As the world gathers in Belém for COP30, one message will echo louder than ever: protecting forests, such as the Amazon, is not just an environmental imperative—it is one of humanity’s best defenses against climate change.
However, when it comes to designing methodologies for projects that protect forests, we, as the leading carbon standards setter, know that this has never been easy. These methodologies for Reducing Emissions from Deforestation and Forest Degradation plus conservation, sustainable forest management, and forest enhancement (REDD+) quantify the climate impact of the emissions prevented when forests aren’t cut down or degraded, i.e., something that projects, if they are successful, prevent from happening. The inherent difficulty is in establishing a counterfactual.
Carbon accounting methodologies that enable projects to quantify the climate benefit of keeping forests standing, generate carbon credits, and sell those credits on the open market to sustain their activities run counter to the current economic paradigm. This current paradigm often makes activities such as mining and large-scale commercial agriculture more profitable than protecting standing forests.
Verra began developing carbon accounting methodologies for projects that reduce deforestation and forest degradation more than 15 years ago. As the data, technologies, and analytical methods that REDD methodologies utilize have evolved over time—and with particular rapidity over the last five years—they have become the object of recent scrutiny and criticism.
In this article, we seek to provide deeper insight into Verra’s new REDD methodology (VM0048 Reducing Emissions from Deforestation and Forest Degradation), including the how, when, and why behind its development. We also share details about the operationalization and approach of VM0048 and its associated module, VMD0055 Estimation of Emission Reductions from Avoiding Unplanned Deforestation.
REDD Project Methodologies and the Need for VM0048
Verra launched its first REDD methodologies (VM0006 Methodology for Carbon Accounting for Mosaic and Landscape-scale REDD Projects and VM0007 REDD+ Methodology Framework [REDD+MF]) in 2010. At the time, the development of REDD methodologies required creating new concepts and tools; rather than quantifying the impact of a specific action, they needed to quantify the impact of something that was not going to happen. To determine the emissions from deforestation that would occur in the project area in the absence of project activities (i.e., the baseline), the methodologies required the establishment of a reference region: an area near the project area that shared relevant characteristics with the project area but was experiencing deforestation. The deforestation rates observed in the reference region provided an estimate of the project’s baseline.
These methodologies enabled the protection of thousands of hectares of valuable forestland, helped to conserve precious biodiversity, and sustained numerous local communities that are the stewards of these forests.
Some elements of this approach left room for discretionary decisions by the project proponent (for instance, the selection and delineation of the reference regions). Nevertheless, as judged by experts and stakeholders, and given the data, technologies, and analytical tools available at the time, this approach provided a credible and robust way of assessing the climate change mitigation impact of REDD projects.
As projects were being implemented, new methodologies were developed in order to better fit different circumstances or geographies. This resulted in several slightly different methodologies being available for the same type of project. Eventually, the Verified Carbon Standard Program included five methodologies for REDD projects, and in 2020—i.e., before the publication of most media articles criticizing Verra’s REDD methodologies—Verra launched an effort to consolidate its various REDD methodologies while also incorporating the latest methodological, technical, and scientific advances. The main outcome of this effort was the creation of the new, consolidated REDD methodology VM0048.
While the REDD methodologies for projects were being developed and implemented, starting in the late 2000s and early 2010s, various multilateral agencies also promoted the development of jurisdiction-wide, government-led REDD programs. Verra took this approach one step further and developed the Jurisdictional and Nested REDD (JNR) Framework, which was launched in 2012. The framework allows for the integration of complementary types of interventions (e.g., regulatory and on-the-ground) at various scales, led by various actors, and supported by diverse funding sources, resulting in more comprehensive and effective forest conservation programs. The integrative nature of the JNR Framework also enables improved consideration of additionality and more thorough handling of leakage, risk of reversals, and safeguards.
When work on VM0048 began, the operational advantages of a jurisdictional approach were incorporated into the new methodology. As a result, VM0048 no longer requires projects to delineate an ad hoc reference region or to collect their own data to construct baselines. Instead, Verra now leads on developing a single, high-quality deforestation dataset for a given jurisdiction, which is used to predict deforestation for the entire jurisdiction. Verra then allocates deforestation data directly to each project within that jurisdiction, so that each project can build its baseline.
Two major advances of the new approach under VM0048 are (1) that it reduces the potential for any perceived or actual conflict of interest, and (2) that it ensures that project-level carbon accounting is consistent across, and aligned within, the jurisdiction.
The development of projects’ baselines under VM0048 includes the following steps:
- Develop a jurisdictional deforestation estimate—that is, project the amount of deforestation that would occur in the coming years across the entire jurisdiction. This involves, first, collecting the jurisdictional deforestation activity data, a statistical estimate of the area that was deforested across the entire jurisdiction over the 10-yr Historical Reference Period (HRP, i.e., the 10 years prior to the start date of the baseline). The average deforestation rate observed during the HRP is then assumed to hold for the following six years and constitutes the jurisdictional deforestation estimate.
