The declaration of the United Nations Decade on Ecosystem Restoration (2021–2030; https://www.decadeonrestoration.org), brought to light humanity's urgency to restore degraded ecosystems while the Bonn Challenge (https://www.bonnchallenge.org) and the New York Declaration on Forests (https://forestdeclaration.org) set a global target of restorating 350 million hectares by 2030. In parallel, national and regional initiatives are also advancing large-scale restoration. For instance, Brazil’s National Plan for Native Vegetation Recovery (PLANAVEG) seeks to restore 12 Mha of native vegetation by 2030 (Brasil, 2025), while the Atlantic Forest Restoration Pact (hereafter, PACTO) aims to restore 15 Mha of degraded forest in the Atlantic Forest by 2050 (Crouzeilles et al., 2019; Toto et al., 2025; https://pactomataatlantica.org.br).As we reach the midpoint of the UN Decade on Ecosystem Restoration, regularly mapping large-scale forest recovery is crucial for transparently assessing restoration progress using replicable methods. Understanding the spatial dynamics of forest recovery helps prevent the loss of regenerated forests – a common issue in tropical regions – and supports restoration goals (Piffer et al., 2022; Reid et al., 2019). These analyses also help prioritize areas with high potential for natural regeneration, a cost-effective alternative to active restoration (Brancalion et al., 2019a), and identify regions where recovery benefits are offset by losses of old-growth or secondary forests, enabling more targeted conservation and enforcement strategies (Chazdon and Guariguata, 2016; Mansourian and Vallauri, 2014).

Despite harboring most of the world’s terrestrial biodiversity and providing vital ecosystem services, such as carbon storage and water supply, tropical forests have been extensively degraded over the past half-century, a process that is still ongoing (Zalles et al., 2021; Amaral et al., 2025). These ecosystems, however, retain high restoration potential (Brancalion et al., 2019b). A global analysis identified 215 Mha of tropical forest areas with high regeneration potential, with Brazil accounting for 20.3% of this total, primarily in the southern Amazon and Atlantic Forest – highlighting the strategic value of prioritizing restoration in these regions (Williams et al., 2024).

The Atlantic Forest with 95% of its area in Brazil, extends into Argentina and Paraguay and harbors a high number of endemic and threatened species (Myers et al., 2000; Muylaert et al., 2018). However, only 23% of its original forest cover remains, mostly embedded within anthropogenic landscapes (Vancine et al., 2024). Numerous studies highlight its potential for natural regeneration (Crouzeilles et al., 2020; de Rezende et al., 2015; Latawiec et al., 2015; Molin et al., 2018; Piffer et al., 2022; Rosa et al., 2021; Tambosi et al., 2014; Tonetti et al., 2022); and it has been designated in 2022 a UN World Restoration Flagship due to multiple restoration initiatives (https://www.decadeonrestoration.org/world-restoration-flagships). This recognition stems from the forest restoration efforts led by the Atlantic Forest Trinational Pact, acknowledged as one of the most promising and globally significant initiatives (Crouzeilles et al., 2019; Toto et al., 2025).

Given these advances, it is crucial to provide an updated, comprehensive, spatially explicit assessment of recent forest recovery across the entire Atlantic Forest domain. To fill this gap, our study uses annual high-resolution (∼30 m) land cover data to map forest recovery between 2011 and 2021 and spatially evaluate the persistence of these recovered areas until 2023. By doing that, we aim to inform restoration policies, guide conservation strategies, and support ongoing regional and global restoration commitments. We also assess recovery within Protected Areas (PAs) and community-managed lands. As in other tropical regions, recovery in the Atlantic Forest is largely driven by natural regeneration (Latawiec et al., 2015; Williams et al., 2024). However, we use the term “forest recovery” broadly to refer to new forest cover established during the study period, without distinguishing between natural regeneration and active restoration.

