Carbon mitigation projects and field interventions
To calculate OARs, we prioritised our search terms to cover the largest project types issuing credits through independent mechanisms and the Kyoto Protocol’s CDM and JI (Table 1 and Supplementary Tables 2 and 3).
Credit issuance is concentrated among the Kyoto Protocol’s crediting mechanisms as well as several governmental, private, and non-governmental mechanisms (which we collectively refer to as ‘independent mechanisms’) (Fig. 1a) and a few sectors (Fig. 1b). The CDM and the JI Mechanism have jointly issued 63% of credits (3.3 gigatons)3,28, whereas independent mechanisms are responsible for 37% of issued credits (1.9 gigatons)29. Credits from chemical processes as well as industrial manufacturing projects were mainly issued under the Kyoto mechanisms, whereas credits from forestry and land use, as well as household and community projects, were mainly issued under independent mechanisms. Projects from renewable energy constitute 29% of the issued credits across these crediting mechanisms. Industrial manufacturing and chemical processes and account for 24% and 22%, respectively. Forestry and land use account for 15%, whereas waste management and household and community account for 5% and 3%, respectively. Domestic crediting mechanisms are excluded from this overview as these only constitute a minor fraction of issued credits1. Figure 1 displays historical averages and, therefore, current issuance volumes might differ.
a Total issuance in gigatons CO2 across the Kyoto Protocol’s two project-based mechanisms (Clean Development Mechanism and Joint Implementation) and the four major independent mechanisms covered by the Berkeley Carbon Trading Project Database29 (American Carbon Registry (ACR), Climate Action Reserve (CAR), Gold Standard (GS) and Verified Carbon Standard (VCS)). b Total issuance in % across different sectors under Kyoto and independent mechanisms. Data is based on Clean Development Mechanism’s (CDM) database for PAs and PoAs28, the United Nations Environment Programme (UNEP) RISOE database3 for JI and the Berkeley Carbon Trading Project Database v929. Other crediting mechanisms are excluded as they only constitute a minor share of issued credits1. The sector ‘other’ contains carbon capture and storage, agriculture and transportation.
Drawing upon the typologies of the Berkeley Carbon Trading Project30, the CDM28, UNEP DTU31 and the Carbon Credit Quality Initiative11, we classify each of the 65 studies in our assessment into one of seven sectors and one of the 21 project types listed in Table 1. We differentiate between studies investigating carbon crediting projects and studies investigating similar field interventions that were implemented without issuing carbon credits (which we refer to as field interventions). We found 14 studies investigating 2346 carbon crediting projects across six project types (Fig. 2a; please note that the sector forestry and chemical processes contain two project types, respectively) and 51 studies investigating field interventions without issuing credits with a total of 1.2 million observations (Fig. 2b). For the other three sectors (waste management, industrial manufacturing and carbon capture and storage), we could not find any studies investigating field interventions and carbon crediting projects that matched our inclusion criteria (see Supplementary Table 1). Overall, we find the strongest concentration of carbon project evaluations in the forestry sector, with equal distribution across the other sectors (Fig. 2a).
a/b Distribution of studies across project types, (c/d) across regions and (e/f) methodology types. Note: k refers to the number in thousands, and m refers to the number in millions. See Supplementary Tables 4 and 5 for a full list of studies. Note: IFM refers to improved forest management. If the same project was evaluated by multiple studies, we count each as a separate project evaluation as the time frame, methodology and other relevant factors might differ. The total project number includes studies that could not be integrated into our quantitative framework (e.g. ref. 18) but are still discussed in the relevant sections.
Studies on carbon credit projects are generally split between different geographies (Fig. 2c); Africa is an exception, with no studies focused solely on the continent but covered in three studies that evaluate multiple geographies. Similarly, most field interventions focus on forestry mainly in Latin America, as most forestry projects have been implemented in the Amazon region (Fig. 2d). Overall, studies of both carbon-crediting projects and field interventions available in the literature mainly rely on rigorous observational studies (Fig. 2e, f). In contrast to randomised controlled trials (RCTs) (in which the experimenter assigns treatment)17, rigorous observational studies build a plausible control group to estimate project impacts8. Only 9 of 65 studies were based on RCTs (mainly evaluating the impact of fuel-efficient cookstoves, with one study in forestry32).
