Land use and land cover changes, including legal and illegal deforestation, are amongst the most important factors that contribute to the social and environmental challenges facing mankind in the 21st century. Deforestation alone is responsible for about 12% of the world’s anthropogenic greenhouse gas (GHG) emissions, whereas another 6% stems from peat oxidation and fires on degraded peatland areas [1]. The combined effects of logging and forest regrowth on abandoned land are responsible for 10–25% of global human-induced emissions [2, 3]. Annual emissions from deforestation in Indonesia and Brazil equal four-fifths of the annual reduction target of the Kyoto Protocol [4].
Linking deforestation with climate change as a mitigation action was one of the key decisions of the thirteenth Conference of the Parties (COP) of the United Nations Framework Convention on Climate Change. The Bali Action Plan agreed:
“Enhanced national/international action on mitigation of climate change, including, inter alia, consideration of…policy approaches and positive incentives on issues relating to reducing emissions from deforestation and forest degradation in developing countries; and the role of conservation, sustainable management of forests and enhancement of forest carbon stocks in developing countries” [5].
These actions are now referred to collectively as REDD+ .
Under the UNFCCC, the REDD+ instrument (Reducing Emissions from Deforestation and Forest Degradation), as agreed at the COP-16 of the UNFCCC in December 2010 [6], is critical for developing countries.
REDD+ includes the implementation of the following mitigation actions:
-
(a)
Reducing emissions from deforestation;
-
(b)
Reducing emissions from forest degradation;
-
(c)
Conservation of forest carbon stocks;
-
(d)
Sustainable management of forest; and
-
(e)
Enhancement of forest carbon stocks.
This means that, potentially, all forest resources in developing countries are subject to accountable mitigation actions. The Cancun agreement also stipulates that robust and transparent national monitoring systems of the above mitigation activities shall be developed. As a consequence, for the implementation of REDD+, it is crucial to determine the spatio-temporal variation of carbon stocks. Obtaining field measurements and developing estimation models to do so is an expensive and time-consuming task.
This systematic review will compare methods of measuring carbon stocks and carbon stock changes in key carbon pools and land use categories/activities identified by the Intergovernmental Panel on Climate Change (IPCC) and the UNFCCCa. Figure 1 illustrates these carbon pools and shows the main fluxes of the global carbon balance that are covered in this systematic review. The systematic review is not designed to provide technical guidance, such as those outlined in the Intergovernmental Panel on Climate Change (IPCC) Good Practice Guidance and Guidelines [7, 8], or to be a sourcebook of methods, such as GOFC-GOLD [9], but is aimed at providing an exhaustive comparative literature review in this field, including the provision of published data on uncertainties.
Carbon stocks in biomass
The discussion below has a focus on tropical forests to highlight current methodological issues that are pertinent to forest science in general. However, this systematic review covers methods used in all forest types in all biomes.
A key challenge for successfully implementing REDD+ and similar mechanisms is the reliable estimation of biomass carbon stocks in tropical forests. Biomass consists of approximately 50% carbon [11, 12]. Uncertain estimates of biomass carbon stocks of tropical forests resulting from difficult access, limited inventory and their enormous extent, [12–14], prohibit the accurate assessment of carbon emissions as much as uncertainties in deforestation rates [15]. The carbon stocks of interest are both above-ground and below-ground.
Although above-ground biomass (AGB) has attracted by far the most research over the years, pools of deadwood and litter could be as large as above-ground biomass. It is essential that a variety of methods to measure deadwood and litter should be reviewed. Deadwood pools, including standing dead trees, fallen woody debris, and decaying and burned wood, are of particular interest in projecting carbon losses from decomposition. They are also often used as an indicator of carbon losses from degradation due to logging [16] or fire [17]. Data collection regarding standing dead trees frequently follow the same protocols as those for AGB inventories but ideally should also include data on levels of decay. Woody debris is most often estimated using the line-intercept method which measures only debris which crosses a transect (e.g. [16]) or through rectangular plots wherein the dimensions of each piece of debris is measured (e.g. [18]). Although some studies have addressed the densities of woody debris of different decay classes [16, 19], more regionally and biome-specific studies would help refine estimates of carbon content of this pool (e.g. [20]).
