Quantifying landscape fragmentation and forest carbon dynamics over 35 years in the Brazilian Atlantic Forest

We used MapBiomas forest maps to assess fragmentation status of AF landscape in 1985 and 2020 at a 30 meter spatial resolution. We categorized forest landscape into seven fragmentation classes (Vogt et al 2007, Soille and Vogt 2009, Riitters and Vogt 2023): core, perforation, edge, bridge, loop, branch, and islet, ranging from lowest to highest fragmentation level (figure 1(a)). Core is the interior forest area excluding the perimeter, islets are small, disjointed forest patches, edges constitute the external forest perimeter, and perforation refers to the internal forest perimeter. The connection classes include bridge, loop, and branch. Bridges connect different core areas, loops connect the same core area, and branches are forest areas linked at one end to perforation, edge, bridge, or loop.

Zoom In Zoom Out Reset image size

After define fragmentation classes, we developed a temporal pathway outlining forest processes. Shifting from core to more fragmented classes indicate primary degradation, while non-core classes into higher fragmentation indicate secondary degradation. Any class transitioning to non-forest is deforestation, while the opposite trend, regeneration. 'Unchanged' forests remained in the same class over time. (figure 1(b)).

Located in a complex landscape formed by heterogeneous environmental conditions and rich forest composition (figure 2). This region is home to over 72% of the Brazilian population and three major South American urban hubs, contributing with approximately 70% of the Brazilian GDP (SOS Mata Atlântica 2022). According to the MapBiomas, there were 337 663 km² of AF cover in 1985 (MapBiomas 2023). Most remaining forests exist as fragments, often smaller than 1 km², and occasionally isolated. Large forest patches persist in areas historically unsuitable for agriculture and human habitation (Ranta et al 1998, Ribeiro et al 2009).

Zoom In Zoom Out Reset image size

We used MapBiomas Collection 6.0 (https://brasil.mapbiomas.org/en/downloads/) Land-use and Land-cover data from 1985 and 2020 (MapBiomas 2023). Based on Silva-Junior et al (2020b) method, we mapped the extent of secondary forests (1986–2020) (https://github.com/celsohlsj/gee_brazil_sv). We then produced maps depicting the old-growth and secondary forests cover between the years 1985 and 2020.

We considered secondary forests as forest pixels that were replaced by anthropic classes (e.g. agriculture and pasture) and then classified again as forest during the period 1986–2020. Old-growth forests were defined as forests that remained in the forest class throughout the time series. Consequently, we created three binary maps: one representing the forest in 1985, another for 2020, both inclusive of old-growth and secondary forests. The third map was built using the map algebra approach by removing the standing secondary forests accumulated between 1986 and 2020 in the 2020 forest map (figure 3, step 1). This approach allowed us to evaluate the impact of secondary forests on fragmentation metrics and connectivity within the AF landscape.

Zoom In Zoom Out Reset image size

To access fragmentation across the entire AF biome boundary, we classified the binary maps with GUIDOS Toolbox MSPA program (Soille and Vogt 2009, Vogt and Riitters 2017). This method utilizes mathematical morphological operators to describe image component connections and geometry. The software automatically assigned each pixel to one of seven exclusive fragmentation types, as depicted in figure 1(a). The MSPA identifies fragmentation classes based on the "edge width," measured from the forest's edge to its interior, in pixels. This determines the spatial configuration of the other classes across the landscape. Based on previous studies, we adopted a 300 meter edge width ⁹ in the MSPA to account for significant edge effects, such as increased tree mortality, forest disturbance, wind turbulence, altered tree recruitment, reduced canopy height, and forest fires in the understory (Broadbent et al 2008, Laurance et al 2011, Numata et al 2011, 2017).

After producing maps containing the fragmentation classes, we calculated the area of each class by forest types (IBGE 2004), protected areas, defined as Strict Use: designated only for biodiversity conservation; Sustainable Use: allows limited resource exploration while designating biodiversity conservation. Furthermore, we analysed these conservation units by jurisdiction responsible for protecting these areas (municipal, state, or federal) (MMA 2023). We also accounted for Regularised Indigenous Lands (FUNAI 2023) (figure 3, step 2).

Based on the 1985 and 2020 fragmentation maps with old-growth and secondary forests, we generate a transition map using mapping algebra functions. This process enabled the pixel-by-pixel evaluation of the fragmentation classes and the quantification of their respective areas (figure 3, step 3).