- Build forest cover benchmark maps that show the location of forest and non-forest at the beginning, midpoint, and end of the HRP, as well as corresponding forest cover transitions (e.g., deforestation and regrowth).
- Develop a deforestation risk map that is a spatially explicit, statistical model of deforestation risk. It is built using the forest cover benchmark map and the jurisdictional deforestation estimate and shows where, across the entire jurisdiction, deforestation is likely to happen based on deforestation patterns observed during the historical reference period.
- Allocate fractions of the jurisdictional deforestation estimates to each project within the jurisdiction proportionally to the assessed risk of deforestation in the project’s location.
- Combine the allocated deforestation estimates (in hectares) with local emission factors (i.e., estimates of the amount of carbon stored in forest carbon pools that would be released to the atmosphere upon deforestation of one hectare of forest). Combining deforestation estimates, or activity data, with emission factors provides an estimate of the total emissions if no project is implemented—the project baseline.
There are several reasons why this data development approach yields robust and credible project baselines, which are the basis for calculating the emission reductions these projects achieve:
- It is data driven. The projected jurisdictional deforestation estimates are built and allocated to projects based on the amount and location of deforestation observed during the HRP in the jurisdiction where the project is located.
- It is accurate. Deforestation estimates are derived from centrally produced, high-quality jurisdiction-wide deforestation datasets compiled by expert organizations. These use state-of-the art data and analytical methods, follow current best practices, and meet high accuracy specifications.
- It is conservative. Prior to allocating activity data to projects, discounts are applied to account for the statistical uncertainty implicit in the estimated deforestation rate.
VM0048 Implementation: The Opportunity and the Hurdles
While VM0048 adopts a credible and robust approach for setting projects’ baselines, it also entails implementing an approach that has never been undertaken before. In developing VM0048 activity data, we have encountered several technical and operational challenges, including the following:
- The approach requires the building of high-specification jurisdictional deforestation datasets for over 90 jurisdictions where REDD projects are located or expected to be located in the near future. This is an unprecedented endeavor that necessitates addressing technical challenges including limited availability of high-quality satellite imagery, the need for identification and quantification of forest coverage in a wide range of environments, and the application of a range of different forest definitions and conditions.
- The approach includes building spatially explicit, statistical models of future deforestation risk with reasonable predictiveness.
- To maximize impact and efficiency, Verra has engaged highly renowned expert organizations to support this data development process. While this approach optimizes resources for the data development process, it requires additional oversight, including the following:
- Reviewing the datasets that are developed by multiple providers and identifying solutions for the challenges that arise during implementation
- Coordinating independent technical assessments to ensure that each jurisdictional deforestation dataset meets Verra’s specifications
- Building and operating the infrastructure for allocating and distributing baseline data to projects
This radically new approach is being implemented while it is being developed and tested in the interest of supporting market stakeholders. At the same time, Verra is committed to producing credible and robust data and will not forsake quality for speed. Generating high-quality data at scale comes at the expense of speed, and accurately predicting timelines has therefore been challenging.
Verra is committed to working through the challenges so that the opportunities inherent in this approach can reach their full potential.
This section provides more detailed insight into the challenges we encountered during the data development process under VM0048 and VMD0055.
Data Quality Challenges
- The data requirements under VMD0055 are extensive, to ensure the development of high accuracy data for the sample-based activity data, forest cover benchmark maps, and associated risk maps for each jurisdiction. In many cases, VMD0055 requires jurisdiction-specific datasets or, at a minimum, validation of existing products to confirm their suitability for use within the methodology.
- Verra is committed to leveraging local data sources and expertise wherever possible to ensure the final activity data best represent local conditions. In addition, stakeholders are encouraged to submit materials and provide feedback on provisional risk maps so Verra can incorporate relevant input before releasing the final maps. While important, these steps prolong the process and make it more complex.
Geographical Challenges
The interpretation of activity data as required under VMD0055 is complicated by biophysical and data-related variability. Key landscape factors affect classification accuracy in Forest Cover Benchmark Maps (FCBMs) as well as interpreter accuracy/interpretation times.
- Sparse canopy: Low-density canopies in dry forests or savannas create ambiguity in distinguishing forest from non-forest and, therefore, lead to uncertainty in estimating both stable forest and forest change.
- Seasonality: Phenological variation alters the appearance of vegetation across time. For example, deciduous forests can resemble non-forest during leaf-off periods. Consistent classification of forest and non-forest requires temporal awareness and avoidance of the use of single-date imagery, which introduces bias.
