Page 35 - 2417

Chaib et al. Carbon Footprints 2026, 5, 23 Page 11 of 23

Table 3. Classification accuracy for the 2022 MapBiomas 10-m land-cover map and global reference
Class Sample size (n) User’s accuracy Overall map accuracy Global reference for overall map accuracy

Natural forest 100 98% ESA WorldCover 10 m 2020 v100: 74.4% [58]
94%
Anthropic use 100 90% Esri 10 m land use/land cover 2020: 75% [59]

changes, the role of the LVI within the applied framework is to establish initial eligibility and contextualize
additionality at project entry [43,52] . The temporal dynamics of land-use outcomes are instead captured through
the annually updated ex post baseline, which operates independently of the LVI and reflects observed
deforestation behavior in the reference region . At the same time, socio-economic constraints and
[43]
climate-related impacts on livelihoods, such as reduced productive capacity or delayed recovery of
agricultural systems, often persist beyond a single year, reinforcing the relevance of vulnerability as an entry
condition [36,38,39] . Under the applied standard, project cycles operate over defined timeframes (e.g., 10 years),
allowing vulnerability to function as an entry condition while enabling its reassessment in subsequent cycles
as socio-economic conditions evolve.

Accuracy assessment and dynamic deforestation baseline
The independent validation of the 2022 MapBiomas 10-m land-cover map using 200 visually interpreted
points yielded an overall accuracy of 94%, with user’s accuracy of 98% for natural forest and 90% for
anthropic classes [Table 3]. These values exceed the class-level user’s accuracy threshold set by the applied
methodology for 10-m products without published regional validation (above 90%) and are within or above
the range of forest and overall accuracies reported for widely used global 10-m land-cover datasets such as
ESA WorldCover and Esri Land Cover [58,59] . Binomial 95% confidence intervals for class-level user’s accuracy
were ±2.7% points for natural forest and ±5.9% points for anthropic classes, confirming the adequacy of the
sample size. Although uncertainty was higher for anthropic classes, confidence intervals remained
sufficiently narrow to support the application of the map within the adopted baseline-adjustment framework.

Across the 47 reference units (Quilombola territories and agrarian settlements), encompassing 531,272 ha of
natural forest in 2022, the weighted mean annual deforestation rate between 2022 and 2023 was 0.74%.
Classification error was accounted for using the probability-based area-adjustment procedure defined in the
avoided deforestation framework , based on the overall map accuracy and the class-specific confusion
[42]
matrix structure. Following this adjustment, the baseline deforestation rate applied to Curiaú was
conservatively rounded to 0.70%.

In 2022, Curiaú’s forest area was estimated at 915 hectares through remote sensing. Based on this dynamic
baseline, an expected deforestation of approximately 6.4 hectares was projected for 2023 in the absence of
targeted interventions. The spatial pattern of 2022-2023 forest loss in a subset of the reference smallholder
and traditional territories is shown in Figure 4.

The official deforestation data for the region of comparison come from PRODES (Satellite Monitoring
Program for Deforestation in the Brazilian Legal Amazon). PRODES provides annual clear-cut deforestation
estimates for the entire Legal Amazon since 1988, based on interpretation of Landsat, CBERS and Sentinel
satellite imagery. For 2022-2023, the published deforestation was approximately 9,000 km , equivalent to
2
approximately 0.18% of the biome’s total area . This macro-scale percentage is not directly comparable to
[60]
the proportional 0.74% reference-region rate used in this study, as PRODES measures absolute annual
clear-cutting across the entire Legal Amazon rather than proportional forest loss within mosaics of
smallholder and traditional territories. For contextual comparison at the jurisdictional scale relevant to the
study area, PRODES reports that, in the same monitoring year, the state of Amapá accounted for only