“The dynamics of social-ecological systems pose a fundamental challenge to attributing changes in carbon stocks to actions taken by carbon offset projects.” Interview with Dr. Pushpendra Rana

“The challenge is not simply poor measurement, weak data, or the absence of improved methodologies.”

Interview with Dr. Pushpendra Rana, lead author of the paper “Why carbon offsets may fail in complex systems: A causal inference perspective”. REDD-Monitor wrote about the paper in February 2026. The interview was conducted by email on 8 March 2026.

REDD-Monitor: Can we start with a brief description of who you are, your work with the Indian Forest Service, and your academic work.

Pushpendra Rana: I work at the intersection of practice and research. On the practice side, I am an officer with the Indian Forest Service, where I have spent more than 15 years working on forest restoration, conservation, and community forest interactions in the Himalayas and other parts of India.

At the same time, I am an academic researcher trained in causal inference, complex systems analysis, geospatial methods, and machine learning. I completed my PhD in Geography at the University of Illinois at Urbana Champaign and later held postdoctoral research positions at North Carolina State University and the University of Illinois. My research focuses on evaluating the real world impacts of environmental policies and climate interventions. I have studied topics such as forest certification, REDD+, tree planting programs, and climate finance initiatives across several countries.

My broader interest is understanding how we can design climate, conservation, and forest management policies that are both scientifically credible and practically effective. That means combining insights from social science, ecology, and data science to better understand how complex social ecological systems behave and how interventions actually work in practice.

REDD-Monitor: You were recently the lead author of a paper published in Environmental Science & Policy, “Why carbon offsets may fail in complex systems: A causal inference perspective” which argues that the problems with carbon offsets cannot be fixed. Can you outline the problem with carbon offsets that the paper focuses on?

Pushpendra Rana: Our paper focuses on a structural tension at the heart of carbon offset markets. A carbon offset credit is meant to represent a precise quantity of climate benefit, for example one tonne of CO₂ that was not emitted or removed from the atmosphere. To justify issuing that credit, a project must show that the benefit occurred because of the project and would not have happened otherwise. In other words, the project must establish a credible counterfactual.

The challenge is that many forest carbon projects operate in complex social ecological systems where many factors simultaneously influence forest outcomes, including economic change, migration, markets, governance, weather, fire, and land use decisions. Having worked inside forest landscapes for many years, I have seen firsthand how forest outcomes emerge from the interaction of these forces. In such systems, it becomes extremely difficult to confidently isolate the effect of a single intervention.

Our paper suggests that the challenge is not simply poor measurement, weak data, or the absence of improved methodologies. It arises from deeper limits of causal inference in complex systems, where key assumptions required for causal attribution or establishing a robust counterfactual, such as independence of units or the absence of unobserved drivers, are often violated.

Importantly, causal inference methods are well suited for estimating average treatment effects across many observations and assessing whether policies work in general. However, they are not designed to reliably determine project level attribution for individual cases, which is exactly what carbon offset systems require when issuing credits. To illustrate the point, I often use a medical analogy. Doctors can estimate that a treatment increases survival on average across a population, but they usually cannot say exactly how much that treatment extended the life of a particular patient because we cannot observe what would have happened otherwise. Carbon offsets demand that kind of individual attribution at the project level, and our argument is that in complex social ecological systems, that level of precision is often fundamentally difficult to establish.

REDD-Monitor: You looked at two carbon projects, one in Brazil and one in India. Why did you choose these projects?

Pushpendra Rana: We purposely selected two projects that are among the stronger examples in the field.

One is a REDD+ project in the Brazilian Amazon, and the other is a Clean Development Mechanism (CDM) afforestation project in the Indian Himalayas. Both projects were well funded, supported by major institutions, and implemented with established methodologies. We also had access to unusually detailed contextual information about both projects.

The idea was to examine “most-likely cases.” In other words, if causal attribution works anywhere, it should work in projects like these. The two cases also represent very different ecological and institutional contexts—the Amazon and the Himalayas, avoided deforestation and afforestation interventions—allowing us to examine whether similar causal challenges arise across different systems.

Our analysis showed that despite these differences, both projects exhibit similar patterns of uncertainty in causal estimates and strong influence from interacting social and ecological drivers.

REDD-Monitor: Why are only two projects enough to reach the conclusion that carbon offsets cannot be fixed?

Pushpendra Rana: The purpose of the cases is not to achieve statistical generalization but to illustrate a theoretical argument about the limits of causal attribution in complex systems. Rather than estimating the average effect across many projects, the cases allow us to examine whether the core assumptions required for project level causal attribution can hold in real world settings. If projects that are relatively well designed and well resourced still struggle to establish stable causal estimates, this suggests that the issue may not simply be poor implementation but a deeper structural challenge.

