Royal Society Publishing


On average, about 45 per cent of global annual anthropogenic carbon dioxide (CO2) emissions remain in the atmosphere, while the remainder are taken up by carbon reservoirs on land and in the oceans—the CO2 ‘sinks’. As sink size and dynamics are highly variable in space and time, cross-verification of reported anthropogenic CO2 emissions with atmospheric CO2 measurements is challenging. Highly variable CO2 sinks also limit the capability to detect anomolous changes in natural carbon reservoirs. This paper argues that significant uncertainty reduction in annual estimates of the global carbon balance could be achieved rapidly through coordinated up-scaling of existing methods, and that this uncertainty reduction would provide incentive for accurate reporting of CO2 emissions at the country level. We estimate that if 5 per cent of global CO2 emissions go unreported and undetected, the associated marginal economic impacts could reach approximately US$20 billion each year by 2050. The net present day value of these impacts aggregated until 2200, and discounted back to the present would have a mean value exceeding US$10 trillion. The costs of potential impacts of unreported emissions far outweigh the costs of enhancement of measurement infrastructure to reduce uncertainty in the global carbon balance.

1. Introduction

The UN Climate Change Conference 2009 (COP15) in Copenhagen, Denmark, failed to reach agreement on binding cuts in greenhouse gas (GHG) emissions, in part owing to the requirement of compulsory reporting of anthropogenic GHG emissions by all countries and the difficulty in verifying the declared emissions. Carbon dioxide (CO2) is the largest contributor to global GHG emissions, and its relative contribution is growing [1]. Only about 45 per cent of the global emissions of CO2 from the burning of fossil fuel, cement production, deforestation and other land use change (LUC) remain in the atmosphere on average [2], while the remainder are taken up by the land and ocean carbon reservoirs—the CO2 ‘sinks’.

The sizes of the CO2 sinks are variable and current estimates have a large uncertainty [3]. Elevated atmospheric CO2 and warming of the climate system also influence exchanges of CO2 between surface reservoirs and the atmosphere [4]. Feedback between climate change and the natural carbon reservoirs could lead to large, abrupt and/or irreversible (on 1000 year time scale) changes in carbon reservoirs, and further increases in surface temperature [57]. Possible consequences of climate change include: failing water supplies in arid regions; flooding of coastal communities owing to sea-level rise; increased wildfires; increase in the number and intensity of extreme surface temperature fluctuations; all of which present high cost to society [8]. It follows that policy- and decision-makers need reliable estimates of the threshold at which possible abrupt feedbacks in the Earth system may occur, so monitoring of the behaviour of CO2 sinks is essential.

In theory, it should be possible to independently verify CO2 emissions using a mass balance approach (CO2 emissions = atmospheric CO2 growth + CO2 sinks). However, the present uncertainty in the estimates of the sinks is too large to provide independent constraints on global emissions, and uncertainty concerning atmospheric transport compounds this problem at finer regional-scale spatial resolution [9]. At the same time, uncertainty in CO2 emissions is too large to enable early detection of unexpected feedbacks in the CO2 sinks, if these were to occur. Thus, a reduction in the overall uncertainty in the global CO2 budget would improve both the capability to verify reported CO2 emissions, and the capability to detect unexpected feedbacks in the CO2 sinks.

This article aims to assess our ability to quantify the dynamics of the global carbon cycle to a minimum level necessary to independently constrain reported CO2 emissions. Additionally, improved emission constraints will increase knowledge of flux responses to climate forcing and will improve our capability to detect climate–carbon feedbacks. The feasibility of uncertainty reduction in the carbon cycle budget and the capital costs of implementing systems to reduce the uncertainty are also explored. In this paper, we assume that improvements in the closure of the global carbon balance will incentivize accurate reporting of CO2 emissions at the country level, and thus constitutes the minimum requirement considered in this paper. Information on the regional carbon balance or measurements to resolve regional CO2 emissions indirectly would provide stronger constraints on regional emissions [10], but these are not considered here except where necessary to reduce uncertainty in the global carbon balance.