A reliable estimation of AGB has to take account of spatial variability, tree and forest metrics (allometric models) and wood. Many studies have been published on AGB estimates in tropical forests around the world (e.g. [21–29]), whereas the volume of literature on below-ground biomass estimates in tropical areas is relatively small (e.g. [30–40]). Indeed, because root systems have particular features and require highly specific procedures [41], measurements are very often time consuming and costly, qualitative, focussed only on one specific application and often not representative of large areas, as they generally involve a small number of root systems. In some cases, however, new methods (e.g. three-dimensional root architecture data analysis) can be used to compute the continuous spatial distribution of coarse root volume, biomass, external surface and specific root length [41].
Several databases provide harmonized above-ground and below-ground biomass information: for example, the World Forest Biomass and Primary Production Database [30], the database and geography of Forest Biomass of Northern Eurasia [42] and the Compartment Database of the European Commission Joint Research Centre. Similar databases for tropical areas would be extremely valuable.
Most studies on tropical forest AGB have been conducted in the Brazilian Amazon and in Southeast Asia. Few studies have reported on AGB for forests in Africa (but see [28]). The large number of published biomass equations [43] indicates that there is a substantial variation in tropical forest biomass [44].
Problems of errors and bias
There are four types of uncertainty associated with AGB estimates of tropical forests (adapted from Chave [45]):
-
Inaccurate measurements of variables, including instrument and calibration errors
-
Wrong allometric models
-
Sampling uncertainty (related to the size of the study sample area and the sampling design)
-
Poor representativeness of the sampling network.
Vieira et al. [46] demonstrated the effect of inaccurate height measurement. A stem with a diameter at breast height (DBH) of 20 cm and a height of 13 m gave an AGB of 153.0 and 127.0 kg, respectively, when using models of Chave [47] and Scatena [48]). With the same DBH but one metre more height, the estimated AGB values become 164.1 and 136.6 Kg, i.e. an increase of around 7% and 5%, respectively. Large footprint LiDAR (Light Detection And Ranging) data (e.g. Geoscience Laser Altimeter System - GLAS) and small footprint LiDAR data (e.g. airborne laser scanner - ALS) can be used to retrieve indirect tree height estimates, however the elevation differences that are present within the footprint, especially for large footprint LiDAR data, can be substantial in comparison with the predominant tree height and make it difficult to estimate accurately tree height [49]. Terrestrial laser scanning can also be used to estimate indirectly tree height at plot level; however as tree height, branching frequency and stand density increase the quality of the information obtained from the terrestrial laser scanner decreases as a result of the inherent occlusion effects and increasing point spacing, and the related uncertainty as to whether the highest returns are echoes from the tree tops or echoes from inside the tree canopy. Because of these data quality problems, using small footprint LiDAR data to retrieve tree height would therefore be preferable [49].
Another important error is the wrong choice of allometric model, which is related to the representativeness of biomass sampling. Allometric equations relate easily-measured variables of an organism (such as diameter and height) to attributes that are more difficult to assess (such as volume, leaf area, and biomass). They aim at facilitating large-scale estimation of complex parameters [50], by, for example, providing ground reference for remote sensing or for estimating regional biomass. Height and diameter are the most common dependent variables for assessing tree biomass, but as height of individual trees has been difficult to measure, most allometric models for tropical forests are based only on tree diameter [51, 52]. Although there are cases where height is the key independent variable for explaining variations in biomass (e.g. in palms), measurements of DBH, which is typically used for trees, explains more than 95% of the variation in tree biomass even in highly species rich tropical forests [53] .
Currently, allometric equations are almost entirely based on Southeast Asian and South American measurements. Some equations are available for African tree species or forest vegetation types [54], but there are no allometries based on destructively sampled trees for Central Africa [47]. Biomass equations for North America are listed in Ter-Mikaelian and Korzukhin [55], and for Europe in Zianis et al. [50] (2005). Similar databases for other parts of the world would be of high value.
Most biomass papers are based on 10–30 sample trees per species, which is far too few for biomass estimation of large countries in the tropics. The accuracy of biomass estimation ultimately depends on the accuracy of the original measurements used to develop biomass assessment tools, such as allometric models, biomass expansion factors (BEFs), and generic equations [56, 57] and species group specific volume-to-biomass models [58]. In Europe, large data sets were compiled from which generic models were devised following several small scale biomass measurement campaigns. This sort of richness has not been replicated in other areas of the world to this extent. Therefore, the lack of representativeness is the major drawback with current biomass equations.