The offset achievement ratio across project types
Carbon project developers quantify emission reductions in line with standards and methodologies developed by carbon crediting mechanisms such as the Verified Carbon Standard by Verra. Following an audit by an accepted third party, carbon credits are issued into a registry8. Yet, these standards and methodologies vary in their robustness and often allow for activities to be credited that would have happened regardless of the offset programme2,23, and provide flexibility to project developers to select methodological approaches and data that maximise credit issuance6,9. It is, therefore, critical to contrast the emission reduction estimates used to determine credit issuance to those achieved based on rigorous academic assessments.
We introduce the term ‘offset achievement ratio’, which compares studies’ quantitative estimates of carbon crediting projects’ emission reductions with those made by project developers to generate carbon credits. An OAR of 50% indicates that the academic literature estimates that only half of the emission reductions claimed by project developers—and issued as carbon credits—were likely achieved. We complement these quantitative estimates with qualitative discussion of other studies including other qualitative and quantitative studies of the quality of offset methodologies and studies that assess field interventions that did not issue carbon credits but may still hold important insights on additionality, conservative quantification, or other relevant factors.
To quantify the OAR, we rely on academic studies that evaluate voluntary, project-based activities that seek to reduce emissions or enhance removals (see Supplementary Table 1 for inclusion and exclusion criteria). We excluded studies that evaluate non-voluntary activities such as mandatory regulations or non-project-based activities (e.g. other forms of carbon pricing such as carbon taxes). We focus on studies that evaluate project impact against a credible comparator. This comparator can include projects, land, or households that were not part of the carbon crediting projects4,5,7,8,17,21; this can include historical data of the same project before it became a carbon crediting project15,16. The comparator can also be values from the scientific literature6. For example, some studies compare individual factors used by carbon crediting projects, such as the share of users that adopt a fuel-efficient cookstove, against the body of knowledge in the published literature6. Studies must also include a quantitative assessment of greenhouse gas emission changes or a comparable environmental metric, such as deforestation rates7,8. Lastly, we only include studies that use RCTs or rigorous observational data (which construct a plausible control group8 or science-based comparator6 to estimate project impacts). The included studies fall into several categories: peer-reviewed articles17, papers aimed at peer-reviewed journals (e.g. working papers)18 and chapters in PhD theses19, which also undergo an academic examination process. We exclude qualitative studies from our quantitative assessment.
Our assessment considers additionality and conservative quantification in determining the OAR. The latter encompasses project, baseline and leakage emissions. Figure 3 illustrates which of these issues have been addressed by the 14 studies on carbon crediting projects that were considered in determining the OAR. Not all studies address all factors that affect a particular source of over-crediting. For instance, Aung et al.17 studied the impact of fuel-efficient cookstoves on firewood usage in households that received the stove and those that did not (i.e. project and baseline emissions). However, the authors do not address other over-crediting factors related to the project emissions and baseline, such as the fraction of non-renewable biomass used to compute credit issuance. In contrast, Gill-Wiehl et al.6 cover all relevant factors relating to over-crediting from baseline and project emissions.
The name of the authors shows the study, which analysed the specific source of over/under-crediting, otherwise, the box indicates ‘No’. The figure excludes several studies that analyse offset quality, namely Calel et al.18 Badgley et al.33 Holm et al.20 and Bomfim et al.45, because they could not be integrated into our quantitative assessment framework, but the findings are reviewed in the discussion section. Reasons for exclusion for each of these studies can be found in Supplementary Table 7. Only the first author’s name is shown due to space constraints. Note that the figure does not cover the field interventions as these did not issue carbon credits and, therefore, could not be integrated into our quantitative framework. IFM refers to Improved Forest Management.