We use the direct matrix approach (Bucki et al 2012) to identify and map dynamic degradation, tracking transitions between fragmentation classes over time. This approach uses fragmentation to access degradation processes, revealing typical paths of change, providing information about forest conditions.

We applied the method used by Silva-Junior et al (2020a) to identify forest edges using all maps from 1986 to 2020. 34 forest edge maps were created, with a 120 meter edge width ¹⁰ . Using annual MapBiomas maps, changes from forest to non-forest cover since 1986 were identified, enabling deforestation mapping since 1986 (figure 3, step 4) (Silva-Junior et al 2020b, Silveira et al 2022) (https://github.com/celsohlsj/amazonia_deforestation).

Our study showed at the biome-scale estimates of C emissions and removals using the map from Brazil's 3rd Greenhouse Gas Emissions Report (MCTI 2016) processed by Google Earth Engine platform. These calculations accounted for emissions from forest edges, deforested areas, and removals from regenerating secondary forests.

In 1985, only 14% (47 195.4 km²) of the remaining forests had core areas. These areas were reduced to 12.5% (40 436.7 km²) by 2020 (see figure 4 for a visual depiction of the forest landscape). The predominant categories in both years remained bridge and islet, making up 72.5% (244 684.8 km²) of the remaining forest in 1985 and 73.5% (238 004.4 km²) in 2020, respectively. The remaining forests, categorized as non-core areas (perforation, edge, loop, and branch), collectively comprised 13.6% (45 783.6 km²) of the remaining forest area in 1985, with a minor rise to 14% in 2020 (45 374.0 km²) (figure 4, table 1).

Zoom In Zoom Out Reset image size

Between 1986 and 2020, secondary AF area accounted for 99 450.5 km². The core class represented 1.4% of the forest, whereas bridge and islet classes dominated with 88.8%, leaving 12.7% for the remaining classes (table 1). Secondary forests contributed to all fragmentation classes (table 1).

Tables 2(a) and (b) detail pixel-based transitions from 1985 to 2020. Core forest areas often shift to the bridge class, while the bridge class tend to shift to deforestation or to transition to the islet class. Bridge class emerges as a crucial transition target, transitioning to classes like perforation and edge, as well as classes like loop and branch (table 2(a)). The islet class exhibits limited potential for transitioning to non-forest compared to other fragmentation classes. Similarly, the background class frequently transitions to the islet class over other fragmentation classes (table 2(a)).

We analysed forest change processes using fragmentation class transitions: primary degradation, secondary degradation, deforestation, regeneration, and unchanged areas (figure 5(a)). Additionally, we mapped forest change across the entire biome extent (figure 5(b)). Of the forest transitions occurring in about 10% of the total biome extent (20 227 003.3 km²), regeneration covered the largest extent, corresponding to 4.3%, followed by deforestation at 3.7%, secondary degradation at 0.9%, and primary degradation at 0.5% (figure 5(a)).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Observed transition classes vary across the biome, showing degradation and deforestation frontiers, regeneration frontiers, and stable frontiers in different AF locations (figures 5(b1)–(b4)). Examples include forest regeneration areas (figure 5(b.1)), consolidated secondary degradation and deforestation areas (figure 5(b.2)), small-scale deforestation and primary degradation within an unchanged forest matrix (figure 5(b.3)), and primary degradation areas with secondary degradation and deforestation (figure 5(b.4)).

We assessed fragmentation classes in strict use, sustainable use, and indigenous conservation units. Strict use mainly includes core, edge, and bridge forests. Sustainable use and indigenous lands contain core, bridge, and islet classes. Regularized Indigenous Lands preserve more core forest (42.9%) than Sustainable Use Conservation Units (26%) (table 3(a)). The analysis of protected areas showed that federal conservation units had 47.1% of core forests, 27.9% of bridges, and 9.1% of islets. State conservation units exhibit 38.8% of core forests, 36.5% of bridges, and 11.5% of islets. Municipal conservation units had 44.2% of bridges, 22.6% of core forests, and 16.6% of islets (table 3(b)).