It is time-consuming and costly to sample sufficient trees to acquire information on species and size distribution in a forest (particularly in a highly diverse tropical forest). Grouping all species, even in species-rich tropical forests, produces regression equations with high r2 (generally greater than 0.95) [53]. Therefore using generic regression equations stratified by, for example ecological zone or species group (broadleaf or conifer), might increase the accuracy and precision of the equations, because they tend to be based on a large number of trees and span a wider range of diameters [53], except in those cases where unique plant forms occur (e.g. species of palms and early colonizers) and developing of local regression equations is recommended.
There is thus a clear need for country- and region-specific studies to address the validity and reliability of allometric models. Ideally, such studies would utilize good ecological plot data, but these are often of poor quality or lacking completely. Commercial inventory data gathered by private companies are therefore used as an alternative and rich source of site-specific data. These are necessary for improving methods for estimating forest carbon, but are generally not available in the published literature or readily accessible from those who hold the data.
Guidelines for measuring wood specific gravity (WSG) in the field exist, but for tropical regions published WSG data are limited to a few commercial timber species that represent only a fraction of the forest biomass. WSG data on other species are scarce or lacking.
BEFs, for example, strongly depend on stand structure [59, 60] and site characteristics [56, 61] and extrapolation with BEFs may lead to biased results when compared with local biomass equations [62], indicating the importance of representativeness and the risks of extrapolation.
Furthermore the biomass stock of tropical forests and its distribution remain poorly resolved at the regional scale [15, 63, 64]. Consensus has also yet to be reached on how much carbon is being emitted by changes in tropical land use (see, for example, [1, 44, 65–69]. There is thus an urgent need for calibrating and improving the methods for determining tropical forest biomass and its spatial distribution [70].
Carbon stocks in soils
Soils are the largest carbon reservoir of the terrestrial carbon cycle. Worldwide they contain three or four times more organic carbon (1500 Gt to 1 m depth, 2500 Gt to 2 m) than vegetation (610 Gt) and twice or three times as much carbon as the atmosphere (750 Gt, see Figure 1) [71]. Carbon storage in soils is the balance between the input of dead plant material (leaf, root litter, and decaying wood) and losses from decomposition and mineralization of organic matter (‘heterotrophic respiration’). Under aerobic conditions, most of the carbon entering the soil returns to the atmosphere by autotrophic root respiration and heterotrophic respiration (together called ‘soil respiration’ or ‘soil CO2 efflux’). The mineralization rate is a function of temperature and moisture levels and chemical environment with factors such as pH, Eh, nitrogen level and the cation exchange capacity of the minerals in the soil affecting the mineralization rate of soil organic carbon (SOC) [72–78]. Under anaerobic conditions, resulting from constantly high water levels, part of the carbon entering the soil is not fully mineralized and accumulates as peat.
Guo and Gifford conducted a meta-analysis of 74 publications on the influence of land use changes on soil carbon stocks [79]; (see also a follow-up study by Laganière et al. [80]). They acknowledge the possible bias in their findings as most data drew from only four countries (Australia, Brazil, New Zealand, and USA) and a limited number of studies. They point to the need for a more comprehensive analysis of some hypotheses generated in their study: soil carbon stocks decline after land use changes from pasture to plantation (−10%), native forest to plantation (−13%), native forest to crop (−42%), and pasture to crop (−59%). Soil carbon stocks increase after land use changes from native forest to pasture (+8%), crop to pasture (+19%), crop to plantation (+18%), and crop to secondary forest (+53%). Most land use on peat soils requires drainage and is associated with a continuous loss of soil carbon stock.