Offset achievement ratio across project types
Overall, we find that carbon-crediting projects achieved considerably lower emission reductions than the number of credits issued to the projects (Fig. 4). We find the lowest OARs in wind power in China and improved forest management (IFM) in the United States, for which no statistically significant emission reductions were documented in the studies (we, therefore, assume an OAR-value of 0% for these projects; see Eq. (1) in the Methods section, as well as Supplementary Table 6 detailing the exact numbers used to calculate the OAR across and within project types). These project types are followed by cookstoves (10.8%), SF6 destruction (16.4%), avoided deforestation (24.7%), and HFC-23 destruction (68.3%) (Fig. 4a). Project-level results (Fig. 4b) show that individual projects may (over-) deliver relative to the issued credits, but the vast majority underachieves relative to the volume of issued credits. For our estimates in Fig. 4a, b, we use the central estimates from the studies. The source data are provided in this paper.
a Estimated offset achievement ratio for project types for which we found relevant studies. b Project-level estimates extracted from relevant studies for individual projects. Only the first authors are mentioned in the study section for visualisation purposes. Two studies excluded that only showed upper-bound estimates refs. 18,35 and two studies that could not be integrated into the framework20,45; see Supplementary Table 7 for detailed exclusion reasons. Estimated average offset achievement ratios in (a) are the weighted average of projects’ OAR based on the issued credits (i.e. projects that issued more credits are weighted more). Each dot in (b) represents one project-level OAR, with the colours corresponding to the underlying studies. Confidence intervals in (a) are the weighted variance (by credit issuance) based on the individual estimates in (b), whereas the centre of the error bar represents the weighted average for each project type. The exact issued credits and the OAR for each project can be found in the supplementary data. ACR refers to the American Carbon Registry, CAR to Climate Action Reserve, CDM to Clean Development Mechanism, VCS to Voluntary Carbon Standard, GS to Gold Standard, JI to Joint Implementation and IFM to Improved Forest Management.
The offset achievement gap
The studies in our assessment cover projects that are responsible for 19% of carbon credits issued across the main international and independent carbon crediting mechanisms (Fig. 5a). Using the OAR estimates, we find that of the 972 million credits issued across the covered project types, 812 million likely do not constitute real emission reductions (Fig. 5b). This offset achievement gap is larger than Germany’s annual emissions. The largest source of non-achieved credits stems from avoided deforestation, wind power and IFM (Fig. 5c). Note that we only include credits from methodologies and projects that are covered by the underlying studies. For instance, we only include credits from Chinese wind power plants under the CDM19 or IFM projects that use California’s Air Resources Board protocol4,5. For avoided deforestation7,8,21 and cookstoves6,17, we apply the OAR to all credits issued to the project types using the studied methodologies, as the underlying studies cover a representative sample of projects.
a Credits covered in our analysis relative to the total volume of issued credits based on sources from Fig. 1a. b Estimated achieved vs non-achieved emission reductions across covered projects based on percentage estimates from Fig. 4a. c Disaggregated shares of estimated achieved and non-achieved emission reductions across covered projects. IFM refers to improved forest management.
Reasons behind low offset achievement ratios across project types
Overall, our assessment indicates that the total achieved emission reductions of the carbon crediting projects for which evidence is available are substantially lower than claimed. We discuss potential sources behind the offset achievement gap across the analysed project types.
Improved forest management (IFM)
Two studies4,5 investigating 106 IFM projects did not find statistically significant reductions in carbon emissions and removals from IFM activities under the ARB protocol. These studies focus on project emissions and baselines. IFM projects involve forest management practices that increase carbon in forests and/or reduce carbon loss in forests. While IFM activities can include extending harvest rotations, reduced impact logging, liberation thinning and converting logged forests into conservation forests, most IFM projects mainly generate credits from avoiding forest degradation. Globally, over three-quarters of carbon credits from IFM projects were issued under the California Air Resources Board’s US Forest Projects Protocol29, and the protocol has been the focus of the two studies of the quality of IFM carbon credits included in our assessment.
Stapp et al.4 analysed 90 IFM projects and overall found no statistically significant evidence of additionality across the United States over the first 5 years of the projects when compared with control lands. The authors document heterogeneous impacts across sub-groups. They observe reduced harvests for land owned by timberland investment management organisations and real estate investment trusts but increased harvesting from other groups. Overall, the positive and negative effects on harvesting balance out across the study sample. The study explains this lack of impact as adverse selection. The baseline is often set as the average carbon per hectare for the forest type in the region of the project. The study finds that lands enroled into carbon crediting projects already had lower rates of harvest over decades before the start of the carbon crediting project compared to the average lands used to set the baseline. Hence, these projects were able to accumulate carbon compared to the baseline before the project started and then generate credits against an average baseline without needing to change how the forests were being managed.