2. Components of the global carbon cycle and current sources of uncertainty

The contemporary global CO2 budget may be broken down into five components (table 1 and figure 1): (i) fossil fuel combustion and cement manufacture-related emissions (FFE) result from the combustion of hydrocarbons extracted from the Earth’s crust, with small contributions (approx. 3%) from cement manufacture and gas flaring; (ii) emissions from LUC result from deforestation and anthropogenic management of terrestrial ecosystems; (iii) atmospheric growth rate (AGR) is the proportion of CO2 added to the atmosphere from anthropogenic CO2 emissions (including oxidation of reduced carbon species such as methane) minus the CO2 removed by the global sinks; (iv) ocean carbon dioxide uptake (OCU) is the net flux of CO2 into the oceans and is primarily a response to the AGR [11]; and (v) terrestrial carbon dioxide uptake (TCU) occurs when ecosystem rates of net primary productivity exceed rates of carbon loss, e.g. through respiration and fire, leading to a net increase in carbon storage. Inter-annual variability (IAV) in atmospheric CO2 concentration is driven primarily by variability in TCU, with smaller contributions from variability in FFE (linked closely to global economies), OCU and LUC [3].

View this table:
Table 1.

Summary of the uncertainty associated with each component of the global CO2 budget and potential for improvements.

Figure 1.

Components of the global CO2 budget (Gt C yr−1). (a) CO2 emissions from fossil fuel combustion and cement production (FFE) and (b) from land use change (LUC), (c) the atmospheric CO2 growth rate (AGR), (d) land CO2 sink (LCU; negative = land uptake), and (e) ocean CO2 sink (OCU; negative = ocean uptake). The land and ocean sinks are shown as an average of several models normalized to the mean ocean and land sinks estimated from observations for 1990–2000. The shaded area is the uncertainty associated with each component. See methods in Le Quéré et al. [11] for the sources of data and the text for an explanation of the associated uncertainty. (Adapted from [11].)

Uncertainty in FFE is ±0.5 Gt C yr−1 (i.e. ±6% of the mean for ±1σ) which stems from inadequate accounting infrastructure and methodologies, and non-standardized reporting; for example, the absence or poor quality of statistics on the rates at which reserves are exploited, stored and utilized, and emission intensities for various fuel types [12]. Non-transparent and non-standardized reporting of emissions estimates are particularly large in fast-growing countries in non-Annex B nations [13]. This uncertainty is increasing through time in absolute terms because both FFE and the share of global emissions of emerging economies are growing (figure 2). Uncertainty in quantification of the AGR is low at 0.15 Gt C yr−1 (approx. ±4% of the mean value). This is mainly owing to the fast mixing rate of the atmosphere, which allows the global average concentration to be well characterized with a sparse but strategic high-precision observational network.

Figure 2.

Uncertainty in the components of the global CO2 budget (Gt C yr−1). (a) CO2 emissions from fossil fuel combustion and cement production (FFE) and (b) from land use change (LUC), (c) the atmospheric CO2 growth rate (AGR), (d) land CO2 sink (LCU), and (e) ocean CO2 sink. See text for the sources of uncertainty.

Uncertainty in OCU is approximately ±0.5 Gt C yr−1 (i.e. ±22% of the mean value) which results mainly from uncertainty in the estimates of the long-term mean sink. Regional and global ocean CO2 fluxes and carbon stocks may be estimated indirectly through measurement of atmospheric oxygen [14], CFCs [15], and through oceanic biogeochemical measurements [16,17]. These methods provide a combined uncertainty of ±0.4 Gt C yr−1 for the 1990s, as reported in the IPCC Fourth Assessment Report [3]. Combined estimates from ocean biogeochemical models have been used to quantify the additional uncertainty in OCU resulting from limited knowledge of IAV [11].

Uncertainty in LUC is approximately ±0.7 Gt C yr−1 (i.e. ±50% of the mean value) which results from uncertainty in the size of carbon stocks, and rates of deforestation and conversion of land for agricultural production or pasture, in some cases followed by later abandonment [1820]. Uncertainty is influenced by current limitations in methodological approaches (especially satellite techniques) to quantify LUC over time.

We estimate the uncertainty in TCU at approximately ±1.6 Gt C yr−1 (i.e. ±59% of the mean value) as explained below. There are no methods to estimate TCU directly from in situ measurements. The sum of TCU and LUC can be estimated from measurement of O2 in the atmosphere [14], which provides an uncertainty for TCU decadal mean of ±1.0 Gt C yr−1 when combined with LUC uncertainty. The variability in TCU could be most accurately estimated from the other terms in the carbon budget (FFE+LUC−AGR−OCU), but here we are primarily interested in quantifying uncertainty in TCU estimated independently from the other terms in the carbon balance. We used an ensemble of five global ecosystem models (e.g. [21]) to quantify the additional uncertainty in TCU caused by our limited knowledge of IAV (maximum mean absolute deviation for 1959–2008 of ±1.3 Pg C yr−1 [21]). The large spread in the model estimates results from a low level of understanding of how rates of photosynthesis and respiration vary with soil moisture, atmospheric CO2 concentration and temperature, and in turn, how the terrestrial carbon sink responds to rising CO2 and temperature on longer time scales.