One of the limitations of many of these studies of the effect of land use change on soil carbon levels is that often the plots were established to study other parameters, such as soil fertility, and early experimental protocols did not consider extensive measurements of soil carbon throughout the soil profile. Many studies only measure the carbon change in the top 20 to 30 cm of the soil profile and therefore do not show the effect of leaching and activity by earthworms, etc., on the movement of carbon down the profile. In addition temporal sampling of SOC measurements tend to be sparse and insufficient in number and interval to estimate SOC decomposition rates and final equilibrium [81]. Eddy covariance measurement of CO2 flux is valuable in this regard if they cover longer time periods spanning a year or more [81]. In this respect the fluxnet network of long term eddy observation represents an invaluable source of information, as illustrated in part in Figure 2
[82, 83].
Mineral soils
Estimates of soil organic carbon (SOC) stock changes are applied to determine long-term carbon fluxes and to design carbon sequestration strategies. Several approaches to estimating these stock changes are currently in use and may provide conflicting results.
On the plot scale there are a number of methods to detect changes in soil organic carbon stocks. Repeated soil samples over a range of years to decades have been noted above, but, in addition, long term flux measurements covering comparable time periods, either by soil chambers or eddy covariance systems, can provide estimates of stock changes. Long-term eddy covariance plays an important role in this context because, depending on its footprint area, it is able to give an integrated picture of the ecosystem under consideration [81].
In comparison to chamber measurements which are at the site or plot scale, eddy flux measurements therefore provide a direct link to a higher spatial level and allow integrative analysis. However, there are a range of uncertainties associated with the method, such as u* correction, gap filling, outlier filtering, advection and flux partitioning etc. [75, 84]. For the estimation of soil carbon stock changes only, systematic uncertainty components are of relevance, such as those outlined in, for example, Lasslop et al. [85].
In order to scale-up plot estimates to the landscape, country or continental scale, additional information about the spatial arrangement of soil types and land cover/land use needs to be considered, which introduces additional sources of uncertainties [81]. One method for estimating SOC stocks of different ecosystems is a regression approach in which regional SOC densities (mass SOC/area) are related to a number of auxiliary variables like temperature, precipitation, age class, and land-use history. An updated methodology applies a geographic information system (GIS) to calculate SOC densities for each forest type within a region from soil databases and satellite-derived land cover information. Campbell et al. (2008) showed large differences in the outcomes of both approaches and identified the need to use direct measurements of SOC in order to determine absolute errors in both approaches [86]. The fact that the methods have been used interchangeably in the past indicates that errors will have been perpetuated in the literature. Both methods are valuable for estimating soil carbon stocks but not for carbon stock changes, because the predictors of both parameters are different.
According to Mäkipää et al. [87] a reliable carbon stock change inventory for Finland with repeated soil carbon sampling would take 10 years and cost 8 million Euros. This high cost would almost certainly prevent more than a few countries from investing in soil carbon inventories; the reality is that a combination of models and additional measurements is needed. Regardless of the methodology applied there is a clear need to identify the uncertainties associated with current understanding of SOC stocks and stock changes [88]. It is important to pay particular attention to changes in soil carbon stock through direct measurements and soil carbon modelling [89] as well as to regional variation of soil carbon stock.
Soil carbon models can be used for estimating carbon stocks and stock change estimation but they need to be parameterized and validated for each land use, vegetation cover, climatic condition and soil type. This requires measurements of the soil carbon in the soil profile as well as spatially. In addition, models divide the soil carbon into theoretical pools with different decay rates and their validation requires measurements of soil carbon types that can be related to these pools [90, 91]. Soil carbon models also face the initialization problem due to the model partitioning the soil carbon into pools with different decay rates. To model soil carbon changes associated with land use or land management change, the initial proportion of these partitions needs to be known for the first land use. For change from a long term land use such as savannah or forest to cropland, soil carbon models assume that at the beginning of the simulation period equilibrium conditions apply, i.e., that plant input and soil carbon stock are in balance given the local climatic conditions. In order to simulate land cover or land use change under constant or changing climatic conditions the model needs a so-called ‘spin up’ period to reach equilibrium condition before any transition takes place. For cases where the prior land use is less than 100 years the land use history must be known in order for this spin-up to be made, accounting for all the historical land use changes. Alternatively the initial pools can be estimated from actual measurements of the soil carbon pools [90]. There are a number of problems and uncertainties related to this assumption and alternative methods are discussed. It is therefore essential to quantify these effects in any soil carbon accounting [88, 90, 92, 93].
