By Aida Mashkouri, Head of Machine Learning and Remote Sensing, Revalue
The voluntary carbon market was built on a powerful promise: that one carbon credit equals one tonne of COâ‚‚ avoided or removed from the atmosphere. It is an elegant idea, but one that depends on trust. And right now, that trust has faltered.
In 2023, The Guardian reported that 90 per cent of rainforest credits were ‘worthless,’ citing flawed science and overstated impacts. The story struck a nerve because, at its heart, it captured a truth many already suspected: that some carbon credits did not live up to their claims.
But the real issue is not always the projects themselves, it is the prediction, the way we calculate what would have happened without them.
The Problem with Static Baselines
Every avoided deforestation project needs a baseline. It is the benchmark that shows how much forest would have been lost without protection. But for years, most projects have relied on what are called static baselines: a single forecast made at the start of the project, often based on historic deforestation data, that is expected to hold true for years or even decades.
That may sound sensible. But in reality, it is an assumption, and one that quickly falls out of date.
Static baselines ignore the changing world around them. Commodity prices fluctuate, new roads are built, governments shift policy, and rainfall patterns move. These are all drivers of deforestation, and they are anything but static. Yet, static models assume tomorrow will look like yesterday.
The result is over-crediting. Projects claiming to avoid emissions that were never actually at risk. That is how trust erodes, even when the intentions are good.
A More Scientific Approach
If the problem is prediction, the answer is measurement.
Dynamic baselines replace those one-time forecasts with continuous comparisons to the real world. Instead of guessing what might have happened, they measure what actually did happen in comparable areas of forest that were not protected; what scientists call ‘control’ or ‘counterfactual’ plots.
It is the same principle used in medicine. You would never approve a drug without comparing it to a control group. So why approve a carbon credit without one?
If deforestation increases in the control areas but remains low in the protected forest, we can measure the true impact. It is not a projection. It is evidence.
This is how science restores trust to avoidance: by replacing assumption with observation.
Building Dynamic Baselines in Practice
We can build these comparisons using advanced data science and machine learning. Today, models can scan millions of hectares of satellite data to identify areas that share the same conditions as the project zone – the same forest type, elevation, proximity to roads and exposure to human pressure. These become the project’s ‘statistical twins.’
A scalable framework is emerging across the market, building dynamic baselines that identify landscapes with similar ecological conditions and exposure to human pressures across large regions. Using multiple satellite data sources and indicators such as forest cover, accessibility, and land-use change, it locates areas that share the same deforestation risk as the project site.
These ‘analogue’ landscapes form adaptive, data-driven baselines that update as new information becomes available, reflecting real shifts in deforestation drivers and reducing the bias found in static or geographically fixed baselines. The approach works across millions of hectares and enables transparent measurement of baseline uncertainty.
To give a sense of the rigour involved, our team ran more than 100 experiments for a single project in Tanzania to refine the model and ensure the most accurate possible match.
Engineering Rigour at Scale
Creating a dynamic baseline at this level of precision requires both scientific and computational power. High-performance computing systems can process terabytes of geospatial data across entire landscapes, enabling models that are both fine-grained and scalable.
New systems integrate high-resolution satellite imagery and LiDAR-based biomass measurements, ensuring that carbon stock estimates are grounded in observed reality rather than generic models.
Restoring Trust in Avoidance
The voluntary carbon market is at a crossroads. Every overestimated baseline, every inflated claim, weakens buyer confidence. But abandoning avoidance would be a mistake. Protecting primary forests remains one of the most cost-effective, scalable and immediate tools we have to slow climate change and protect biodiversity.
Dynamic baselines offer a way to do it credibly. They bring the transparency and scientific rigour the market has been missing, replacing static assumptions with adaptive, data-driven evidence.
Science does not earn trust by declaring certainty. It earns trust through transparency and by showing what is known, what is not, and what is being improved. To create the most rigorous and trusted nature credits on the planet, we must systematically reduce uncertainty until every credit reflects measurable, verifiable impact.
We’re not claiming this is possible in every project today, nor that the work is finished. Far from it. But dynamic baselines mark an important step forward for avoidance credits, and there remains huge scope for innovation, refinement and collaboration across the sector. The goal for all of us should be to demonstrate what is possible now – and to keep improving methods and standards through constructive collaboration.
Aida Mashkouri is Head of Machine Learning and Remote Sensing at Revalue
Any opinions expressed in this commentary reflect the views of the authors and not of Carbon Pulse.
Carbon Pulse allows subscriber companies to submit one piece of ‘Contributed Content’ (op-eds, advertorials, tenders/RFPs, etc) per year. This post appears in front of our paywall, so it’s readable by anyone. It also appears in our CP Daily newsletter once. Beyond that, or for non-subscribers, we allow companies to purchase ‘Sponsored Posts’. These posts also appear in front of our paywall, while we feature them in our daily newsletter for three consecutive days (instead of one).
You can read more about our Contributed Content/Sponsored Post offering here: https://carbon-pulse.com/advertising-brochure/