Using a comparable approach, Coffield et al.5 also find no evidence of additionality from 16 ARB IFM projects in California. The study found no statistically significant evidence of increased carbon accumulation after project initiation compared to similar control areas. Similarly, it found no evidence of reduced harvesting compared to past harvesting rates in the project areas and compared to harvesting rates of similar control areas. Lastly, while Badgley et al.33 (72 projects analysed) could not be integrated into our quantitative assessment, the authors also document systematic over-crediting in California’s carbon offset programme due to adverse selection.
Other studies of ARB IFM projects have found additional sources of over-crediting, suggesting that even if some projects changed their forest management practices, the emission reductions or removals would still likely be overestimated due to methods for assessing leakage25 and for quantifying reversal risk and associated contribution of credits into the insurance buffer pool13. No studies to date have conducted quantitative assessments of the quality of credits under other IFM protocols. However, similar issues of lenient baselines, low leakage deductions and low deductions for reversal risk into the buffer pool have been documented for most protocols24.
Wind power
Two studies18,19 investigated 1966 wind power projects registered under the CDM in India and China. These studies only investigate the additionality of these projects. Globally, around half of credits from wind power projects were issued under the CDM, 63% of which were generated in China. We use only the data by Chan and Huenteler19 to estimate the OAR of wind power projects, because ref. 18 only identify the most obvious cases of non-additionality but provide no central additionality estimates for all projects.
Chan and Huenteler19 investigated the additionality of 2051 wind projects, of which 1494 were financed in China under the CDM between 2007 and 2012. They found no statistically significant evidence that projects that received funding from the CDM were less financially viable than those constructed without support. However, they show that projects under the CDM used more foreign technologies and larger wind turbines, potentially increasing technology transfer. In addition, they document a small positive effect on CDM projects being sited in previously undeveloped areas. Yet, these positive effects can only be ascribed to CDM financing if projects were additional, which appears not to be the case.
Calel et al.18 investigate the additionality of 1350 wind projects in India, of which 472 were financed under the CDM between 2000 and 2013. They developed a new conceptual framework called Blatantly Infra-marginal Projects, which identifies particularly obvious cases of non-additionality. The approach allows the authors to identify projects that were less financially attractive but were built even without selling carbon credits. For around half of these projects, they identified that these projects had lower capacity factors, were in less windy locations and were sited further away from electrical substations, and hence, overall, likely to be less financially attractive than the CDM projects.
The authors indicate that low additionality is likely due to the capital intensity of this project type. Utility-scale renewable energy projects require high up-front investments and a secure cash flow to secure funding from banks and investors34. As revenue streams from selling carbon credits are often low compared to revenues from electricity sales and carbon credit prices may fluctuate substantially, as in the CDM, revenues generated by carbon credits are unlikely to affect the financial viability of renewable energy projects substantially19,23. We do not assess small-scale projects, such as off-grid energy, due to a lack of studies.
Cookstoves
Information from two studies6,17 investigating 52 projects was used to estimate an average OAR of 10.8% (Supplementary Table 6 explains how we post-process and synthesise the results from these studies; this is the weighted average across projects covered by studies). Aung et al.11 assess project and baseline emissions for one CDM project. Gill-Wiehl et al.6 analysed 51 projects (40% of all issued credits across independent crediting mechanisms from five key methodologies) and assessed all relevant factors (apart from additionality and leakage) in the quantification of emission reductions, including fraction of non-renewable biomass, adoption/usage rates, and emission factors. Distributing fuel-efficient cookstoves seeks to reduce greenhouse gas emissions by subsidising households in low- and middle-income countries to switch to a less GHG-intensive fuel or a more energy-efficient stove. Most cookstove projects are registered under the Gold Standard (GS), the VCS or the CDM and rely on GS and CDM6 methodologies.
Aung et al.17 ran an RCT to evaluate the climate impacts of one CDM-approved stove replacement project in India. The author team randomly assigned 187 households to either receive a fuel-efficient replacement (96 households) for their traditional stove or to serve as a control group. Overall, Aung et al. find no statistically significant impact on fuelwood usage between the intervention and control groups (hence, we assume an OAR of 0%). They document that 40% of households that received the fuel-efficient stove continued using the traditional stove. They hypothesise that the lack of reductions might also be due to households cooking larger meals with the improved stoves (‘rebound effect’), thereby eliminating any efficiency-based reductions in fuelwood consumption.
