A Systemica realizou recentemente um estudo técnico para avaliar o impacto da mudança metodológica nos projetos REDD AUD localizados no Brasil, no contexto do mercado voluntário de carbono. Para isso, foram aplicadas a versão mais recente da nova metodologia REDD da Verra (VM0048), seu módulo AUD (VMD0055) e a ferramenta de alocação correspondente (VT0007), a fim de derivar dois modelos distintos para mapeamento e alocação do risco de desmatamento não planejado. O primeiro modelo segue o algoritmo de referência do VT0007, conforme a versão mais recente publicada em 21 de fevereiro de 2024. O segundo modelo, chamado de modelo alternativo, foi desenvolvido adaptando uma versão preliminar do VT0007, datada de 21 de abril de 2021, para torná-la compatível com a versão atual da ferramenta.
A principal diferença entre os dois modelos está na variável utilizada para determinar espacialmente o risco de desmatamento futuro. O modelo de referência usa a distância até áreas não florestadas como variável preditiva, enquanto o modelo alternativo considera a taxa local de desmatamento para essa mesma função. Assim, embora os modelos sejam semelhantes em diversos aspectos, a diferença fundamental nas variáveis preditivas pode resultar em estimativas distintas para projetos individuais.
Although the distance to non-forest is a very effective predictive variable, reducing the number of model parameters and enabling computational gains in implementation and processing, our experience shows that the use of a density-like variable, such as the local deforestation rate, implies gains in the robustness of the model, making it less sensitive to errors in land use classification, for example.
Despite their differences, both models presented very similar accuracy at the jurisdictional level, according to the comparative procedure determined by VT0007. In this context, it is worth noting that VT0007 itself prescribes the construction of at least two alternative risk maps for comparison with the benchmark map, with the most accurate among the models being chosen as the baseline for the jurisdiction. This highlights the importance of preparing alternative risk maps, aiming to understand the possible range of variation in the predicted baseline resulting from all the potential maps to be proposed by Verra’s data providers, who will ultimately be responsible for determining the adopted risk mapping algorithm.
In our study, data from Brazilian AUD projects available on the VERRA registry system were compiled, considering only those that were registered, in the process of being registered or in the process of being verified. Additionally, projects that did not have vector data of the project area available on the platform, or that presented a vector that differed from the project area defined in the Project Description (PD), were excluded from the analysis. As a result, a total number of 20 AUD projects were included in our analysis after applying the exclusion criteria. In all these projects, discontinued methodologies, mostly VM0007 and VM0015, were employed for baseline quantification.
The benchmark and alternative risk map models have been used to allocate the predicted deforestation for the next 6 years in each project analyzed. These results were then compared with the baselines presented in the examined PDs. It is important to emphasize that the risk maps derived for this study are only an estimate of the maps being developed by Verra’s data providers, which will most likely be produced with different datasets than those used here.
The figure below shows the average over projects of the predicted unplanned deforestation in three scenarios: old REDD methodologies, benchmark and alternative risk maps.
As expected, the discontinued AUD methodologies, which are based on the idea of a localized reference area, predicted greater unplanned deforestation when compared to the new AUD methodology, where risk maps and deforestation rates are assessed at the jurisdictional level. Overall, the benchmark model presented a loss of 42% of predicted deforestation in relation to the PD values, while the alternative model shows an even greater loss of 69% when compared to the old methodologies.
In view of these large potential losses in credit generation, many projects are expected to lose economic viability at current VCU prices and, therefore, have a higher risk of non-permanence due to the transition to the new REDD methodology. Despite this negative impact on ongoing projects, it is expected that the proposed methodological changes, which remove the responsibility from project developers for generating the risk maps, will ensure greater quantification integrity to AUD projects, as the jurisdictional approach to deforestation prediction provides greater consistency and conservativeness to baseline determination, while minimizing challenges of harmonization between private projects and jurisdictional programs.
In conclusion, our analysis demonstrated that the baselines derived from the old AUD methodologies are significantly different from two deforestation-prediction models based on Verra’s new REDD methodology (VM0048). We believe that these methodological changes are fundamental for the generation of high integrity carbon credits, aligned with the Core Carbon Principles (CCP). However, it is not yet possible to fully anticipate the impacts of methodological changes on the supply and price of carbon credits from AUD projects, as Verra’s endorsed risk maps have not yet been published, which is expected to happen later this year.