3. Feasibility for uncertainty reduction

This section summarizes existing methods that can be used to improve the estimate of each component of the carbon cycle and assesses the respective potential for uncertainty reduction in their mean and IAV, time scale of implementation, and associated economic and societal costs. We provide qualitative estimates only based on our own judgement, where low effort refers to efforts to maintain infrastructure of measurements that are already in place, medium efforts require an expansion of existing infrastructure that can be done by existing institutions, and high effort requires major expansion or the development of new infrastructure beyond existing capacity (table 1). We relate the potential uncertainty reduction to our assessment of the approximate cost above current investments in order to identify the most efficient strategy.

Reduction of uncertainty in FFE requires improved measurement and accounting infrastructure, particularly in developing countries; furthermore, the degree of under-reporting may increase once a global carbon taxation system is in place [22]. Measurement of radiocarbon (14C) can distinguish FFEs of CO2 from non-FFEs and sinks, and may provide a new tool that could reduce the associated uncertainty in atmospheric model inversions [10]. The overall effort is assumed to have a high economic cost (table 1) and will be challenging in terms of societal costs. Political commitment and restructuring of internal energy consumption/production accounting procedures would be required. However, the restructuring will need to occur only once and subsequent recurrent annual costs would be lower.

Improving understanding of the consequences of LUC, such as deforestation, for carbon stocks is challenging, especially on the global scale. Reducing LUC uncertainty requires improved land conversion statistics [23] and greatly improved understanding of carbon allocation to different pools following conversion, for example, quantification of the amount of carbon released to the atmosphere during biomass burning and wildfires [24]. We assume that restructuring of internal reporting statistics is costly. Dedicated satellite sensors are expensive, but they can measure changes in vegetation cover over regional scales, as is already routinely established in Brazil [25,26] and other regions [27].

Improved conversion statistics would resolve only part of the problem. Quantification of carbon stocks using bottom-up methodologies (which accounts for vegetation density and degradation) requires forest and soil inventories to be compiled every 10 years [28]. High resolution measurements of carbon stocks and fluxes are needed over both short and long time scales, for example, to determine the impacts of fire and the variability of carbon stocks, respectively. Physical challenges include measurement of soil carbon to depths of several metres, over highly variable terrain, often in remote locations. While the cost of such inventories is low to medium, the human effort is medium as the measurements are labour intensive. Inventories and their associated costs would need to be repeated each decade, while the reporting statistics would need to be maintained on an annual basis.

Regarding the partitioning of CO2 emissions, as previously described, the global AGR is known with very low uncertainty and requires only maintenance of the current network. The value of enhanced regional network would come into play when regional CO2 budgets are required and to validate land and ocean models, but not directly for global budgets. Reducing uncertainty in OCU from observations directly would be possible, both for the long-term mean and for the IAV, particularly through the expansion of repeated pCO2 measurements from ships of opportunity [29]. Expanding the current network comes at low cost and low effort. One ship line costs approximately US$150 000 per year and sufficient coverage to recover IAV in the North Atlantic and Pacific [29] could be achieved at approximately US$2–3 million yr−1 above current investments (A. Watson 2009, personal communication). The Northern Hemisphere observation-based estimates can be complemented with inversion analysis of atmospheric CO2 concentrations over the Southern Ocean (e.g. [22]).

At present, process-based and inversion models of ocean carbon fluxes provide the only method to estimate the IAV in the global ocean CO2 sink. There is no systematic effort to validate the models and only a few models (four to six) produce annual estimates of the global CO2 sink [11]. Such a small model ensemble is too limited to encompass all possible sources of errors. Improvements in models and data/model integrations through systematic validation would come at low cost and low effort. Nevertheless, the potential for uncertainty reduction in OCU is relatively small, and may be realistically limited to reduce the uncertainty to about ±0.3 Gt C yr−1, based on the formal analysis of Watson et al. [29] for the North Atlantic and the difficulty to expand the data coverage in remote regions.