While Aung et al. only analysed one project, Gill-Wiehl et al.17 assessed the overall quality of a substantial portion of cookstove credits on the voluntary carbon market, covering 51 projects, five key cookstove methodologies and a comprehensive set of factors. The authors recalculate the likely emission reductions of these analysed cookstove projects by scrutinising key methodological assumptions made to issue credits. Overall, the authors find that the project sample likely only achieved 10.9% of the claimed emission reductions, though there is a large variation between methodologies (please note that the OAR of 10.8% calculated for the overall project type is the weighted average by issued credits from refs. 6,17). For instance, Gold Standard’s Metered methodology35, which assesses fuel use directly, features the lowest over-crediting risks of all methodologies.
Hence, while efficient cookstoves have been found to offer considerable sustainable development benefits, the literature suggests that their low carbon credit quality is due to a lack of rigour and flexibility in how methodologies allow projects to (1) determine the fraction of non-renewable sources of fuelwood and other biomass (fNRB), (2) assess actual use of the new and old stoves and (3) translate these values into changes in fuel consumption. Only Gold Standard’s Metered methodology accurately assesses stove use and fuel consumption by directly metering stove or fuel use. All other methodologies use methods with known biases or inaccuracies. To some extent, all use infrequent and simple surveys, which are vulnerable to bias when respondents give answers they believe the project developer wishes to hear36. Kitchen performance tests can have similar biases, when stove users change their behaviour when they are observed. Some methodologies also use stove efficiency ratings determined in laboratory settings that can be artificial and inapplicable to real-world conditions.
In addition, numerous other studies have evaluated one or a few factors in the emission reduction calculation and compared them to carbon crediting projects or methodologies’ approaches, finding over-crediting from the choice of fNRB37 and methods to track adoption/usage rates38 and under-crediting from emission factors39. Rigorous evaluations of field interventions have found substantial variation in the achieved emission reductions40,41,42,43,44, which are rarely on par with the levels claimed by carbon crediting projects6. Studies investigating the additionality and leakage of cookstove projects are still nascent in the literature but analysing these factors would be important to fully assess the achieved emission reductions6.
Avoided deforestation
Three studies7,8,21 investigating 48 projects that seek to avoid deforestation were used to estimate an average OAR of 24.7% (see Supplementary Table 6 for description). For 26 projects, two independent estimates exist on their OAR (Fig. 6). Projects that seek to avoid deforestation employ various approaches, mostly to protect rainforests in the Global South, such as improved monitoring and control of deforestation in the areas and encouraging sustainable land uses7. All projects covered by our assessment that seek to avoid deforestation are registered under one of several VCS methodologies (e.g. VM0015, VM0007).
a Offset achievement ratio across studies. b Average 5-year issuance in million tons CO2. Based on West et al.7,8 and Guizar-Coutiño et al.21. Credit data is based on the VCS registry50 and contains all credits that have been issued. Please note that for the main estimates in Fig. 4, we use the estimates from the full sample of projects, which is an OAR of 24.7% (compared to an OAR of 14.5% on average for this subset of 26 projects). Average 5-year issuance is based on the total issuance from project start year to 2024 (calculated by dividing total issuance by total years since inception, multiplied by five).
West et al.7,8 investigated 36 projects (of which 32 projects contained sufficient data for analysis) across multiple jurisdictions and found an overall achievement ratio of 8.2%. The authors argue that a central reason for the low achievement ratio is the inherently flawed methodological frameworks used to calculate credit issuance. Specifically, project developers use deforestation baselines informed by historical trends in chosen reference areas defined at the outset of the project, which often result in unrealistic scenarios7,8,9. West et al.7,8 recalculate the achieved emission reductions based on control areas not enroled in the project. Guizar-Coutiño et al.21 investigated 40 projects (of which 35 contained sufficient data for analysis) and found a higher average OAR (42%) for a partially overlapping set of analysed VCS projects as in West et al.
