO₂, several other gases have significance because they may (i) participate directly in land-air carbon fluxes (CH₄, CO, NMVOCs); (ii) act as tracers of anthropogenic emissions which alter CO₂ concentrations by non-terrestrial processes (CO, NMVOCs); (iii) provide tracers of biomass burning, a part of the terrestrial carbon cycle requiring explicit identification (CO, NO_x, CH₄, NMVOCs); or (iv) constitute significant greenhouse gases in their own right (CH₄, N₂O). Other significant constituents include the ¹³C and ¹⁸O isotopes of CO₂, because they contain information on the relative magnitude of air-sea gas exchange and terrestrial carbon exchange, ecophysiological properties such as water use efficiency, C₃/C₄ ratios in plant communities, and ratios of soil evaporation to transpiration. The priority order among these gases for measurements is situation-dependent but is likely to be (1) CO₂; (2) CH₄, CO, N₂O; and (3) NMVOCs and isotopic constituents. This order will change in response to circumstances.

(b) Intercalibration issues:

Intercalibration among different measurement networks is a serious problem and is currently no better than around 0.2 ppm for CO₂. Significantly better accuracy, to 0.1 ppm or less, is required for many atmospheric inversion methods. A global inter-comparison programme called GLOBALHUBS, run from CSIRO Atmospheric Research in Australia addresses this problem. Based on the current status of observation programmes and inversion methods, the following accuracy targets appear realistic: 0.2 ppm for CO₂ flask and continuous data, 0.05 ppm for ¹³C, and 0.1 ppm for ¹⁸O. While less accurate data would also be useful because of the spatial and temporal dynamics of the land-atmosphere interactions, the higher accuracies will be essential to discern longer-term trends and their spatial characteristics at the regional or smaller scales.

(c) Site locations and sampling strategies:

For atmospheric concentration measurements obtained for terrestrial carbon estimation, site selection criteria and sampling strategies should be different from those which apply in the present global networks and are based mainly on sampling in the marine boundary layer. There are two primary reasons. First, the terrestrial atmospheric boundary layer has strong diurnal variability because daytime CO₂ draw-down by photosynthesis is associated with strong, deep convective mixing whereas nocturnal CO₂ build-up from respiration is associated with the formation of shallow stable boundary layer. These differences lead to a highly asymmetric CO₂ signal with time through the full day-night cycle (the "rectifier effect"; Denning et al., 1995, 1996, and 1999). Second, terrestrial ecosystems exhibit high horizontal heterogeneity in trace gas exchange. These factors have several implications for the sampling strategies:

• Continuous observations, at least for CO₂, are crucial at a 'reasonable' number of sites. Where these are supplemented by flask sampling, mid-afternoon samples are most appropriate because boundary layer development and mixing is usually greatest at this time.

• As in the present global networks, accurate long records at fixed locations will be crucial for detecting and interpreting long-term variations and trends in terrestrial biospheric functioning.

• Aircraft profiles of CO₂ (and where possible, other trace gas) concentrations through and above the atmospheric boundary layer will be needed, at least at some sites, for two reasons. First, such data are useful for extending near-surface observations to represent the entire boundary layer. Campaign-style studies largely accommodate this aspect. Second, higher-altitude (upper-air) tropospheric profile data is needed, in addition to surface and boundary-layer data, to improve the constraints on top-down atmospheric inversions. In contrast to the campaign studies, the upper air observations need to be ongoing.

• Uncertainty analyses of present atmospheric inversions suggest that the most critical locations for additional terrestrial observations of atmospheric gas concentrations are in continental locations, especially in the tropics (Rayner et al., 1996; Bousquet et al. 2000; Gloor et al., 2000). The placement of stations is complicated by the need to obtain "regionally representative" samples with minimal local influence, which proves difficult for most continental locations. For example, the siting of observations near continental coastlines is attractive to provide access to both continental and oceanic air (depending on wind direction) from a single station, but the influences of terrain variability and associated mesoscale circulation (such as sea breezes) complicate the measurements and must be carefully addressed at such sites. Gloor et al. (2000) estimated that adding 12 routine vertical profiling sites would reduce the mean error in regional fluxes by a factor of five, to about 0.2 GtC yr^-1 for 17 regions.