In the history of science, carefully chosen cases have often been sufficient to challenge prevailing assumptions. For example, Ignaz Semmelweis showed through observations in a single hospital in Vienna that handwashing dramatically reduced maternal mortality, challenging prevailing medical beliefs even before germ theory was fully understood. Similarly, John Snow’s investigation of a cholera outbreak in London in 1854 revealed that contaminated water rather than “bad air” was responsible for the disease, fundamentally transforming public health thinking. The value of such cases lies not in their number but in their ability to reveal whether core theoretical assumptions hold under real world conditions.

In our study, the cases allow us to examine the mechanisms behind the uncertainty, including nonlinear dynamics, interacting drivers, and unobserved confounding, rather than only measuring outcomes. These dynamics are not unique to the two projects we studied. Similar structural features, including multiple interacting drivers and nonlinear processes, are common in forest based climate interventions implemented in complex social ecological systems around the world, and may therefore lead to similar challenges in establishing stable causal attribution.

REDD-Monitor: Carbon trading proponents argue that the methodologies behind carbon offsetting projects have improved since the two projects were developed that you look at in your paper. Do these new methodologies address the problems that you raise?

Pushpendra Rana: No, the newer methodologies do not address the problems we raise in the paper, for two main reasons. First, they do not overcome the fundamental limits of causal inference. Carbon offset policies require the construction of credible counterfactuals—estimates of what would have happened in the absence of a project—in order to attribute climate benefits to a specific intervention. Causal inference methods are very useful for estimating average treatment effects across many observations, helping us understand whether policies tend to work in general. However, they are not designed to reliably determine individual project-level attribution, which is precisely what carbon offset systems require when issuing credits for specific individual projects.

Second, the dynamics of social-ecological systems pose a fundamental challenge to attributing changes in carbon stocks to actions taken by carbon offset projects. In such systems, individual project counterfactuals cannot be observed with confidence, and core causal assumptions—such as non-interference between units and the absence of unobserved drivers—are often difficult to satisfy. Our analysis using advanced causal and machine learning methodologies shows that carbon outcomes in nature-based offset projects emerge from nonlinear and interacting dynamics typical of complex social-ecological systems, where many variables jointly influence outcomes. This creates substantial uncertainty about the causes of observed changes and makes it very difficult to attribute changes in carbon storage to specific interventions such as offset-funded programs, even when advanced analytical methods are applied.

Focusing only on developing better methods—when key assumptions cannot be reliably enforced or tested in real-world settings—may ultimately slow the search for more robust and effective climate mitigation solutions.

REDD-Monitor: Please explain the methods that you used to demonstrate that social-ecological complexity create high levels of uncertainty and make it very difficult to conclude that any changes in carbon storage are the result of the carbon project.

Pushpendra Rana: We used a combination of modern causal inference and machine learning approaches to test whether credible project level attribution could be established in complex social-ecological systems. These included Directed Acyclic Graphs to map potential causal relationships, propensity score matching to construct comparable control groups, and Double Machine Learning models to estimate nonlinear effects and examine how multiple interacting drivers influence forest outcomes. We also conducted sensitivity analysis to assess how unobserved factors might affect the estimated impacts.

Together, these approaches allowed us to examine whether the causal effects attributed to the projects remain stable when accounting for complex relationships among variables, nonlinear dynamics, and potential hidden drivers. The methods are described in detail in the paper to ensure easy replicability.

REDD-Monitor: Can you give some examples of the type of complexities encountered in forest carbon projects?

Pushpendra Rana: Forests are embedded in coupled social ecological systems, where ecological processes and human decisions interact. In these systems, forest outcomes are shaped by multiple factors operating simultaneously and often in complex ways. For example, nonlinear dynamics mean that relatively small changes in drivers, such as road access or grazing pressure, can produce very different forest outcomes.

At the same time, multiple interacting causes including agriculture, livestock, fire, markets, governance arrangements, and climate conditions all influence patterns of forest change. Interventions can also generate spillovers or leakage, where protecting forest in one location shifts deforestation pressures elsewhere. In addition, important drivers such as migration patterns, local institutions, or informal markets may strongly influence outcomes but remain difficult to observe or measure.

Because these processes interact, causal effects can vary across locations, over time, and across combinations of variables. In such systems, uncertainty is not simply a measurement problem. It emerges from the way multiple drivers interact, making stable attribution of outcomes to a single intervention extremely challenging.

REDD-Monitor: Have you had any response from carbon certification companies such as Verra to your paper?

Pushpendra Rana: The conversation around the effectiveness of climate interventions is evolving rapidly, and many organizations are actively exploring ways to improve credibility and transparency. My hope is that the paper contributes to that discussion by highlighting the importance of aligning policy design with the realities of complex systems.

Climate change is an urgent challenge. If we rely on mechanisms that systematically overestimate their impact, we risk delaying the deeper emissions reductions that are necessary. We have not yet received responses from certification bodies, but the goal of this work is not to criticize particular actors. Rather, it is to encourage more robust and credible approaches to climate mitigation.

In short, forests should absolutely be financed and protected. But climate policy may work better when emissions reductions and nature-based investments are treated as complementary strategies rather than interchangeable ones. The challenge ahead is not whether to finance forests, but how to do so in ways that respect the complexity of the systems we are trying to protect.