Organic (peat) soils
Only recently has science recognized the importance of organic (peat) soils for greenhouse gas emissions and climate change. With some 500 Gt of carbon stored on only 4 million km2 (= 3%) of land, peatlands constitute the world’s most dense terrestrial carbon stocks [94]. In the case of peat swamp forest, emissions from peat oxidation and peat fires following drainage may be significantly larger and longer-lasting than above-ground emissions from clearing or burning forest vegetation. Peat oxidation currently leads to worldwide emissions of some 1.3 Gt CO2 per year, whereas peat fires contribute another 0.6 Gt CO2 per year on average [93]. During the 1997–1998 El Niño drought, peat fires in Southeast Asia emitted some 1.8 Gt CO2
[95–97], which is equivalent to 10% of the total global anthropogenic emissions for the same year.
Many variables linked to peat oxidation are not well understood and few reliable measurements exist for many of them. Uncertainty begins with the extent of peatlands worldwide, and especially in the tropics and with the amount of carbon stored in the peat layer. The degree of peat humification has strong influence on the mass of peat and carbon per volume, the hydraulic conductivity and the moisture retention capacity. Knowledge of the 3D topology of peatlands is important for hydrology and modelling, but peat depth and peatland shape have been measured only in a few locations (e.g., mapping by petroleum exploration companies in Borneo forests). Sampling sufficient locations to allow for spatial modelling is a time-consuming and costly exercise. New technologies may be capable of reducing time and effort.
Even less is known about emissions factors, which are essential for reliably estimating GHG emissions. Emission estimates from peat fires have large uncertainties, because of the highly variable mass of peat combusted and the various gases emitted depending on fire severity, water table, peat moisture and fire history. Data on most of these parameters are scarce or lacking. Long-term GHG emissions from biological oxidation of peat are even more significant than the emissions from peatland fires [97, 98]. Very few long term (> 1 year) measurements exist to assess emission rates under different water management regimes. A recent review shows that drained peatlands emit in the range of 9 CO2 t/ha/yr from peat oxidation for each 10 cm of additional drainage depth [97]. The role of tropical peat swamps is crucial not only in terms of GHG emissions but also for REDD+, as their peat carbon stock is on average 10-times larger than their above-ground biomass stock [94] and significant amounts of carbon are released by fire and bacterial decomposition. Emissions from drained peatland occur worldwide. The largest emitters include Indonesia, the European Union, Russia, China, USA, Malaysia, Mongolia, Belarus and Uganda [98].
It is important to make the distinction between litter and soil when assessing terrestrial carbon stocks and to ensure that accurate data are collected and analysed. Litter includes leaves and other fallen plant material (including fine woody debris of diameter less than 2 cm). Litter may be equivalent to only a small fraction of AGB in some ecosystems (e.g. 2% for montane forests in Mexico [99]) whereas it can be substantially higher in others (e.g. 30% in sugarcane fields). Research on carbon in litter has been as neglected as that of below-ground biomass, but it is a pool that must be taken into consideration when estimating carbon losses and movement between pools [100]. This is important for the parameterization of soil carbon models. Some estimates of the litter pool in forests use quadrats to assess the litter mass per unit area at a given point in time [100]. However, this method may suffer from imprecision owing to the difficulty of distinguishing between litter and soil organic matter. Litterfall traps, which can monitor the input of litter falling over time, may be more accurate in distinguishing between pools. Many studies are available which address decomposition rates and the implications for carbon cycles [101, 102].
Carbon stocks in agriculture and croplands
According to the Fourth Assessment Report from the Intergovernmental Panel on Climate Change (IPCC), agriculture, forestry and other land uses (AFOLU) account for approximately 30% of the total anthropogenic GHG emissions [103]. Of these, agriculture accounts for about 60% of N2O and 50% of CH4 emissions, whereas deforestation and land use change are responsible for most of the CO2 emissions [103]. It is recognized that the relationship between agricultural practices that affect CO2, CH4, and N2O emissions can be especially complex in cropping and grazing systems. This systematic review, however, will focus on carbon measurements and make recommendations that future systematic reviews should look at assessment methods for other GHG emissions. Agriculture has the potential to mitigate between 5.5 and 6 GtCO2e/yr [103], which means that there will be a need to assess carbon accurately and reliably to comply with international monitoring requirements or trading schemes.