• There is a major need for an objective, model-based analysis of both measurement locations and temporal sampling requirements. This also extends beyond atmospheric sampling to other components of the observation system.

Eddy Correlation Flux Towers

Present status:

Investigators can now apply the eddy covariance technique to acquire nearly continuous measurements of carbon exchange between the atmosphere and biosphere.

Regional collections of eddy covariance flux towers were formalized into the EUROFLUX (Europe) and AmeriFlux (North and South Americas) networks in 1996 along with MEDEFLU (Mediterranean region) started in 1998 and followed by AsiaFlux and OzNet (Australia) in 2000. A variety of organizations within each country typically fund the towers; for example, the Department of Energy, Department of Agriculture, NASA, and National Science Foundation (NSF) fund the towers in the USA. Although some towers have been in operation for many years, 1996 marked the start of a community effort to collect continuous measurements of ecosystem carbon and energy exchange to understand the controls on carbon fluxes. In 1997 the FLUXNET project was established to compile the long-term measurements of carbon dioxide, water vapour, and energy exchange from the regional networks into consistent, quality assured, documented data sets for a variety of ecosystems world-wide (Baldocchi et al. 1996, Running et al. 1999).

FLUXNET is a "partnership of partnerships", formed by linking existing sites and networks. As of early 2000 there are over 130 flux towers in FLUXNET. Measurements and terminology from existing but disparate sites and networks are brought together and harmonized into a common framework, thereby substantially increasing the usage and value of the flux data and information for the global change community. The core FLUXNET variables include both meteorological model-driving inputs (photosynthetic active radiation, air temperature, precipitation, relative humidity, wind speed and direction above the canopy, barometric pressure, soil temperature, and carbon dioxide concentration) and flux model checking variables (net ecosystem exchange [CO₂ flux] sensible heat, and latent heat from eddy correlation; net radiation; and soil heat flux). In addition, associated site vegetation, length of growing season, stand density, stand age, leaf area index, leaf nitrogen, edaphic, and hydrologic characteristics are compiled.

Gaps and proposed solutions: Significant issues that will challenge the use of flux data in TCO include the following.

(a) Intercalibration

A fundamental goal of the networks is to establish and maintain long-term intercomparability of results between the sites. Intercomparability is achieved through consistency in measurement techniques, strict attention to calibrations (and traceability to standards), and site intercomparisons in, for example, software processing of standardized flux data files (distributed by EUROFLUX) and comparisons of flux system response to a roving standard (as implemented by AmeriFlux). At present, the flux measurement community has agreed on common measurement techniques (http://www-eosdis.ornl.gov/FLUXNET/fluxnet.html). Most flux groups use common measurement techniques (open or closed path infrared gas analyser, 3-D sonic anemometer) and data processing routines. Resources must be devoted to verifying the overall comparability of flux measurements.

(b) Representative Sites

There are gaps in the distribution of flux towers, most notably so in savannah and desert biomes, in urban areas, in all successional states, and in managed systems. Funding for new flux towers may help to fill these gaps.

(c) Nighttime and Complex Terrain Bias Errors

At night, CO₂ flux measurements are subject to error and underestimation, as turbulent mixing is low. Drainage of CO₂ in sloping terrain is another compounding factor and has recently been investigated. Compensation for the expected under-estimation of nighttime net ecosystem exchange include the use of spatially extrapolated chamber measurements of leaf, soil and bole respiration, modelled values, or the u* correction (a relationship inferred from nighttime measurements during high turbulent mixing; e.g. Goulden et al., 1996). This correction remains an open research issue.

(d) Incomplete data

Data from eddy covariance measurements are usually reported in half-hour increments with an objective to collect data 24 hours a day and 365 days a year. However, the average data coverage during a year is only 65 percent due to system failures or data rejection. No universal method has emerged for the filling of missing or rejected data. Therefore, gap-filling procedures need to be established for providing complete data sets (Falge et al., 2000).