Yet, we found that studies diverge somewhat in their OAR assessments, even if the same offset project is analysed. For the 26 projects that were analysed by West et al.7,8 and Guizar-Coutiño et al.21 the weighted average OAR is 14.5% (with West estimating 10.5% and Guizar-Coutiño 18.5% for the fully overlapping set of VCS projects). The estimates are moderately correlated, with a correlation coefficient of r = 0.4 (Fig. 6). Several reasons could explain this divergence, such as differences in methodology, selection of control groups and pixel vs. area-based approach. The observed divergence underscores the challenge of estimating baselines and the OAR of avoided deforestation projects. Estimates are very sensitive to the creation of the control group, a non-trivial task due to the unobservable nature of these groups and the necessity of their construction via statistical methods. Overall, while the findings from West et al. and Guizar-Coutiño et al. diverge, they indicate that forest protection was much less effective than the volume of issued credits indicates.
Yet, West et al. and Guizar-Coutiño do not assess project developers’ assumptions regarding the carbon contained in the forest areas, which can further lead to over-crediting (see Fig. 3). Bomfim et al.45 assess project developers’ estimates of the carbon per hectare in protected forests. If these estimates are overstated, then the issuance of credits will also be inflated. Based on a representative sample of 12 projects across four key VCS methodologies, the authors show that project developers have significant leeway in assessing carbon content in forests. They found that project estimates were 23%–30% higher than values drawn from scientific literature. We do not consider this potential additional source of overestimation in our OAR calculation, as more research would be needed to ascertain the carbon rates per hectare on a project level.
Further to the carbon crediting project evaluations, a large literature exists that assesses the effectiveness of interventions seeking to avoid deforestation or similar environmental degradation46. Studies have found a wide variance in the effectiveness of these interventions. For projects that have low performance, studies have documented various reasons, such as poor administrative targeting (i.e. the project does not protect the forest most at risk), adverse-self-selection (those without intention to deforest self-select into programmes) and non-compliance (many schemes do not have appropriate measures to sanction non-compliance)46.
Chemical processes
Based on two studies15,16 evaluating HFC-23 and SF6 projects in chemical processes, we derive OARs of 16.4% for SF6 and 68.3% for HFC-23 destruction. These studies investigate project and baseline emissions but do not address leakage (see Fig. 3). The projects were registered under the CDM and JI.
Schneider15 analysed assumptions about baselines made by 19 HFC-23 destruction projects under the CDM. For two projects (CDM 151, CDM 1105), the author observed monitoring periods in which the projects could not issue carbon credits due to methodological constraints. For these two projects, we leverage historical data and data observed in periods without carbon credit issuance to compute the OARs.
Schneider and Kollmuss16 investigated four projects, three abating HFC-23 and/or SF6 under the JI mechanism in Russia and one trifluoroacetic acid (TFA) plant in France. We exclude the plant in France due to lacking historical data. To calculate the OAR for these plants, we follow a similar approach as in Schneider15 (see Supplementary Table 6).
Generally, HFC-23 and SF6 abatement projects have a high likelihood of additionality as there is commonly no business case for these interventions in the absence of financial or regulatory incentives. Yet, the high carbon credit revenues can lead to perverse incentives to increase waste gas generation beyond levels that would occur without carbon credits. The two CDM projects lowered their HFC-23 waste gas generation in periods when they could not claim carbon credits15. The CDM Executive Board revised the respective methodology to address this issue, but most plants never applied the new methodology as they stopped issuing credits due to a lack of demand. The HFC-23 and SF6 projects under JI abruptly increased their waste gas generation at the point in time when plant operators could generate (more) credits by producing more waste gas16. For the HFC-23 projects, changes in waste gas generation were more moderate than for the two SF6 projects for which waste gas generation also exceeded Intergovernmental Panel on Climate Change (IPCC) default values by up to 85 times, leading to lower OAR values for the SF6 projects compared to the HFC-23 projects.
Next to these two studies that qualify for our analysis, several studies have assessed the quality of projects abating nitrous oxide (N2O) from adipic acid and nitric acid production2,3,22. These studies indicate that carbon leakage may have led to some over-crediting from CDM projects abating N2O from adipic acid production. For N2O abatement from nitric acid production, older CDM methodologies (AM0028 and AM0034) involve considerable uncertainty regarding N2O generation in the baseline and pose some risk of over-crediting, whereas a new methodology version (ACM0019) is likely to lead credit fewer emission reductions than are actually occurring.