We build on a previous assessment of forest dynamics in the Brazilian Atlantic Forest (hereafter BAF; Crouzeilles et al., 2019), and based our analysis on land cover maps from MapBiomas Collection 9, which provides annual land cover mapping for Brazil from 1985 to 2023 (∼30 m pixel resolution; https://brasil.mapbiomas.org/en/). The overall accuracy for the BAF map is 91.67% (https://brasil.mapbiomas.org/estatistica-de-acuracia/colecao-9/). To estimate recovery, we considered only pixels classified as (1) “Forest Formation,” (2) “Mangrove,” or (3) “Wooded Sandbank Vegetation”. We followed an approach similar to Crouzeilles et al. (2019). Under that criteria, for an area to be classified as recovered, it must be (i) previously classified as agriculture, pasture, planted forests or mosaic of uses for at least five consecutive years; (ii) classified as forest for at least three consecutive years; (iii) contain at least five connected pixels (∼0.45 ha) also classified as recovered forest; and (iv) to be classified as forest in the last year of analysis. Our analysis covers the period from 2011 to 2021, as 2023 is the most recent year of land cover data available, and we applied a three-year minimum forest cover threshold as an inclusion criterion to consider a pixel as a regenerated area.

Alongside estimates of total forest gain per municipality (Fig. 1A), we quantified the total loss of recovered forests (Fig. 1B), the proportion of loss relative to gains over the period (Fig. 1C), and the proportion lost in relation to forest cover in the baseline year, 2010 (Fig. 1D). To detect municipalities with spatially aggregated patterns of recovering forest gain or loss, we applied a modified version of Lee’s L statistic to calculate a Local L statistic and pseudo p-value, indicated for measuring spatial association of continuous variables (Lin et al., 2020). These correlations were conducted using the lee.mc function from the spdep R package (Bivand and Wong, 2018), following Oliveira-Dalland et al. (2022). To identify regions with significant spatial association between variables, this analysis allowed us to classify clusters as 'High' or 'Low' based on values significantly higher or lower than expected for the two variables. For instance, a 'High/High' cluster indicates that both variables (e.g., forest cover in 2010 and recovery) exhibit high values and are significantly spatially clustered, while 'High/Low' means that the first variable has a high value but is surrounded by areas where the second variable has a low value, and so on. The statistical significance of these groupings was determined by the pseudo p-value. We assessed spatial associations between the following variable pairs: (1) forest cover in the baseline year (2010) and recovery from 2011 to 2021 (Fig. 1E); (2) baseline forest cover and loss of areas that had recovered during the study period (Fig. 1F); and (3) forest recovery followed by subsequent loss (Fig. 1G). These analyses allowed us to identify regions where forest dynamics (e.g., cover, recover, or loss) were unusually high or low compared to surrounding regions. From the total of recovered forests, we quantified the amount in Brazil’s Protected Areas (PAs), Indigenous Lands, Agrarian Reform Settlements of small farmers, and Quilombolas Lands (Afro-Brazilian traditional communities). The databases used and their respective sources are presented in Table S1, and spatial operations were performed using the terra package in R (Hijmans, 2023).

Fig. 1.

A - Forest recovery (hectares) in Brazilian municipalities during the study period (2011 to 2021) and that persisted until 2023. B - Total loss of forest that was recovered during the study period. C - Proportional loss of recovered forest relative to the recovered area. D - Proportion of recovered forest compared to forest cover in the initial year (2010) considering only municipalities above the 50th percentile of forest cover in 2010 (>4075 ha). Municipalities below this threshold are shown in gray. E - Spatial associations between total forest cover in 2010 and recovered forests were classified into categories (e.g., High / High indicates areas where both variables are high; High/Low where 2010 forest cover is high but recovery is low; and so on). F - Total forest cover in 2010 and recovered forest loss. G - Recovered forests and deforestation of recovered forests. Bivariate spatial correlations in E to G were calculated using the Lee’s spatial correlation coefficient.