(e) Footprint and regional scaling

Towers typically sample fluxes within a kilometre of the tower, based on changing wind conditions. The characterization of the source area, or footprint, requires a detailed inventory of the vegetation and soils contained in the source area and the pattern of changing wind conditions. This information can be used with soil-vegetation-atmosphere transfer models as part model validation and scaling up to the region.

(f) Data Availability

Flux data are slow in becoming available to the broader scientific community. Although significant flux data are becoming available, additional incentives are needed to ensure the flow of data into the regional networks and ultimately into FLUXNET for distribution and archiving. Fluxes and ancillary information are unified in FLUXNET into consistent, quality assured, documented, readily accessible datasets via the World Wide Web (http://www-eosdis.ornl.gov/FLUXNET/).

Meteorological Variables

Present status

Current sources for meteorological forcing variables (Table 1) are a combination of existing ground-based meteorological networks, remote (satellite, radar) observations, and assimilation/interpolation from numerical weather models in nowcasting mode (ECMWF, NOAA, etc.).

Gaps and proposed solutions

(a) Precipitation: Precipitation data for terrestrial biospheric models are not available at the spatial resolution (< 1 km) needed to resolve topographic and other forms of landscape heterogeneity, nor with the temporal resolution (< 1 hr) needed to resolve short-term responses of water fluxes (especially canopy interception, infiltration and runoff) to intermittency in precipitation. The current global data sets have a spatial resolution of 0.5 to 1.0o and temporal resolution of days to months. (http://orbit35i.nesdis.noaa.gov/arad/gpcp/). Two alternatives exist in principle: (i) increase the spatial and temporal resolution of precipitation data, and/or (ii) develop improved parameterisations in ecosystem models for the statistical treatment of subgrid-scale processes in space as well as time. Both approaches are important because the available precipitation data are currently far from sufficient in spatial and temporal resolution, and will unlikely become adequate in the foreseeable future.

Because of its importance in many fields, the study of precipitation, both observationally and statistically, rapidly evolves. Resources are becoming available through these developments which need to be harnessed in the development of a strategy for terrestrial carbon observations. These developments include the following:

• There are developments in statistical downscaling of precipitation in both space and time ("weather generator"), incorporating techniques such as correlation of statistical attributes of precipitation with orographic variables (elevation, aspect, distance to coastline, others). This work needs to ensure realism in the statistics - not only of precipitation, but also its correlation with other variables such as radiation.

• Maintenance of the present surface measurement network represents a major challenge. It must deal with the problem of station closures and continue to address the long-standing but difficult question of gauge corrections.

• The GPCP is producing a global, on-line precipitation data set (5-day, 0.5o, retrospective to 1998,ongoing;http://orbit35i.nesdis.noaa.gov/arad/gpcp/).

• Regional studies under the GEWEX programme have produced intensive data for the midwestern USA and other regions (http://www.ogp.noaa.gov/mpe/gcip/index.htm).

• Coverage of North America and Europe by weather radar is now complete, and will rapidly extend to other parts of the world. Archiving and distribution of these data would augment precipitation data in the context of global terrestrial carbon observations.

• At least two satellite-borne remote sensing techniques are under active development: TRMM (active radar) and DMSP (passive 2-band microwave).

• Global NWP models (ECMWF, UKMO, NOAA, etc.) overcome most temporal and spatial consistency problems through their global domain coverage and high frequency of output reporting. However, (1) all relevant variables for surface water, carbon and energy balance determination need to be archived; and (2) caution is in order because properties and assumptions in the NWP land-surface scheme influence these outputs. Temporal consistency of the global NWP outputs may also be an issue of concern.

• Mesoscale NWP models provide even higher resolution but to minimize the problems of region-to-region inconsistency, these models should be provided with boundary conditions from a global model.

