Dynamic Baselining for Next Gen ARR projects: How Stocking Index Quality Drives Impact & Integrity
April 28, 2026
By Florian Reber, SVP Business Development & Partnerships, Chloris Geospatial
I spend a lot of time talking to ARR project developers about VM0047. Over the past year, the conversation has changed. A year ago, most questions were about the methodology itself: how does performance benchmarking work, what does a valid donor pool look like, how do you satisfy statistical matching requirements? Those questions have largely been answered.
Today the questions are sharper, and they tend to focus on the data. Specifically, whether the data underlying a dynamic baseline is actually fit for purpose. That sounds like a technical question. In practice it's a commercial one. Methodology quality and data quality are separate problems, and conflating them is one of the more consequential mistakes a developer can make in project design.
The stocking index problem
VM0047's performance method requires tracking carbon stock change in project and control plots, year over year. The baseline is only as credible as the signal it's built on, and most common remote sensing metrics struggle to detect biomass growth - the incremental, year-on-year woody accumulation that ARR credits depend on.
Vegetation indices like NDVI or NDFI saturate quickly in denser canopies. They're noisy, sensitive to seasonality, cloud cover, and soil background. Canopy height models tell you something grew taller but ignore the structural and density components that drive actual biomass. None of these metrics were designed to reliably detect the subtle growth signal a dynamic baseline requires.
VM0047 explicitly allows projects to upgrade their stocking index as better data becomes available. That's a designed feature of the methodology, not a workaround. Projects that launched with NDVI-based approaches aren't locked in. But the question of what to upgrade to, and how to evaluate the options, is one the market hasn't fully worked through yet.
The direction Verra has signaled is toward direct above-ground biomass estimation: a stocking index that integrates height, structure, and density rather than proxying for any single one. A well-constructed AGB time series doesn't saturate in maturing canopies, and it continues to resolve year-on-year growth at exactly the point in a project's life when that signal matters most.
What validation actually means
Every biomass estimate is exactly that: an estimate. In carbon markets, nobody measures biomass at scale directly; that would mean felling trees and weighing them. The key point to prove then is whether your estimate is tracking reality closely enough to support crediting decisions.
That is why data validation matters. And not all validation is equal. Internal consistency checks, confirming a model behaves predictably across different inputs, tell you something. They don't tell you whether the underlying numbers are right. Meaningful validation requires testing against fully independent, higher-quality reference data: field measurements, airborne surveys, datasets that played no role in training the model being evaluated.
It also matters what you're validating. Validating biomass stock, a snapshot of carbon present at a given moment, is different from validating biomass change, which is the quantity that actually drives crediting in a performance-based methodology. A model can be reasonably accurate on stock while being poorly calibrated on the year-on-year growth signal. For VM0047 purposes, the latter is what matters. It's also the harder thing to get right.
The field is still early here. Most accuracy reporting in the remote sensing biomass space focuses on stock, uses internal consistency checks, or tests against data that overlaps with training. Independent, multi-biome validation of carbon change specifically is not yet standard practice. That gap will become more visible as VM0047 issuances scale and investor scrutiny increases.
When evaluating a data provider, three questions reveal most of what you need to know: What reference datasets were used for validation? Were they fully independent of training data? Was change validated, or only stock?
The ex ante problem
There's a related challenge that the methodology doesn't yet fully address, and that comes up consistently in conversations with developers and investors: forward projections.
A dynamic baseline tells you what happened. What investors increasingly want to know is what will happen: how many credits this project will generate, and how much confidence sits behind that number. Most developers, when pressed, are working from growth assumptions that are difficult to defend with empirical evidence. That's not a criticism of the developers, it reflects a genuine gap in the available data infrastructure.
Building defensible forward projections requires an empirical record of how forests in similar conditions have actually grown over time across different climate years, disturbance events, and recovery trajectories. You need to know not just what a model predicts, but how much variance there is in real outcomes around that prediction, and what drives it.
That kind of historical depth is rare in the remote sensing biomass space, which is relatively young as a commercial field. Most products have a few years of data behind them. Where longer records exist, a decade or more of annual biomass estimates across diverse biomes, they carry information that point-in-time models simply can't access: how forests respond to drought years, how recovery rates vary by ecosystem type, what the realistic range of outcomes looks like for a project of a given size in a given region. That historical record is the foundation on which credible forward projections get built. Without it, ex ante estimates rest on assumptions that are hard to stress-test and harder to defend under scrutiny.
Where this leaves you
VM0047 is a well-constructed methodology and the market around it is maturing quickly; the first credits were issued under it just days ago. But good methodology design and good data design are distinct disciplines, and they deserve equal rigor.
The questions worth asking any data provider: What are you actually measuring, and how does it behave in maturing canopies? How was accuracy established and against what reference data? Was change validated independently, or just stock? What is the uncertainty on individual annual estimates? What does your historical record look like in ecosystems like mine?
These aren't easy questions to answer well. But they're the questions a sophisticated investor or verifier will eventually ask, and the time to work through them is in project design, not at verification. The gap between a credible dynamic baseline and a technically compliant but fragile one is often larger than it looks from the outside. It tends to show up at the worst possible moment.
If you're designing a VM0047 project and want to pressure-test your data choices, I'm happy to go into specifics on stocking index selection, baseline uncertainty, and what fit-for-purpose looks like across different biomes and project sizes.