Post et al. (1999) outlined the elements of a plan for monitoring and verifying SCS at regional scales. The plan had four basic elements: (1) selection of landscape units suitable for measuring and monitoring SCS, (2) development of measurement protocols, (3) utilization of remote sensing information and simulation models, and (4) development of a methodology to scale up results to represent the entire SCS project. The selection of landscape units for measuring and monitoring SCS will depend on the general responsiveness of a region to SCS practices such as climate and soil properties, management history, and availability of research data. Further, an initial estimation is required on how many farms could be involved in a project. For example, Izaurralde et al. (1998) estimated that a project promising to deliver 500 Tg C in the form of SCS would require the participation of about 1000 farms. This assumed a conservative SCS rate of 1 Mg C ha-1 and an average farm size of 500 ha. This element will require the participation and interaction of regional agronomists, soil consultants, and farmer organizations who as a group can decide on the selection of the best pilot areas and the extent to which the results can be extrapolated.
Another element of the plan deals with the development of methods for detecting changes in SOC that would take place during several years after the implementation of an SCS practice. Changes in SOC can be detected as changes in stocks or fluxes. Changes in stocks can be calculated from direct measurements of SOC stocks present when the SCS practice is implemented and a number of years after its implementation. Eddy covariance methods can also be used to determine net ecosystem exchange, which is the difference between net primary productivity and soil respiration, and thus provide an indirect but accurate estimate of SOC change (Baldocchi and Wilson, 2001). Eddy covariance methods are attractive because they would provide estimates of SCS over fields, but their use will remain experimental until they prove their applicability for these types of projects. A methodology for direct measurement of SOC changes should include the following: (1) selection of sampling sites; (2) soil sampling protocols (depth and volume of sampling); (3) ancillary soil measurements (bulk density); (4) sample treatment and analyses; (5) ancillary field measurements (plant biomass and yield, plant residues, and soil erosion); and (6) calculation of results including adjustments for marginal costs of SCS. Details and examples for some of these components will be described in the next section.
Satellite remote sensing and simulation models have key roles in monitoring and predicting SCS at field scale and for extrapolating SCS from field to regional scales (Post et al., 1999). Remote sensing methods can be useful for monitoring net primary productivity, estimating crop yields and leaf area index (LAI), and — in combination with ground-based data — deriving maps of land cover and use history. Currently, however, it is not possible to directly estimate SOC stocks unless ground-based information on SOC values is available (Chen et al., 2000).
Simulation models have a key role to play in understanding and predicting SCS at project and regional scales. Currently, there are numerous soil and ecosystem models capable of describing SOC trajectories in response to varying climate, soil, and management conditions. McGill (1996) reviewed ten soil organic matter (SOM) models, and classified them in terms of their treatment of environmental (weather) drivers; temporal scale (day, month); vertical distribution (i.e., soil horizons); soil properties (clay content); and litter description (kinetic vs. biochemical). Subsequently, Smith et al. (1997) compared the performance of nine of them against 12 data sets from seven long-term experiments representing a range in land uses (grassland, cropland, woodland), climatic conditions within the temperate region, and treatments (nutrient sources and rotations). While no one model performed consistently better than the others across all data sets, a group of them (RothC, CANDY, DNDC, CENTURY, DAISY, and NCSOIL) had lower model errors than another group (SOMM, ITE, and Verberne). Well-tested models will be particularly useful not only for understanding interactions between biophysical and management variables but also for projecting SCS over large areas. The issue of upscaling is very important because it requires the integration of information from a variety of sources such as field measurements, geographic information systems and associated databases, computer models, and remote sensing.
Table 19.1 characterizes a series of feasibility and pilot projects designed to evaluate SCS under a variety of climate, soil, and management conditions. The table also provides
Table 19.1 Examples of Feasibility and Pilot Projects on Soil Carbon Sequestration
McConkey Cropland (2000)
Pacific Northwest, Scholz United States (2004)
Vergara Sánchez et al. (2004)
Crop/natural fallow secondary forest
Agroforestry Communal Field
Direct seeding/ cropping intensification Direct seeding/ cropping intensification Fruit tree intercrops with annual crops/ conservation tillage Direct seeding
Stratified random sampling
Soil Carbon Analysis
Dry combustion No
Gridded within fields
Dry combustion SPOT
Nutrient management, N fixation agroforestry Tree conservation/ Random ridge tillage within fields
Paired sampling Random, re- Wet and dry sampled combustion
Wet combustion Landsat Yes
Agriculture to grassland
Cropland State/private Region
Notes: SPOT = Système Probatoire pour l'Observation de la Terre; AVHRR = Advanced Very High Resolution Radiometer.
Wet and dry combustion, loss of ignition, infrared spectroscopy Carbon flux towers
4 to 5 million ha
CQUESTER 2600 ha
1 million ha
3.3 million ha
AVHRR, CENTURY Landsat c
•fe. a information about management practices, sampling techniques, analytical procedures, and the use of models and remote sensing tools. Details on soil sampling, soil analysis, and the use of simulation models and remote sensing for assessing SCS are discussed in the next section.