We estimated that 1.67 million hectares (Mha) of forest recovered in the Brazilian Atlantic Forest (BAF) between 2011 and 2021 and remained stable through 2023. This area represents approximately 4% of the total forest cover in the domain (Fig. 1A). The amount of forest recovery observed over the past decade is equivalent to half of the forest that regenerated between 1985 and 2019 and persisted until 2019 (Piffer et al., 2022), highlighting a strong potential for further expansion of restored areas. The largest concentrations of recovered forest were found in the states of Minas Gerais (26.4% of the total), Paraná (18.6%), Bahia (12.9%), and São Paulo (12.7%; Fig. 1A).

Recovered forests are concentrated in large clusters, mostly in municipalities that already had high forest cover (Fig. 1E). Many of these clusters are in hilly areas with limited suitability for mechanized agriculture, where well-connected forest patches facilitate seed dispersal and forest recovery on abandoned lands (de Rezende et al., 2015; Latawiec et al., 2015; Piffer et al., 2022; Tonetti et al., 2022). However, we also identified areas with high forest cover but low forest recovery – mainly in Rio Grande do Sul, Santa Catarina, and parts of Minas Gerais (Fig. 1E, light blue) – suggesting strong agricultural pressure that likely restricts regeneration (Latawiec et al., 2015). In contrast, small clusters of municipalities with low forest cover in 2010 showed notable recovery (Fig. 1E, yellow), indicating that restoration efforts can succeed even in degraded landscapes and enhance ecosystem services such as soil protection, erosion control, and carbon sequestration (Chazdon and Guariguata, 2016).

Most forest recovery (75.2%) occurred in areas classified as “Mosaic of Uses” in 2010 – landscapes where pasture and agriculture could not be distinguished (https://brasil.mapbiomas.org/en/). The second most common preceding land cover was pasture (16%). As reported in other studies, our findings suggest that forest recovery largely occurred on abandoned and degraded pastures (Crouzeilles et al., 2019; de Rezende et al., 2015; Latawiec et al., 2015; Molin et al., 2018; Piffer et al., 2022; Rosa et al., 2021). Promoting forest recovery in these areas is central to large-scale initiatives like PLANAVEG, which aims to restore 12 million hectares (Brasil, 2025), and PACTO (Crouzeilles et al., 2020). Agricultural lands (coffee, soy, sugarcane, temporary crops) accounted for only 1.1% of prior land use, and planted forests for 0.62%. The remaining 7% of recovered areas were classified as forests in 2010, which indicates they were deforested and later regenerated during the study period.

Regarding the PAs, 141.3 thousand hectares (Kha) were recovered within sustainable-use areas, such as Environmental Protection Areas (equivalent to IUCN category V; Table 1). These areas, designed to reconcile agriculture and other economic activities with biodiversity conservation, often enable forest recovery to occur alongside human use (Gonçalves-Souza et al., 2021). A smaller portion (23.9 Kha) of recovery was found in strict protected areas (IUCN categories I–IV), such as National and State Parks, where forest cover is typically higher and more effectively protected (Tonetti et al., 2024). Community-managed lands, including Agrarian Reform Settlements, Indigenous, and Quilombola lands, accounted for 50.3 Kha of recovered forests (Table 1).

About 50 Kha of recovery annually did not persist until 2023 (Fig. 2). In total, 568 Kha of recovered forest was later lost. These losses are spatially clustered, often occurring in areas with high forest cover in 2010 (Fig. 1F) and significant recovery (Fig. 1G). This pattern suggests that, despite the high potential for forest recovery, some regions experience low forest persistence – losing up to 64% of the recovered forests in certain areas (Fig. 1C). Notably, clusters of forest loss also appear in landscapes with low forest cover or recovery (Figs. 1F and G, yellow), posing a threat to biodiversity and ecosystem services, especially where forest cover is most needed. Previous studies have also reported that many regenerated areas in the BAF are later deforested (Piffer et al., 2022; Rosa et al., 2021). These findings highlight the urgent need to promote the long-term persistence of regenerating forests to ensure lasting benefits such as carbon sequestration and biodiversity conservation (Tonetti et al., 2022; Brancalion et al., 2024).