It is critical to improve estimates of TCU through approaches that are independent of emissions data. Direct measurement of changes in biomass and soil carbon content is labour intensive and impractical on the ecosystem scale, so quantification will rely on indirect approaches. Evaluation of carbon fluxes and the calculation of TCU may be improved through the refinement of atmospheric inversions of measured CO2 concentration. Improvements in models and data/model integrations through systematic validation, particularly with regional patterns based on inversions of atmospheric CO2 and flux tower data or though formal data assimilation systems, would provide constraints on the model IAV by allowing for a better representation of the regional response of productivity and respiration to warming and changes in water availability. Whereas it is challenging to determine how such constraints would reduce the uncertainty in IAV, there is a wealth of information that could be used to constrain model estimates and reduce uncertainty in IAV by up to half its current value (from ±1.3 Pg C yr−1 to approx. ±0.7 Pg C yr−1). The uncertainty in total TCU (mean+IAV) would decrease from ±1.6 to approximately ±1.2 Pg C yr−1, further if the uncertainty in LUC and O2 detection method can also be improved. Inherent limits in observational coverage and associated uncertainty would preclude further uncertainty reduction. Efforts to improve IAV estimates would come at low cost and low effort as long-term datasets and models are mostly in place.

When the sum of the possible uncertainty reductions in the partitioning of OCU and TCU is considered, the combined uncertainty decreases from ±1.7 to ±1.2 Gt C yr−1. This is close to the uncertainty in total emissions, which is currently ±0.9 Gt C yr−1 and increasing. Such significant uncertainty reduction in annual estimates of the global carbon partitioning would provide incentive for accurate reporting of CO2 emissions at the country level, as inadequate reporting, if it was widespread, would lead to inconsistencies between the reported emissions and the estimated partitioning, and in turn, lead to further questioning of the available data on a regional scale. It follows that reduction below a threshold uncertainty would be feasible at low cost and low effort, but as it is primarily based on model improvements that would be constrained by enhanced observations and understanding, major coordinated effort across the carbon cycle community and sustained long-term observations would be required.

4. Financial and human benefits of reducing current uncertainty

In this study, the economic benefits of implementing an effective monitoring system were evaluated using the PAGE09 (Policy Analysis of the Greenhouse Effect) model, which is an updated version of the PAGE2002 integrated assessment model [3032]. PAGE2002 was previously used to perform calculations of the economic impacts and social cost of anthropogenic CO2 emissions for the Stern review [8] and the Asian Development Bank’s review of climate change in Southeast Asia [33]. PAGE09 applies simple parametrizations of global climate and financial systems, and accounts for the uncertainty in scientific and economic understanding, outlined primarily in the Fourth Assessment Report of the IPCC [34]. The description includes representation of carbon cycle feedbacks and sea-level changes, and assumptions about climate sensitivity based on the current literature, i.e. the global temperature rise in response to a doubling of CO2 (mean of 2.6°C; 5–95% range corresponds to 1.6–4.2°C with a maximum up to approx. 7°C). Positive impacts, e.g. on terrestrial ecosystem productivity owing to CO2 fertilization, are also included. The model explicitly calculates the dependence of climate-related impacts on gross domestic product per capita and as changes in expected utility, and accounts for the impacts of nonlinear and transient carbon cycle feedbacks resulting from increased global warming. The PAGE09 calculated impacts do include the possibility of economic benefits for small temperature rises and can also account for the economic burden of developing non-carbon-based infrastructure (although we do not calculate the latter). The results are reported as the total global net present value (NPV) of impacts in US$ in year 2005, which is a parameter that aggregates the computed losses owing to climate change over the future two centuries, and discounts it back to the present day.

The potential effects of underestimates in national emissions reporting (intentional or non-intentional) were evaluated relative to the cost of implementing effective monitoring infrastructure. A series of model scenarios were evaluated (figure 3): (i) fully effective monitoring infrastructure that measures emissions with zero uncertainty; (ii) current monitoring infrastructure which provides estimates of FFE with uncertainty of ±6.5 per cent (table 1); under this condition it was assumed nations emit an additional 5 per cent more than reported; (iii) fully effective monitoring infrastructure in place by 2020; nations assumed to emit an additional 5 per cent more than reported until 2020; and (iv) the same as (iii) but unreported emissions stop in 2030. All regions were assumed to sign up to quantitative targets of a 30 per cent decrease in CO2 emissions from 2008 levels for Annex 1 countries, and a 30 per cent increase from 2008 levels for non-Annex 1 countries.