19.3 DETECTING AND PREDICTING SOIL ORGANIC CARBON CHANGES: SAMPLING, ANALYSIS, MODELS, AND REMOTE SENSING
Soil sampling for determination of SOC content represents the most direct way to determine whether a particular soil management or land use practice has caused a net change in SOC storage. Soils are three-dimensional natural bodies that vary in time and space in response to biophysical properties (i.e., climate, topography, parent material, and vegetation) and management (fertilization, tillage, land use change, etc.). Detailed soil surveys can greatly assist in the design of a sampling scheme. Once the general properties about the field to be sampled are obtained (soil series, soil phases, landscape characteristics, etc.), the next step consists of deciding the sampling plan. According to Petersen and Calvin (1996), the sampling plan should help select which units of the soil population are to be included in the sample. The best sampling plan would be the one that provides the maximum precision at a given cost or, conversely, the lowest cost under a specified precision and error. There are three sampling plans recommended: (1) simple random, (2) stratified random sample, and (3) systematic. In the simple random plan, each unit of the total n units being sampled from the population has an equal chance of being selected. A stratified random scheme would be more desirable when there are variations that can be predicted from knowledge about soil or landscape properties. This stratification would help reduce a source of variation of the sampling error. A systematic sampling plan occurs when soil samples are taken at regular distances from each other, either in one or two dimensions. In principle, systematic sampling would yield more precise results than the first two sampling schemes (Petersen and Calvin, 1996).
The number of samples (n) to be taken from a field depends on the variability in SOC content as well as on minimum difference that needs to be detected. For example, Izaurralde et al. (1998) used a one-tailed t test to calculate the number of soil samples needed to detect, with a 90% confidence, a 0.1% increase in SOC with a known variance of 3.3 (g C kg-1)2. They calculated that for each representative parcel of land the baseline sampling would require 54 samples, while the final sampling would require another 54 samples, a large number indeed. Similar calculations were carried out by Garten and Wullschleger (1999), who evaluated the statistical power to detect significant SOC differences under switchgrass (Panicum virgatum L.). They calculated the smallest difference in SOC that could be detected between two means for a given variance, significance level, statistical power, and numbers of samples. They concluded that while differences of about 5 Mg SOC ha-1 were detectable with reasonable numbers of samples (n = 16) and good statistical power (1 - P = 0.90), the smallest difference in SOC inventories (1 Mg SOC ha-1) would be detectable only with large numbers of samples (n > 100).
In order to reduce the number of samples required and to minimize soil variability, Ellert et al. (2001) proposed a high-resolution method to detect temporal changes in SOC storage by comparing the quantities from a sampling microsite (4 x 7 m) at two sampling times separated by periods of 4 to 8 years. In this method, one to six microsites are selected in such as way so as to represent the dominant soils found in fields ranging from 30 to 65 ha. Guidance for the location of the microsites is obtained from experienced pedologists. The location of each microsite is recorded by survey methods, including geographic positioning systems. The authors also made useful recommendations regarding core size and number, time of sampling, depth of sampling, and ancillary measurements. This methodology was successfully applied to the PSCBP in 1997 to 2000, and allowed for the statistically significant detection of SOC storage gains as small as 1.2 Mg ha-1, only 3 years after the implementation of a SCS practice.
Upscaling point measurements of SOC storage to the field level requires confidence in the assumption that the properties of the point measurements, including their measurement errors, will hold across the area of prediction. This confidence has been growing by an increased understanding of the relationships of soils in the landscape. The spatial dependence of soil attributes, including SOC content, has been studied with a variety of techniques or tools, including soil and topographical surveys, geostatistical techniques, remotely sensed data interpretation, as well as ground and monitoring devices. Like other disciplines, soil science has greatly benefited from advances in computation and information technology (e.g., McBratney et al., 2003). A few examples of these approaches follow.
Pennock et al. (1987) proposed a segmentation procedure to describe landscapes into functional units (i.e., landform segments such as shoulder, backslope, footslope, and depression). Pennock and Corre (2001) used it to study the comparative effects of cultivation on soil distribution and SOC storage, and to understand the main landscape features controlling soil emissions of N2O. This approach was used in the PSCBP to help delineate the sampling areas for monitoring SOC changes. MacMillan et al. (2000) expanded on Pennock's approach and developed a model, which based on digital elevation models (DEMs) and fuzzy rules, could identify up to 15 morphologically defined landform facets. A consolidation in the number of land-forms can be obtained to provide units at a farm field scale that are relevant for benchmark soil testing, application of simulation models, and precision farming.
Geostatistical methods are being increasingly used to predict soil attributes. Odeh et al. (1994) compared various interpolation methods (e.g., multilinear regression, kriging, co-kriging, and regression kriging) in their ability to predict soil properties from landform attributes derived from a DEM. The two regression-kriging procedures tested performed best, and thus showed promise for predicting sparsely located soil properties from dense observations of landform attributes derived from DEM data. Triantafilis et al. (2001) had success in using regression kriging to predict soil salinity in cotton (Gossipium hirsutum L.) fields with electromagnetic induction data. They attributed the success of the method to the incorporation of regression residuals within the kriging system. Hengl et al. (2004) tested a framework based on regression-kriging to predict SOM, soil pH, and topsoil depth from 135 soil profile observations from the Croatian national survey. These research results are promising, as they anticipate the possibility of implementing these algorithms in a GIS, thus enabling the interpolation of soil profile data from existing data sets (Hengl et al., 2004). The challenge remains, however, of developing rapid methods to accurately estimate SOC stocks in space and time (including uncertainties) at a relatively low cost.
Was this article helpful?