(b) Radiation: The key radiation variables are incoming solar, photosynthetically active radiation (PAR), and net radiation (or estimates of the upward and downward longwave components). Excluding the obvious diurnal cycle, the temporal and spatial variability of radiation variables is not as great as for precipitation but still remains a significant issue. There are also far fewer long-term, directly measured records for radiation than there are for most other meteorological variables. Strategies to deal with these issues include the following:

• A set of high-quality, long-term measurements of solar, net and PAR radiation needs to be established in the context of a terrestrial carbon observations, to calibrate satellite-based estimates and to improve local parameterizations of the long-wave terms. Consistency and quality of calibration is vital. These measurements can opportunistically be located at flux tower sites, although additional measurement sites are desirable.

• NWP and mesoscale models produce outputs that include radiation variables. However, the qualifiers regarding the usefulness of NWP-derived estimates (see a) above) also apply here. In addition, NWP archives are required to include all the terms in the surface radiation budget; they are usually calculated but not always archived.

(c) Temperature and Humidity: For these variables the effects of terrain heterogeneity are smaller than for either precipitation or radiation, although they are still potentially important. Orographically sensitive interpolation of data from existing meteorological networks (or NWP outputs in nowcasting mode) is a reasonable approach to obtaining data at the spatial and temporal resolutions needed for terrestrial carbon observations.

(d) Wind: Wind data are important for three reasons. First, they are necessary to specify aerodynamic transfers in models of land-atmosphere exchanges. The observation issues here are similar to those for temperature and humidity, resulting in the need to include orographic effects and to consider the role of atmospheric stability. Second, wind data are needed to determine the Green's functions in atmospheric inverse approaches (Enting, 2000). These are usually obtained from GCMs or NWP models, but there is a need to archive and interpret the sub-grid scale transports used in constructing the wind fields as these play a significant role in the forward calculation of the Green's functions. Third, wind information is needed to interpret flux tower data (flux measurements by eddy covariance, eddy accumulation, mass balance or profile methods) in any circumstances except for flat, homogeneous terrain. The acquisition and interpretation of wind data for this purpose is best undertaken in campaign mode rather than through long-term observations, though long-term measurements in the vicinity of some flux towers may be beneficial.

(e) Wet and dry deposition: Data on wet and dry deposition of nutrients and contaminants may be an important biogeochemical forcing input for terrestrial biosphere models. The main present requirement is to access existing networks. Additional measurements, for instance at flux tower sites, may require implementation as significance of this forcing becomes better understood. These requirements should be considered from a regional or biome perspective; for example, nitrogen deposition is known to be important for the boreal biome (McGuire et al., 1992).

An important overarching issue is commonality between the requirements of terrestrial carbon observations and 'terrestrial water' observations. While the latter is now largely carried out at national or regional scales, many of the compelling reasons for establishing a global carbon observing system (Chapter 2) extend to water as well. Linkages between carbon and water cycles and observations include:

• The close process links between the cycles of energy, water and carbon;

• The need in carbon cycle modelling for good specifications of water plant availability, including soil moisture and depth to groundwater table if the latter is accessible to plants;

• The improvement in modelling the water cycle through linkage to carbon cycle modelling, through the stomatal coupling between transpiration and carbon assimilation;

• The dependence of both carbon and water exchanges on similar suites of meteorological forcing variables.

Surface Fluxes and Stocks

Present status

Surface measurements and monitoring of carbon fluxes and stocks has a rich history. However, there are large gaps in the data in terms of (i) complete above and below ground components, (ii) spatial and temporal consistency, and (iii) completeness of an adequate spatial and temporal coverage. The surface measurements are produced by scientific research studies, inventories focused on commercial interests such as forest inventory or yield, and broader surveys or compilations, e.g. country-level statistics assembled by FAO.

