New research shows that social-ecological complexity creates “high levels of uncertainty” and makes it “very difficult” to attribute changes in carbon storage to carbon offset projects
Social-ecological complexity “makes any carbon additionality estimation fragile”.
“Social-ecological system dynamics present a fundamental challenge to the attribution of changes in carbon stocks to actions taken by carbon offset sellers.” That’s the first line of the abstract of a recent paper published in Environmental Science and Policy. It’s written by Pushpendra Rana of the Indian Forest Service, Forrest Fleischman of the University of Minnesota, and Amit Sharma of Microsoft Research, India.
The authors look at two cases, in Brazil and India, to illustrate that, “carbon outcomes in these nature-based carbon offset projects emerge from non-linear and independent dynamics that are the result of the inherent complexity of social-ecological systems, where large numbers of variables jointly influence causal processes”.
This creates “high levels of uncertainty”, they write, and makes it “very difficult” to attribute changes in carbon storage to carbon offset projects.
Project-based offsetting “consistently struggles to meet the expectations of credibility”, Rana, Fleischman, and Sharma write. “To achieve effective climate change mitigation, policymakers need to focus on policies that do not depend on inherently uncertain causal attribution.”
“The credibility of nature-based carbon offsets relies on demonstrating that an offset led to measurable increases in carbon storage,” they write, “and the extent to which offsets do so remains controversial.”
Counterfactual baselines
To evaluate whether offsets are additional, permanent, and avoid leakage requires analysts to make accurate predictions about counterfactuals — what would have happened if the offset project did not happen. This is mostly determined by setting up a counterfactual using an ex ante model projection of the baseline. This model is based on assumptions and predictions. The changes resulting from the offset project are determined by ex post measurements.
In 2016, Larry Lohmann discussed the problem of counterfactual baselines following on from a comment on REDD-Monitor:
Rana, Fleischman, and Sharma note that recent independent evaluations show that most offsets fail to meet additionality, leakage, and permanence criteria.
The carbon trading industry response is that these problems can be addressed with improved methods. The carbon certification organisation Verra, for example, has introduced a new methodology for afforestation, reforestation, and revegetation that uses remote sensing and AI and aims to estimate carbon additionality and how many carbon offsets are generated.
Rana, Fleischman, and Sharma write that,
Nature-based carbon offset projects occur in complex social-ecological systems that exhibit nonlinear dynamics with thresholds, time lags, surprises, reciprocal feedback loops, heterogeneity, and legacy effects. In these systems, the impact of one variable depends on the values of other variables and, moreover these variables interact in a non-linear way to affect outcomes. These nonlinearities are likely to lead to frequent violations of core assumptions required for robust causal inference such as excludability, non-interference (or Stable Unit Treatment Value Assumption, SUTVA), and unobserved confounding (or absence of simultaneous influences) on outcomes. Due to such violations, causal inference in real-world situations remains shallow and fragile. Routine violations of these assumptions mean that it is difficult for offset projects to support claims of additional carbon storage regardless of the sophistication of the causal inference methods they use.
They give the example of uncertainties in carbon storage due to wildfire, drought, disease, and landslides, which are substantial and difficult to model.
Predictions of baseline deforestation are usually based on extrapolation of historical trends. But such predictions “ignore a wide variety of economic, political, and biophysical contexts,” the authors write. These “may influence changes in forests but do not appear in historical trends due to nonlinearities or changing social or climatic conditions in complex systems”.
This can result in the generation of carbon offsets where there is no real reduction in deforestation in the project area.
The case studies
The study looks at the Jari Pará REDD project in Brazil and the World Bank-funded Biocarbon Project in India, an early Clean Development Mechanism project. The Biocarbon project aimed to establish tree plantations on about 12,000 hectares of land to generate carbon credits.
REDD-Monitor wrote about the Jari Pará REDD project in March 2023 — after Verra suspended the project over a land dispute:
Verra carried out a review and reinstated the project in February 2025.
The baseline scenario for the Jari Pará REDD project was set up using historical average rates of deforestation. Rana, Fleischman, and Sharma found that,
In Jari Pará, the effect of REDD+ on forest loss is substantially modified by the distance of the unit of analysis from the road, livestock farming and number of farms. REDD+ reduces forest loss in locations that are near roads and experience high livestock farming. The effect of the number of farms is non-linear with REDD+ reducing forest loss in locations with many farms but increasing forest loss in locations with a smaller number of farms. Despite general trends, it is hard to precisely estimate and predict the nature of such non-linear relationships in advance.
The authors write that forest loss in remote areas is “likely driven by the higher costs and practical difficulties of monitoring and enforcing rules”. However, remote areas with high livestock farming saw less forest loss.
The results show that “social-ecological complexity poses fundamental challenges for causal attribution in carbon offset projects”. This limits the credibility of causal claims in nature-based offset protocols. “These findings,” the authors write, “suggest that nature-based offsets are a risky way to achieve climate change mitigation targets.”
Rana, Fleischman, and Sharma write that,
We attribute the wide uncertainty bounds in our two case studies to social-ecological complexity and the assumptions that require us to make in order to control for non-linearities, unobserved confounders and interactions in nature-based carbon offsets. Such wide bounds may indicate the inability of any modeling approach to estimate stable treatment effects given the complex nature of effects from multiple confounders and the presence of non-linearities.
Regarding REDD, the authors note that project baselines are “model dependent”. The uncertainty of these models in complex systems “makes any carbon additionality estimation fragile and vulnerable to manipulations by project developers”.
“Claims cannot be credibly established”
While the authors only evaluate two carbon offset projects, they argue that the dynamics they report are likely to be present in most projects that aim to store more carbon by manipulating social-ecological systems.
The justification for an offset is that a specific amount of carbon is stored as a result of a carbon project. This specific amount is then sold as carbon credits to allow the same amount to be emitted somewhere else. “In our cases,” the authors write, “such claims cannot be credibly established.”
The problems highlighted in this paper cannot be addressed by tweaking the carbon markets, improving baselines, determining additionality, preventing leakage, or ensuring permanence. The inherent uncertainties associated with social-ecological systems remain.
At a jurisdictional level, the difficulties in understanding causality are likely to be more severe.
“Given the fundamental causal dynamics, attribution of change to a particular intervention funded by the carbon market is likely impossible,” the authors write.
[W]e believe our findings are likely representative of nature-based solutions in any complex social-ecological system. This presents clear challenges for Voluntary Carbon Markets, carbon trading under the Paris Agreement, the California carbon market, and similar programs which rely on specific attribution of carbon mitigation to a complex and dynamic system.
Rana, Fleischman, and Sharma conclude that,
We conclude that social-ecological complexity fundamentally undermines the reliability of causal attribution in offset projects and considerably limits their ability to meaningfully contribute to mitigating climate change. In our two study cases in Brazil and India, carbon out comes were strongly influenced by nonlinear dynamics, multiple causal pathways, and confounding variables that were not observed. We argue that non-linearities, unobserved confounders and interactions are likely in social-ecological systems, and we show that when they are present, they strongly influence the actual effects of nature-based carbon offset projects.