We found that Bridge and islet classes were dominant in 1985 and 2020. In the AF biome, Rainforests had the largest core area (26.7% Dense Rainforest, 8.3% Araucaria-Mixed Rainforest, 7.1% Open Rainforest), while Seasonally Dry formations had the lowest (5.8% Seasonally Dry Semideciduous Forest, 3.7% Seasonally Dry Deciduous Forest). Over 35 years, proportions of the different fragmentation classes stayed almost stable. The core area of Open Rainforest class increased by 4.4%, Araucaria-Mixed Rainforest reduced by 3.7%. The bridge class area in Araucaria-Mixed Rainforest and Seasonally Dry Deciduous Forest shrank (4.9% and 2%, respectively), while this fragmentation class in Open Rainforest expanded by 3%. Islet area increased for Seasonally Dry Deciduous Forest (2.2%) and Araucaria-Mixed Rainforest (7.9%) but was reduced by 16.7% in Open Rainforest. Secondary forests contributed to all fragmentation classes, expanding the area of bridges and islets (table 4).

Our estimates show that secondary forest regeneration removed 1346 TgCO₂, from 1985 to 2020, offsetting around 82.62% of the total emissions in the AF biome during the analysed period. This estimate excluded all emissions from deforestation before this period. Our findings demonstrated parity between C emissions from deforestation (818 TgCO₂) and those arising from the forest edge effect (810 TgCO₂) (figure 6(a)), emphasizing the substantial impact of both processes on C emissions. This suggests that despite long-term trends of deforestation reduction, the AF continues to lose C stocks due to the edge effect. Consequently, if fragmentation continues, the dominant source of emissions may no longer be attributed to deforestation, but rather to edge effects.

Zoom In Zoom Out Reset image size

We calculated net emissions by considering emissions from both deforestation and edge effects, and removals from secondary forests in each AF state. The consistent pattern of equivalence between deforestation and edge emissions held for each State (figure 6(b)). We showed the overall net balance, for the period analysed, indicates a source of 283 TgCO₂ to the atmosphere. Individually assessing each state, only five (from a total of 15)—Santa Catarina, Bahia, Paraná, Mato Grosso do Sul, and Rio de Janeiro—were unable to have secondary forest removals completely offsetting emissions from deforestation and edge effects. In contrast, Pernambuco, Alagoas, and Minas Gerais outperformed other states in offsetting recent deforestation emissions (figure 5(b)).

We spatially analysed emissions from deforestation, edge effect, removals from regeneration, and the resulting net balance (figures 7(a)–(d)). Net emissions were characterized by the dominance of removals in low latitudes in the northeast region, in the central part of the biome, mainly in the Midwest portion and in the southern regions. We observed prominent emissions in the southern region of Bahia, the northeast region of Minas Gerais, and in eastern São Paulo, Paraná, Santa Catarina, and Rio Grande do Sul, which are areas near metropolitan regions. (figure 7(d)).

Zoom In Zoom Out Reset image size

We provided a comprehensive biome-scale perspective on AF fragmentation using the MSPA approach (Vogt et al 2007), updating the previous assessment by Ribeiro et al (2009) and Rezende et al (2018). Our results show a decrease in total mapped forest cover from 1985 (337 663.8 km²) to 2020 (323 815.1 km²) (table 2). Although the proportions of the fragmentation classes remained consistent, there was a loss of 6758.7 km² in the core area, a reduction in the bridge area of 14 352.3 km² and an increase of 7671.9 km² in the islet area (table 1).

Core areas provide essential habitats that support larger populations, enabling species to thrive and maintain healthy population dynamics (Gaston et al 2006, Belote et al 2017). These large, mature fragments are vital for preserving taxonomic, genetic, and functional diversity (Smith et al 2021, Schweizer et al 2022). Moreover, old-growth forests enhance seed input and facilitate the recolonization of neighbouring fragments (Rodrigues et al 2009).

The bridge class comprises most remaining forests (38%), prone to converting into loops, branches, or islets, intensifying fragmentation, and deforestation. Bridges serve as transitional boundaries, offering opportunities for regeneration into core forests. They aid in countering fragmentation effects by reducing isolation and enhancing connectivity (Bennett 2003, Wickham and Riitters 2019).

Islets constitute 35.3% of AF and face threats from both nature- and human-related causes. Our transition matrix reveals a risk of substantial loss of forests, as only around 13% transitioned to classes with low fragmentation level. Islets are vital for mitigating the non-forest matrix's impact on ecosystem processes.