Gaps and proposed solutions

The following describes some major gaps in information and potential ways to address these:

1. Forest stocks and productivity data at global to sub-national levels

Gaps:

• Limited access to the original plot-level measurements;

• Exact coordinates of plot data are not released due to confidentiality concerns;

• Not all the biomass is measured: aboveground components only focus on commercial tree species, focus on merchantable volumes, does not include litter production; thus the best way to use biomass data for total flux estimates is not well established;

• A variety of inventory methods are used with varying degrees of uncertainty;

• Accurate data on stock changes (due to harvest, fires, other disturbances) are not available, particularly at sub-national levels;

• Inventories usually have a good statistical design to estimate volumes and growth for large areas of forest; they do not provide information at a local level.

Solutions:

A two-prong approach is required: (i) increase access to quality forest biomass data, and (ii) develop methods for using the existing forest data and inventories to improve estimates of carbon fluxes. Some options are:

• Determine availability of the Food and Agriculture Organization (FAO) forest and other carbon-related statistics at the sub-national scale as part of Forest Resources Assessment (FRA) 2000 and other ongoing programmes. Such data are often collected but are mostly not centrally available in a country, even in the form of metadata;

• Work with forest inventory data for selected countries (e.g. USA, Canada) to demonstrate the potential use of inventory data for global terrestrial carbon observations;

• Explore the potential for acquiring and using long-term mensurational data for sub-national scales, including review of inconsistencies, deficiencies, etc. with various country programmes;

• Explore use of the data in combination with models based on land use, remote sensing, or other approaches to downscale national level inventory data to finer resolution.

2. Below-ground coarse and fine root biomass, root turnover rates

These observations are generally made at flux tower sites, but the characteristics of the distribution over large areas are not known.

Gaps:

• Scarcity of measurements, especially in tropical deciduous and boreal deciduous needleleaf systems (e.g. larch);

• Difficulty in performing measurements with consistent methods.

Solutions:

• Refer to the status of data and procedures to estimate root biomass based on soil and climate (Jackson et al., 2000);

• Promote the development of new measurement tools.

3. High resolution forest inventories

Depending on the resolution of the Kyoto Protocol reporting requirements, there will be a need for repeated measures of biomass/carbon with high degree of accuracy for small forest parcels. Traditional forest survey methods are generally too expensive to meet this need. Vegetation Canopy Lidar from aircraft or satellite provides the potential for the survey need (see also Satellite Observations). This issue will require attention once the Kyoto reporting requirements are agreed upon.

4. Soil carbon

In addition to point/soil profile measurements available at national or global (Soil and Terrain Database, SOTER) levels, a method for the spatial distribution of soil carbon has been developed by IGBP (Global Soils Data Task, 1999). The quality of the output is limited by the available site soil carbon information.

Gaps:

• Lack of measurements for many locations;

• Lack of soil depth data to compile an accurate soil carbon inventory;

• Inherent heterogeneity of soils at local scales.

Solutions:

• Ensure active sites measure soil carbon (at the sites and in surrounding areas, if feasible, using standardised methods);

• Promote new soil surveys specifically for soil carbon;

• Develop new soil carbon measurement techniques.

5. VOCs and other greenhouse gasses (methane, CH₄, NO_x, N₂O)

Gaps:

• Generally, estimates of NPP do not consider VOCs but these may be significant. For example, using recent emissions data and estimates of biome-specific ecosystem properties such as foliar density and emission responses to climatic factors, Guenther et al. (1995) produced a global model of total biogenic volatile organic compounds (BVOCs) fluxes. They estimated combined emissions of isoprene, monoterpenes, and other reactive volatile organic compounds to be 0.31, 0.15, and 0.21 Mg C ha-1 yr-1 for tropical rain forests, tropical montane forests, and tropical seasonal forests, respectively (Guenther et al., 1995);

• Lack of measurements in time and space;

• Measurements are difficult and costly.

Solutions:

• MOPITT (http://terra.nasa.gov/Gallery/MOPITT/) will provide relevant measurements from satellite for CO and CH₄;

• Review of the TRAGNET model. The United States Trace Gas Network (TRAGNET) (http://www.nrel.colostate.edu/PROGRAMMES/ATMOSPHERE/TRAGNET/TRAGNET.html) measures fluxes of CO₂, CH₄ and N₂O between ecosystems and the atmosphere to determine the factors controlling these fluxes. There are 25 sites representing a variety of regionally important ecosystems. Gas samples are taken from permanent chambers at prescribed intervals, typically one hour or less. Gas chromatography analyses the samples for CH₄, N₂O and CO₂;

• Development of a fast-response VOC measuring system.