Figure 3.

Projected drop in economic losses from improved emissions monitoring if cheating is eliminated immediately, by 2020 or by 2030. All regions were assumed to sign up to quantitative targets of a 30% decrease in CO2 emissions from 2008 levels for Annex 1 countries, and a 30% increase from 2008 levels for non-Annex 1 countries. Three scenarios are shown: (blue) drop in losses if there is no cheating and nations emit as reported; (green) under current monitoring infrastructure, it is assumed that nations under-report and globally emit an additional 5% of CO2 until the implementation of a fully effective monitoring infrastructure by 2020, after which emissions are as reported; and (red) as before if the fully effective monitoring infrastructure is in place by 2030. The shaded areas show the 25–75% confidence intervals.

The total global NPV of climate change impacts associated with FFE, aggregated from present until 2200, and discounted back to the present day, is just under US$200 trillion (5–95% range is about US$30–670 trillion). Under scenario (ii), regions emit an additional 5 per cent more CO2 than reported which leads to extra economic losses: the mean extra annual impact increases from about US$2 billion in 2010 to nearly US$20 billion in 2050 (the 5–95% range is from about US$2 billion to over US$45 billion in 2050). The mean NPV of these losses aggregated from today until 2200 and discounted back to the present day is just over US$10 trillion, reflecting much greater annual losses in the second half of this century, and throughout the twenty-second century. If improved monitoring is introduced so that 2030 is the last year in which regions can ‘cheat’ on emissions reporting, then the mean drop in losses is predicted to increase from about US$0.5 billion in 2040 to US$4 billion in 2050, compared to the case with no new monitoring infrastructure (figure 3). The drop in losses increases greatly after 2050 (not shown) to about US$400 billion in 2200. The mean NPV of the impacts decreases by about US$7.5 trillion, which corresponds to about 75 per cent of the impacts from cheating on reporting if there is no addition to monitoring infrastructure. However, if monitoring infrastructure is improved so that 2020 is the last year in which regions can cheat on reporting, the drop in losses is greater, increasing from about US$3 billion in 2040 to US$8 billion in 2050. The mean NPV of the aggregated impacts decreases by about US$8.5 trillion, which is about 85 per cent of the losses related to cheating in the absence of new monitoring. In both cases, the annual mean drop in losses increases greatly after 2050, to greater than US$400 billion in 2200.

It follows that the benefits of improved monitoring (i.e. which will act to mitigate the cost of potential impacts) far outweigh the costs of enhancing monitoring infrastructure by orders of magnitude. Investment in monitoring is necessary now to reduce the uncertainty and limit the economic impacts from climate change related to uncertainty in emissions reporting.

5. Conclusions

We reviewed the sources of errors in the various components of the global CO2 budget, and analysed the potential to reduce the errors to a level necessary to provide independent constraints on reported CO2 emissions. Useful uncertainty reduction can be achieved with existing methods through a globally coordinated long-term effort at low cost and low effort. This basic uncertainty reduction would be sufficient to identify incoherence in reported emissions and/or unexpected behaviour in the carbon sinks, but it would not provide further indications of the origin of the problem. Further reductions in uncertainty necessary to constrain the CO2 emissions at the regional level require a far greater expansion of observations and models, which includes the refinement of atmospheric inversion techniques, additional space-based sensors and improved data assimilation methods; this comes with an associated high cost and high effort [28,35]. Our analysis is in line with a more detailed study by Pacala et al. [10] who conclude that expanded measurement infrastructure and improved inventories (focused carbon stocks in soils) are required to reduce uncertainty on FFE at the regional level, especially in developing countries. The cost of reducing uncertainty in global carbon balance with current methods is extremely low in comparison to the potential financial impacts of undetected and unreported FFEs, which could reach close to US$20 billion annually by 2050.


A.J.D. acknowledges the Natural Environment Research Council for support during the writing of this manuscript. Development of the PAGE09 model received funding from the European Community’s Seventh Framework Programme, as part of the ClimateCost Project (Full Costs of Climate Change, grant agreement 212774; www/, and from the UK Department of Energy and Climate Change. A.D.F. received funding from the European Community’s Seventh Framework Programme (FP7/2007–2013) under grant agreement no. 238366. The authors thank Ralph Keeling and an anonymous reviewer for thoughtful and constructive reviews that improved the manuscript, and Euan Nisbet for editorial coordination.



View Abstract