Strategically using islets as stepping stones enhances population permeability through the matrix, benefiting endangered species in fragmented forests. This approach strengthens landscape resilience against fragmentation and degradation threats (Baum et al 2004, Antongiovanni and Metzger 2005, Pardini et al 2009, Brancalion et al 2018, Cardoso et al 2022).

Isolation and patch size significantly influence biodiversity in fragmented landscapes (Fahrig 2013). Prioritizing islet integrity and ecological restoration can enhance structural connectivity, benefiting conservation, C goals, and cost reduction (Strassburg et al 2019, Grantham et al 2020, de la Sancha et al 2021).

Around 86% of mapped secondary forests belong to the bridge and islet categories. Their contribution to the 2020 fragmentation classes increases progressively from core to islet, enhancing landscape connectivity. Secondary forests contributed 3.5% of the entire core area mapped in 2020. They serve as crucial stepping stones, aiding species movement (Crouzeilles et al 2017, Matos et al 2020). Strategic placement of secondary forests near remnant patches or creating corridors can further reduce isolation and enhance connectivity (Marshall et al 2022, Schweizer et al 2022, Wills et al 2022).

Landscape planning in the AF should prioritize spatial configuration, connectivity promotion, and core area enhancement and preservation. Understanding fragmentation spatial patterns and processes leading to temporal changes can help identifying highly vulnerable areas, informing effective conservation strategies (Lira et al 2012, Maxwell et al 2020).

Our results underscore the importance of protected areas for core forest preservation and landscape connectivity. Secondary forests play a vital role in the mosaic of Protected and Indigenous areas, highlighting their significance. However, concerns raised due to recent trends in deforestation that were identified following our mapping in 2020. These findings emphasize the necessity for collaboration among federal, state, and municipal authorities in effective forest resource management.

Our findings revealed marked alterations in AF's landscape due to shifts in land use. Native vegetation decreased from 360 000 km² in 1985–333 000 km² by 2022. Activities such as agriculture, and expansion of urbanized regions can be associated to this decline. Pasture areas reduced from 390 000 km² to 300 000 km², while agricultural land doubled from 120 000 km² to 240 000 km². Native vegetation predominantly regrew in former pasture areas (MapBiomas 2023).

Landscape planning is crucial to address declining forest areas. Observing the changes in land cover in AF highlights extensive transitions from pastures to agriculture, mosaic of uses, and native vegetation. These three classes dominated the shifts from pastures to other uses from 1985 to 2022 (MapBiomas 2023). Approximately 65.6% of Brazil's pastures (2010–2018) showed signs of mild, moderate, or severe degradation (Santos et al 2022). Utilizing MSPA to map fragmentation classes is advantageous for managing the AF landscape. Leveraging the regrowth potential in degraded pastures near bridge, loop, and branch areas, which serve as connectors, can facilitate the expansion of connections between forest fragments through natural regeneration.

AF has the potential to adopt crop-livestock-forest systems, replacing degraded pastures and monocultures across extensive areas. These combined systems can prevent additional fragmentation or deforestation (Rodrigues et al 2023). Recognized for its integrated role in enhancing water, energy, and food security, this approach is becoming a crucial strategy to engage stakeholders and local communities in seeking nature-based solutions (Melo et al 2021).

Our analysis also permitted a comprehensive understanding of the components of the net C emissions in the AF biome. Secondary forest regeneration offsets around 82.62% of total emissions. From this total C emissions, deforestation and edge effect contributes equally. In Amazonian forests, edge areas contribute to 37% of deforestation (Silva-Junior et al 2020a), underscoring that the Atlantic Forest exhibits an even more pronounced influence from edges, approaching nearly 50%. This highlights the ongoing C stock loss, despite nearly 90% of the forest land use remained unchanged over the last 35 years. This pattern is mainly associated to the extensive fragmentation of AF landscape, which allows edge effect to become a major contributor to C loss in this biome. Therefore, urgent efforts are needed to reduce deforestation to nearly zero, with forest regeneration aiming at reducing fragmentation and forest edge area, contributing positively to the stability of biome's functions in the future. An additional area of approximately 1.96 million hectares (19 600 km²) of forests would be required to grow in the coming years to offset carbon emissions from 1985 to 2020 in the AF. An area about a third of Belgium.

It is a challenge to reduce emissions from deforestation in tropical forests. Smith et al (2021) found a strong negative relationship between the loss of primary forests and the recovery of secondary forests for countries in the Amazon region. AF stands as a hope spot for regeneration (Rezende et al 2018), with its natural regrowth forests being a cost-effective strategy for restoration goals (Crouzeilles et al 2020).