6. Wetlands and coastal estuaries

With some exceptions, existing observations are inadequate to obtain accurate or representative spatial and temporal estimates of carbon fluxes in wetlands. The gaps concern both the distribution and functioning of wetlands (Sahagian and Melack, 1996).

Gaps:

• Aquatic issues have not been dealt with in terrestrial inventories.

Solutions:

• Satellite sensors (e.g. SAR) may provide information on wetland distribution and dynamics, especially concerned with water table.

There is a need to build linkages with aquatic communities to ensure that this component is included.

7. Missing biomes

Inventories of biomass are often poorly characterized for unique forests such as woodlands/savannahs, urban forests (human managed) and crops (especially in the tropics). Data for these ecosystems are often available from research studies, but are not compiled or archived systematically.

8. Comments

An overall approach to acquiring much of the desired in situ information could be to ensure that the variables will be measured at the existing sites within the networks associated with the TCO, such as the FLUXNET tower sites, EOS core test sites, the GTOS NPP sites, IGBP transects, etc. Sites that measure ecosystem fluxes are of particular importance since they provide the basis for enhancing the value of other site observations through process models tested against the flux measurements. The specific measurements are in Table 1 but should include where possible: soil carbon, root biomass and turnover, litter fall, phenology, decomposition (litter bags), canopy chemistry. Most of these are low cost and relatively simple measurements. In addition, the core variables of aboveground NPP, LAI, and others should be measured at all sites.

The acquisition of in situ observations around the globe is complex, regarding both the observations themselves and a strategy for their continued acquisition and availability. Some of the considerations or incentives to acquire these data include:

• Create a structure within TCO to coordinate the collection of common measurements and compilation of the resulting data;

• Prepare a manual, such as the BigFoot Field Guide (Campbell et al., 1999), to help standardise the measurements in different regions and biomes, and a way to compile the resulting data;

• Ensure that participating sites receive recognition for their contributions;

• Provide analytical services as needed, e.g. soil analysis;

• Provide an educational component, such as training workshops;

• Provide return-in-kind (e.g. remote sensing data) for access to in situ data;

• Use successful networks as models, e.g. the flask network, DIVERSITAS litterbag studies, etc.

Research studies have collected a large amount of information that has not been readily available. Therefore, another general approach to locating and accessing this extensive information is to collaborate with scientists in the countries of interest. A successful example of this approach is the IGBP-DIS Global Primary Production Data Initiative (Scurlock et al., 1999) which resulted in a comprehensive global database of NPP. Activities that may be useful to promote collaboration of this type include synthesis workshops and exchanges of students and researchers.

Satellite Observations

Satellite data are important in both top-down and bottom-up approaches (Table 1). For top-down, NWP or GCM models presently make the most extensive uses of these data, although their value for trace gases estimation has also been shown in research mode (Reichle et al., 1994; Connors et al., 1994) and will increase in the future (e.g. MOPITT; http://terra.nasa.gov/Gallery/MOPITT/). The section on page 28 discusses additional requirements for top-down satellite observations and the needed technological developments. The following sections deal primarily with bottom-up observation issues.

Status and recent progress

Satellites provide an important measurement technology for a number of essential variables, especially those used in upscaling from sites to globe. Table 1 identifies the observation requirements that may be met through satellite remote sensing, and Table 2 lists variables for which data products have been produced from satellite measurements. From Table 2, it is evident that remarkable progress has been achieved in converting raw satellite measurements into products useful for terrestrial carbon assessment. However, the quality of the products obtained so far (third column of Table 2) needs further improvements; this is discussed further below.