Fig. 2.

Area of recovered forest per Brazilian state (A), and annual areas of recovered forest for each year of the studied period (B) represented by the blue portions of the bars. Portions of recovered forests that were subsequently lost and did not persist until 2023 are in red (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article).

The BAF was previously identified as a global restoration hotspot, with estimates of 673–740 Kha of native forest recovery between 2011 and 2015, and a projected total of 1.35 1.48 Mha by 2020 (Crouzeilles et al., 2019), which is confirmed by our results. This ongoing recovery suggests that the high potential for forest regeneration regeneration in BAF can significantly contribute to achieving major restoration targets. However, the high amounts of forest loss associated with low temporal persistence compromises the potential of recovered areas to sustain biodiversity and ecosystem services and threatens the achievement of restoration goals (Chazdon et al., 2016; Piffer et al., 2022; Resende et al., 2024).

To fully realize the restoration and conservation potential of the BAF, it is essential to safeguard young regenerating and old growth forest patches in a combined and policy-wise manner (Chaves et al., 2022; Resende et al., 2024). The domain is protected by the Atlantic Forest Law (AFL, Federal Law No 11.428; Brasil, 2006), which prohibits the cutting of mature forests and those in an advanced stage of regeneration, while allowing the cutting of vegetation in an intermediate or initial stage under specific technical and environmental criteria. Although the AFL is considered a milestone for the protection of the BAF, the criteria used to classify forests according to their regeneration stage remain vague (Resende et al., 2024). In addition, it should also prevent deforestation of early-stage regenerating forests, which play a crucial role in enhancing landscape connectivity through time in dynamic landscapes, especially relevant in this highly fragmented domain (Vancine et al., 2024).

Promoting the persistence of recovered forests is essential for biodiversity conservation and ecosystem service provision at the landscape scale (Resende et al., 2024). Beyond strengthening environmental legislation, financial mechanisms – such as Payments for Ecosystem Services (PES), carbon and biodiversity credits – can incentivize forest recovery (Brancalion et al., 2024; Barbosa et al., 2025). PES schemes should also support the long-term persistence, ecological quality, and multifunctionality of regenerating forests (Ruggiero et al., 2021). To guide and evaluate these efforts, continuous monitoring of recovering forests enables the creation of spatial databases capturing forest regrowth across ages and environments, supporting large-scale remote sensing–based restoration monitoring. In this context, the PACTO-led restoration governance model offers a valuable example (Pinto et al., 2014; Toto et al., 2025).

The future of Atlantic Forest restoration depends on approaches that combine the natural regrowth of native vegetation with active restoration measures (Toto et al., 2025). Regions with suitable conditions and existing remnant vegetation can leverage on this passive recovery potential to scale up restoration efforts across the domain. However, highly degraded areas – often characterized by extensive, long-term agricultural occupation with little to no native vegetation nearby – will require active restoration strategies (Chazdon and Guariguata, 2016). These areas are typically those with minimal native vegetation cover and the greatest risk of ecosystem service collapse. Moreover, many degraded forests suffer from issues such as excessive liana infestation and the loss of key native species, requiring intervention through control and enrichment planting with native species that have been locally extirpated. Such measures are essential in regions where natural regeneration alone is insufficient to prevent ecosystem collapse (Chazdon and Guariguata, 2016).

As the world’s largest tropical forest nation, Brazil is uniquely positioned to lead global efforts to address the climate and biodiversity crises (Barbosa et al., 2025). As it prepares to host COP30 in Belém, the country has a chance to showcase large-scale forest recovery. The BAF offers a compelling case for demonstrating how restoration can succeed in fragmented, densely populated landscapes. By leveraging nature’s regenerative capacity, the BAF can serve as a global model for science-based restoration, offering critical lessons for restoring biodiversity and ecosystem services in human-modified regions.