Implications for carbon crediting mechanisms
We synthesised the extant literature relying on experimental or rigorous observational methods, covering 14 studies on 2346 carbon mitigation projects and 51 studies investigating similar field interventions implemented without issuing carbon credits. Our analysis covers about one-fifth of the credit volume issued to date, almost 1 billion tons. We estimate that less than 16% of the carbon credits issued to the investigated projects constitute real emission reductions, with 11% for cookstoves, 16% for SF6 destruction, 25% for avoided deforestation, 68% for HFC-23 abatement and no statistically significant emission reductions from wind power projects in China and IFM projects in the United States.
Our assessment, therefore, documents substantial and systemic quality problems across all analysed project types, which further strengthens the evidence by previous cross-cutting analyses of the CDM and the JI2,47. Carbon credits are issued based on standards developed by carbon crediting mechanisms. The quality of carbon credits hinges on the robustness of these standards, the choices made by project developers in applying these standards and the thoroughness of the checks by third-party auditors and the carbon crediting mechanism. Our assessment highlights that many project developers pick favourable data or make unrealistic assumptions6. Some methodologies make use of outdated data or inappropriate methodological approaches4, which can lead to adverse selection35 or perverse incentives12,15,16. Our results also indicate that there is substantial heterogeneity across project types and methodologies.
The reviewed studies suggest that existing approaches to assess additionality have led to many non-additional projects being registered. To address this issue, carbon crediting programmes could limit eligibility to project types that have a high likelihood of additionality and of being effectively supported by revenues from carbon credits. For example, following criticism regarding additionality, Verra and the Gold Standard excluded wind power projects in most countries from eligibility. However, newer crediting mechanisms, such as the Global Carbon Council, include these projects in their scope. This change would result in a much narrower set of eligible project types.
Our findings also suggest that the standards and methodologies to quantify emission reductions need to be considerably improved. Such improvements should address a range of issues, in particular reducing project developers’ flexibility in making favourable methodological assumptions to maximise credit generation6,8,21; using conservative assumptions and data based on the latest scientific evidence6,18,19,45; and addressing the risk of adverse selection4,5 and perverse incentives15,16. Carbon crediting programmes may also exclude project types from eligibility where it is very difficult to ascertain whether calculated emission reductions result from the mitigation activities or exogenous factors that impact emissions, an issue that has also been referred to as ‘signal-to-noise’ issue.
Various other studies, not included in our analysis, suggest that quality issues also persist for many other project types not covered by our analysis2,3,11,47,48. Our estimate that 812 million carbon credits do not represent actual emission reductions should, therefore, be considered as a lower bound as many more credits currently traded may not constitute real emission reductions.
In addition, questions around additionality and leakage remain only partly addressed by the literature24,25 and our analysis does not cover two other potential sources of over-crediting: permanence and double counting. For instance, Holm et al.20 assess the non-permanence risk for 57 VCS forestry projects. Project developers need to make non-permanence risk assessments which inform the number of carbon credits set aside to insure against future reversals. Holm et al.20 recalculate the assessments made by project developers based on the latest scientific literature and find that project developers were issued on average 26.5% more credits than an appropriate risk management would demand. Cookstoves projects also face a non-permanence risk as more fuel-efficient cookstoves lead to the preservation of carbon stocks in surrounding forests, but this risk is not accounted for by any of the carbon crediting mechanisms. Double issuance presents another risk as more than half of cookstove projects are co-located in areas where projects seek to avoid deforestation49. Hence, our estimates would likely be even lower if these factors were considered.
Our findings also suggest that more research is needed to better understand the quality of credits across different project types. For instance, for renewable energy, the extant literature providing quantitative assessments of achieved emission reductions focuses primarily on grid-connected wind power projects18,19, though the literature on small-scale renewable energy is scant. More work is also needed to explore the full sources of over/under-crediting of projects with existing evaluations.
Demand for carbon credits is expected to grow significantly over the next decades, with increased demand from voluntary carbon market buyers, domestic compliance markets, CORSIA and countries using Article 6 of the Paris Agreement1. Yet, our results substantiate doubts about the environmental quality of carbon credits from the project types we study. These quality issues need to be addressed for carbon crediting mechanisms to meaningfully contribute to climate change mitigation.