For terrestrial carbon observations, several kinds of sensors are required (Table 3). They differ in terms of the spectral bands (from about 0.4 mm to 21 cm), the illumination source (passive or active), spatial resolution (from ~25 m to ~1000 m), and the control over which region is imaged (fixed or remotely pointable). The conceptually most important sensor types are listed in Table 3. In virtually all cases, the technology is changing, thus the characteristics of specific sensor types also evolve. The current representatives of the various types are listed in Table 4 for sensors generally available to date (fine- and coarse- spatial resolution, SAR) and in Table 5 for recent, innovative concepts (very high spatial resolution, multi-angle, lidar, hyperspectral). Future research should focus on an effective use of data from these new sensors.

Compared with the situation 5-10 years ago, substantial progress has been made in several areas related to the use of satellite data for studies of the terrestrial biosphere. They include:

• The reduction in costs associated with the Landsat-7 data policy, leading to vastly improved data availability and the possibility of obtaining accurate information on land cover and land-cover change, by taking proper account of the spatial heterogeneity in carbon density and the processes governing carbon exchange;

• Improvements in data acquisition strategy. For example, a new approach has been developed for Landsat 7 which aims to maximize the number of useful (atmospherically uncontaminated) images over land (Arvidson et al., 1999, 2000);

• Availability of large-area radar data sets. Several campaigns have been carried out by space agencies to provide continent-wide mosaics of SAR data (the Amazon Basin, boreal forest of North America, etc.; examples can be found at http://trfic.jpl.nasa. gov/GRFM/worldmap.html);

• Convergence on land cover classes. There seems to be gradual convergence to the IGBP land cover classification scheme. GOFC and FAO have agreed that the FAO Africover scheme should be included in any classification exercise. Progress has also been made in novel land cover products, e.g. fractional cover, leaf type, etc.;

• Agreement was achieved on LAI as a product and progress was made on the LAI measurement and validation protocols (e.g. Chen et al., 1999; Campbell et al., 1999);

• New, larger datasets are being planned for a more routine production. For example, global burn scar products will be produced with SPOT-VEGETATION, MODIS and ATSR data;

• New sensors are being developed that will provide critical information for TCO (e.g. VCL; Table 5).

In spite of the above progress, gaps still remain in several areas. Much more needs to be done for satellite data to fulfil their potential in determining the distribution of carbon sources and sinks around the globe. Some of these gaps are discussed below.

Gaps and solutions

Gaps:

The most serious gaps or problems regarding satellite observations for terrestrial carbon include:

• Lack of commitment to long-term data continuity for fine resolution sensors (Landsat - type) and SAR (JERS-type) sensors. This problem is well understood and is related to programme planning and priorities of individual space agencies. IGOS-P is intended to assist in resolving this problem.

• Lack of commitment to long-term data continuity for critical observations begun with MODIS (in particular, into the NPOESS series). Building on the long (1983) AVHRR series, EOS/MODIS will greatly improve the quality and quantity of products needed for terrestrial carbon studies (Table 2). However, there is presently no assurance for data continuity beyond approximately 2005 (the nominal lifetime of the Terra satellite).

• Lack of consistency in, and access to, existing long-term archives. This problem exists in most archives, although its severity varies with agency and sensor. The access to older data is especially difficult. The problem includes not only storage media, but also access to metadata such as calibration information. Nevertheless, solutions are possible if enough attention is given to this challenge.

• Lack of commitment to providing information products, as opposed to data only, and to ensuring that in situ observations are used effectively in preparing the satellite-derived information products. The provision of information products that also incorporate in situ observations is fairly common for meteorological and oceanographic applications (e.g. Reynolds and Smith, 1994; http://www.nodc.noaa.gov/dsdt/oisst/index.html; http://ibis.grdl.noaa.gov/SAT/). However, such products are not yet available for the terrestrial environment over large areas. Current plans of some programmes include the generation of experimental data sets (e.g. EOS Terra; http://terra. nasa.gov/) but in situ data are intended to be incorporated in a limited way and as a research activity only. The incorporation of surface observations is exacerbated by the difficulty of accessing timely in situ data from various parts of the globe.