Our study demonstrates that secondary forests are dominant in the classes bridge and islet, which are more susceptible to deforestation than others with better connectivity. This result confirms the vulnerability of secondary forests to subsequent deforestation. Factors such as steeper slopes, proximity to rivers and existing forests, higher agricultural productivity areas, and lower rural-urban population ratios increases the persistence of regenerated forests (Piffer et al 2022).

Another important point is the need for not only preserve forests in lower fragmentation class, but also those in higher fragmentation classes to increase the resilience of AF functions to climate change. A recent model predicting the impact of climate change on AF above-ground biomass (AGB) showed that the most significant loss of AGB, up to 40%, compared to the baseline value, is likely to occur between latitudes 13° and 20° south (Ferreira et al 2023). Our results showed deforestation and degradation leading to core area loss in southern and south-eastern forests at these latitudes, which may decrease the capacity of forests to counteract the effects of changes in climate.

Over the past 35 years, our findings reveal that despite the AF's success in preserving much mature forest and expanding secondary forest coverage until 2020, deforestation, edge effects, and degradation have eroded connectivity and reduced core areas (Rosa et al 2021). This scenario jeopardizes Brazil's commitment to reforesting 12 million hectares. Additionally, MP 1150/2022 (National Congress 2022), aiming to relax restrictions on native vegetation removal placing core, bridge, and islet classes at risk, protected only by AF law.

If implemented, this measure could authorize deforestation for infrastructure projects, bypassing environmental assessments and ecological compensation requirements (Ribeiro et al 2023). The proposed law might also extend the timeframe for environmental restoration, obstructing financing agencies from enforcing restoration as a precondition for credit (Disclosures 2017).

The dataset used as input to fragmentation mapping relies on MapBiomas Collection 6.0, which maintains an overall accuracy for AF biome by 90,6% ± 2,9% (https://brasil.mapbiomas.org/en/analise-de-acuracia/). No changes to the spatial resolution of the input were made in any steps of the methods. Since Guidos toolbox can be settled to work in the same resolution of the input that is provided, we guarantee the same accuracy for our fragmentation mappings of 1985, 2020 and secondary forests. The method used to map secondary forests based on MapBiomas was validated by comparing secondary forest areas with TerraClass secondary forest map using a bootstrap approach with 10 000, resulted in an R² = 0.6(0.03) and p < 0.001 (Silva-Junior et al 2020b).

The model used to estimate carbon loss due to edge effect was proposed by Silva-Junior et al (2020a)—the model was built based on light detection and ranging data, which provides information on the forest structure in high detail. The model in question corresponds to a Michaelis–Menten kinetic equation, which was applied to the average carbon loss stratified by the chronosequence of the age of the forest edges, having an R² = 0.780 and a mean squared error of 5.767 Mg C.

Continuous Fragmentation: Over the last 35 years, the AF biome witnessed the loss of crucial core (33.6%) and bridge (38.9%) areas (table 2(a)).

Contribution of Secondary Forests: Regeneration by secondary forests significantly contributed to all fragmentation classes from 1986 to 2020.

Parity in Carbon Emissions: Deforestation (818.41 TgCO₂) and edge effects (809.78 TgCO₂) almost equally contribute to carbon emissions, requiring the growth of approximately 1.96 million hectares of forest for offsetting.

Reach zero deforestation: Even with 90% stability in forest change, deforestation impacted the reduction of total forest, reflecting the loss of important fragmentation classes.

Core and bridge preservation: Essential classes to maintain the structural integrity of the existing forest. We Mapped 40 436.7 km² of total core forest, those, only 25 749.5 km² are within conservation units, leaving 36.3% unprotected. And might face the imminent risk of degradation or deforestation.

Islet preservation and management: It is the second largest class (35.4%). In Seasonally Dry formations, this class dominates, representing mainly what is left of them. The use of these areas in the implementation of restoration projects to enhance forest connectivity is urgent.

Promoting Sustainable Forest Management: Orientate secondary forests growth potential aiming on the replacement of degraded pastures with agroforestry systems highlighting non-forest matrix integration. Reducing turnover rates and ensuring long-term sustainability of regenerated forests.

These actions are central in sustaining AF's persistence and regeneration, aligning with Brazil's dedicated sustainable development strategies.