Smith et al. review current methods available for measuring components of the carbon budget in croplands, and examine some of the tools used for scaling the carbon budget of croplands from ecosystem to continental levels [104]. They highlight the complex nature of the evidence that needs to be gathered at each site before a full carbon budget can be made. They also point at the “impossibility” of up-scaling results from sites to the continental level (Europe is the focus in their synthesis), even with an extensive network of sites, because of the lack of spatially explicit information on management and soil for agricultural systems. However, they point out the value of such site-specific data for calibrating and validating ecosystem models for continental projections.
Using remote sensing to estimate carbon stocks
Interest in the possibility of using remove sensing to determine carbon stocks in terrestrial systems has been growing in recent years for example, [54, 70, 105–108]. Remote sensing (space-borne or air-borne) usually provides continuous spatial information over landscape-size areas (size depends on sensor characteristics) in contrast to field inventory where information is generally limited to plots or small areas. Carbon stocks can be estimated by applying carbon density values from ground data or national forest inventories across land cover/vegetation maps obtained by remotely-sensed data. Spatial vegetation information from optical satellite sensors can be related to ground-based measurements to estimate carbon stocks. Direct measurements of AGB are limited to small forest areas, because site-specific allometric equations cannot be generalized for a forest or region and space-borne instruments cannot measure tropical forest biomass directly. The use of space-borne radar backscatter data is becoming popular as a method for estimating woody biomass over large areas in the tropics because of its capability of penetrating through the forest canopy and all-weather acquisition.
Published studies very often use national forest inventory data to verify results of remote sensing estimates of carbon. Many claim to show strong correlation. However, limitations are reported in the literature, in particular the weak, or absent, relationship between radar backscatter and AGB associated with saturation, and errors in geo-location: for example, old Global Positioning System (GPS) instruments used in constructing inventories may introduce uncertainty in establishing the ‘centre of plot’ location, compass direction, etc. [52].
There are a number of approaches to estimating AGB at larger spatial scales with remote sensing data by extrapolating those obtained from field plots. Such methods include multi-stage sampling, multiple regression analysis, non-parametric k-nearest neighbour technique (k-NN), neural networks, or indirect relationships between forest attributes, determined by remote sensing, and biomass. An increasing number of studies use fine resolution imagery such as QuickBird, a high-resolution commercial earth observation satellite, launched in 2001, aerial photographs or IKONOS, a commercial earth observation satellite, which launched in 1999 to collect publicly-available high-resolution imagery at 1- and 4-metre resolution, for modelling tree parameters or forest canopy structures, though these are not applied to large areas owing to cost and technical demand. Medium spatial resolution imagery such as Landsat has been widely in use since 1972. Where optical sensors have limitation, radio detection and ranging (radar) and light detection and ranging (LiDAR) data are being used.
Most studies on AGB estimates have not provided accuracy or precision assessments with respect to ground data [109]. Rosenqvist et al. undertook a qualitative review of remote sensing techniques for use under the Kyoto Protocol but did not provide an assessment of their operational status for use at national scales [110]. For the UK and countries with similar reporting requirements, Patenaude et al. made quantitative assessments of the accuracy and comparative costs of optical, radar and LiDAR techniques for reporting deforestation through land-cover classification analyses and quantification of forest above-ground carbon stocks [111].
The accurate assessment of above-ground forest biomass and carbon stock over large areas requires a grid of ground sample plots (with very precise location or a nested sampling) together with a map of vegetation types and/or cover classes. Classification and mapping can be done either from ground sampling or on the basis of satellite imagery or aerial photography. More precise vegetation classification and a denser network of sample plots would give more precise estimates at higher costs.
The need for a multi-disciplinary systematic review of terrestrial carbon
It is clear that a wide range of efforts have been and are being undertaken in public and corporate research to provide methods and data for carbon stock assessments in different pools. There is a huge body of knowledge collected over decades. There has been a proliferation of scientific and technical papers, but monitoring of forests is still ‘insufficiently accurate or precise for an international protocol that would administer finances based on monitoring results of forest area or forest carbon storage’ [112], and there is no reason to suppose that the situation for other pools is any better. The adequacy of current or potential systems for reliably assessing carbon stocks at national, regional or local levels (under the REDD+ framework or elsewhere) has not been systematically evaluated, nor has the scientific underpinning of these approaches been properly examined. It has been argued that a REDD+ system must allow and account for variability in methodologies and accuracy. The latter is inevitable with such wide differences between countries and assessment methods, but flexibility must come with knowledge of the limits of confidence in these variable approaches if REDD+ is to be credible, transparent and fair.