- Different forest types: Jurisdictions contain heterogeneous forest structures and forests at different stages of degradation. Spectral and structural variation among forest types complicate interpreter consistency and classification accuracy in FCBMs.
- Low canopy cover thresholds: National definitions with canopy cover thresholds below 20–30% amplify error. Small deviations in human or automated interpretation of land-cover classes can cause misclassification, and low thresholds blur the distinction between natural variability and actual deforestation.
- Cloud cover: Persistent cloud cover in tropical regions obstructs clear imagery acquisition, reducing usable observations and introducing spatial or temporal gaps in sample coverage and maps.
Data Availability Challenges
- High-resolution imagery is not always continuously available for all jurisdictions and/or there may be temporal and spatial gaps.
- In some geographies, data availability for forest cover maps, validation datasets, high-resolution imagery, and ancillary layers supporting risk mapping are limited, fragmented, or non-existent, constraining consistent application of VMD0055 requirements.
Activity Data Challenges
- Under VMD0055, activity data—the extent of deforestation—must be collected following established best practices, which rely on photointerpretation of high-resolution imagery to statistically estimate the extent of stable and change classes (e.g., stable forest, deforestation, degradation). This process requires the distribution of thousands of sample plots across each jurisdiction, with each plot labeled according to the applicable national forest definition—typically based on canopy cover thresholds within a defined area (for example, 30% canopy cover within 0.5 hectares).
- This represents a substantial manual effort, often requiring several months of work to accurately annotate all plots. Interpreters must classify each plot and be trained to apply definitions consistently across diverse landscape types.
- The task is further complicated in jurisdictions with low canopy cover thresholds, naturally sparse vegetation, frequent cloud cover, or limited availability of suitable high-resolution imagery.
Forest Cover Benchmark Map Challenges
- Many of the same difficulties encountered during manual photointerpretation apply to the creation of FCBMs. Sparse or heterogeneous canopies, seasonal variation, and divergent forest types complicate image classification. Separability between forest and non-forest classes is particularly challenging in jurisdictions where sparse forests predominate, increasing omission and commission errors. Limited availability of high-quality reference data and cloud-free imagery further constrains model training, validation, and consistency across jurisdictions.
- VMD0055 imposes strict accuracy thresholds for jurisdictional FCBMs, which are difficult to achieve in certain geographies and forest types. Accuracy is evaluated using sample-based comparison of mapped classes against observed data. Two accuracy components are required:
- Deforestation (start to end of HRP): Minimum 70% accuracy
- Forest cover at end of HRP: Minimum 90% accuracy
- In jurisdictions where at least 50% of forest areas have canopy cover below 50%, the following reduced accuracy thresholds apply:
- Deforestation: ≥60% accuracy
- Forest: ≥80% accuracy
Meeting these standards is challenging in heterogeneous landscapes, areas with low canopy cover forest definitions, and regions affected by persistent cloud cover or limited reference data.
Risk Mapping Challenges
- Modeling risk of deforestation in a spatially explicit manner under VMD0055 is a time-intensive process that typically requires at least three months per jurisdiction. The work involves extensive research, data collection, and iterative testing before a final map can be produced.
- Identifying relevant risk factors for deforestation requires literature review and careful evaluation of which drivers are most applicable within each jurisdiction. Once factors are identified, appropriate geospatial data must be located, compiled, and formatted for analysis.
- Developing the risk model involves testing the predictive power of each variable and constructing statistical models that capture empirical relationships between risk factors and observed deforestation. Model development is inherently iterative, requiring repeated testing, refinement, and validation to determine which configuration performs best against out-of-sample data.
Data Review and Feedback Challenges
- All developed datasets—including Forest Cover Benchmark Maps, activity data, and risk maps—are subject to third-party review. When reviewers identify major findings or inconsistencies, the data developer must implement corrections. These revisions often require reprocessing, revalidation, and documentation updates, extending overall timelines.
- Provisional datasets are opened for stakeholder feedback, which adds another layer of review and iteration. Feedback must be collected, organized, and evaluated for relevance and technical merit. Substantive comments are then referred to the data development team for potential incorporation into updated versions. This process extends timelines but often greatly improves data quality.
Related resources:
View REDD Case Studies
REDD+ Project for Caribbean Guatemala: The Conservation Coast
From Ridge to Reef: Conservation and Community in Guatemala
FUNDAECO is uniting rural Guatemalan communities to conserve trees, expand healthcare access, and protect precious biodiversity.
Corazón Verde del Chaco Project
Protecting the Beating Green Heart of South America in Paraguay
While ecological efforts in the neighboring Amazon receive more global attention, the Gran Chaco is no less important. More than twice the size of California, it is South America’s second-largest…