• Lack of globally applicable, robust algorithms and other infrastructure needed for low-cost, large volume production of information products. Numerous algorithms have been developed based on existing satellite data and applied to partial (in spatial and/or temporal coverage) data sets. Their limitations are due to the input data (calibration, resolution, spectral or angular coverage) or adequate validation in a broad range of terrestrial environments. Progress is being made in both areas, with the recent or planned satellite launches and more systematic international coordination of validation activities under CEOS and individual programmes (e.g. http://modarch.gsfc.nasa.gov/MODIS/LAND/VAL/;http://www.ceos.org/). Previous limitations due to large data volumes and processing requirements melt away as the computing power and communication bandwidths increase. However, reliable product availability requires agency commitment and support, and this has been lacking so far for terrestrial products.

• Poor inter-agency coordination of observations. To date, the coordination of terrestrial data acquisition among space agencies has been minimal. This leads to sub-optimum use of the existing space assets, and to data and products with built-in limitations. It is especially important where data from various sensors could be used in a complementary fashion, e.g. tradeoffs between spatial coverage and timeliness of acquisition (Ahern et al.,1998). IGOS-P should be an effective mechanism for making rapid progress in this area.

• The unavailability of satellite data for important terrestrial variables. The current and near future satellite programmes cannot provide data for several ecosystem variables that are important for upscaling. They include aboveground biomass, soil moisture, leaf nitrogen, water table depth (especially for wetlands), and precipitation (at high spatial resolution). These areas require focused investigations regarding options for sensing technologies, to lead to the design of suitable satellite missions. In addition, some clearly promising techniques (e.g. lidar for biomass and canopy structure) require further technological development to improve the information content of the data (increase of coverage, resolution, etc.).

Solutions:

• Ensure commitments to long-term continuity of data from fine resolution and SAR sensors;

• review NPOESS specifications from the TCO perspective to ensure that critical observational needs are met;

• insist on the provision of information products, not just data;

• invest in reprocessing archived satellite image series such as AVHRR;

• identify a small set of C-related products which satellites can deliver (based on Table 2);

• establish a process that will lead to consensus algorithms for specific products. In many cases, combinations of data from multiple satellites are necessary or highly desirable;

• ensure provision of data in common formats to facilitate their integrated use. A suitable starting point would be co-registered fields of radiances and then reflectances. These will require reliable radiometric calibration and geometric correction (ortho-rectification to a common geometric base);

• desirable early information products include LAI and land cover (fractional cover, other parameters) from each satellite.

Specific steps regarding most of the above proposed (and other potential) solutions need further discussion before devising plans for implementation.

As noted above, sound global validation strategy is an essential component of the use of satellite data for terrestrial carbon observations. The approach should focus on network sites where in situ measurements and process studies are combined with the available satellite data for algorithm development and comparison of products with independent estimates. As part of the algorithm intercomparison and validation strategy, action is also needed to set up a community process to define and implement priority locations for acquisition of high and very high resolution data. Such data could be purchased from commercial operators, obtained by coordinated targeted observation from multiple sensors with a restricted duty cycle, or assembled by separating out relevant data from unwieldy data streams into a separate archive.

Table 2. Products derived from satellite observations

¹ Maturity: 1 = can be produced now,; 2 = within 5 years,; 3 = after >5 years.

² Quality in production mode: A = excellent; B = satisfactory; C = fair.

³ Sensor type needed: Fine = pixel spacing ~25 m; Coarse = pixel spacing ~ 250-1000 m; F= fixed pointing (nadir); P = programmemable pointing. Fusion of data from two or more sensors often required to generate a product.

⁴ The precise definition of the suite of vertical and horizontal structure products will needs further more discussion and negotiation within the community; clumping index is a possible additional product.

⁵ These products would be very valuable, but may be difficult to achieve with current and foreseeable technology.

Table 3. Generic sensor types

* with 4 or more independent looks

Table 4. Current specific sensors

Table 5. Specific sensors - new or anticipated capabilities