There is clearly a need to critically review the accuracy and precision of various remote sensing techniques against ground observation and among methods, and their applicability in geographically varied regions.
Question development
At an initial workshop held in FAO headquarters in March 2009, it was agreed that an international participatory initiative should be undertaken to scope the potential for using an evidence-based approach to validate the knowledge base on carbon monitoring. A project contact group of 50 people was chosen to represent a range of organizations with academic, policy, consultancy and/or training focus in different countries.
The contact group participated in iterative discussions (mostly by email and telephone) to develop a shared understanding of the problems which need to be addressed and to develop a series of possible review questions. In addition to discussions the group shared knowledge, mainly published papers but also ‘grey literature’ (project reports and discussion documents), which contributed to an understanding of the potential scope of the review question. The review questions were further developed by two scoping groups (one meeting in person and the other via a teleconference) during the Bonn climate change meetings in June 2009.
Review authors met in November 2009 at a workshop in FAO, Rome for three days to frame the review questions precisely.
The broad review question and three sub-questions agreed after these extensive discussions were further clarified as follows:
Broad question:
“How do current methodologies compare in their ability to measure and assess terrestrial carbon stocks and changes in carbon stocks with accuracy and precision?”
Where accuracy is a relative measure of the exactness of an estimate against true values, precision is the inverse of uncertainty with a measurement or estimate (e.g., the standard error of the sample mean). The term “methodologies” includes methods (including direct measurements, sampling design, remote sensing and models) and systems that aggregate methods to measure and assess carbon stocks.
Sub-questions:
-
1.
“How accurate and precise are methodologies used for the conversion of in situ measurements into carbon stock estimates at the plot or site level?”
The term methodologies includes direct measurements of variables in the field (in situ) and methods that convert them into carbon stock estimates at the site level. “Site” refers to sample or assessment plot. This question also looks at the geographical validity of methodologies developed at the site-level and examines the applicability of methodologies to different land use categories in different environments, ecosystems and countries.
-
2.
“How accurate and precise are methodologies for generating carbon stock estimates for larger geographical areas (landscape level) from site-level data?”
The term “landscape level” encompasses the spatial scales from site to sub-national and national levels through forest inventories, stratification, other sampling schemes and modelling. This question also looks at sampling and stratification by remote sensing and examines methodologies which attempt to assess changes in soil carbon with land use conversion.
-
3.
“How accurate and precise are direct remote sensing methodologies for estimating carbon stocks?”
This question includes carbon stock estimates from direct measurements of variables by remote sensing instruments, coupled with field measurements and methodologies to convert measurements into stock estimates. Ground-based measurements such as terrestrial LiDAR will be included, and field biomass components will be identified from the source material where this information is available.
The initial set of questions included the concept of repeatability, in addition to accuracy and precision. Repeatability is a measure of the practical aspects of using a given method to assess carbon, including cost, and human and/or technical capacity. At an early phase of the review during pilot literature searches, however, it became clear that the concept ‘repeatability’ was not easy to define in a way that would not introduce unnecessary bias into the review, and seriously reduce the number of papers available for analysis of the more immediately critical questions relating to accuracy and precision. The pragmatic decision was taken to remove this element of the review question (and sub-questions).
Objectives
The primary objective of this systematic review is to compile the evidence relating to the question:
How do current methodologies compare in their ability to measure and assess terrestrial carbon stocks and changes in carbon stocks with accuracy and precision?
The three secondary questions identified aim to focus the research and reporting of the review:
-
1.
How accurate and precise are methodologies used for the conversion of in situ measurements into carbon stock estimates at the plot or site level?
-
2.
How accurate and precise are methodologies for generating carbon stock estimates for larger geographical areas (landscape level) from site-level data?
-
3.
How accurate and precise are direct remote sensing methodologies for estimating carbon stocks?