The study area encompasses the entire Atlantic Forest biome (Federal Decree 750/1993; Law 11.428/2006), covering 148 million ha across 3,429 municipalities in 17 Brazilian states (Supplementary Fig. S1). Extending from 5 ° N to 33 °S and 35 ° W to 52 ° W (IBGE, 1992), it spans wide climatic and topographic gradients, with altitudes from 0–2,200 m, rainfall from 800–4,000 mm, and mean temperatures between 15–25 °C (IBGE, 1992; Câmara, 2003). This diversity supports several forest types, including Dense, Mixed, Semideciduous, and Deciduous Forests (Marques and Grelle, 2021).

Historically, the biome suffered intense deforestation driven by timber extraction, sugarcane, mining, coffee, and later urbanization and livestock expansion (Solórzano et al., 2021). Today, only about 22% of forest vegetation remains, mostly as small, fragmented patches within agricultural and urban matrices (Rezende et al., 2018; Vancine et al., 2024).

Municipalities were used as analytical units since they are the smallest governance entities in Brazil with available ecological and social data (SOS Mata Atlântica, 2013). Of the total, 390 municipalities were excluded due to limited area or partial inclusion (<50%) within the biome. Land use and cover limits followed IBGE (2010, see Supplementary Fig. S2) to assess the distribution limits of the Atlantic Forest.

We searched for databases that could represent provision, regulation, and support services (according to MEA, 2005) in all municipalities of the Atlantic Forest. We used as criteria in the choice of data: (1) the quality of the data source (whenever possible, official data), (2) the existence of data available at the municipal level, (3) the possibility of spatializing the data for use at the municipal level. The ecosystem services and the sources were defined as follows (Supplementary Table S1):

Provisioning services: We included the annual production of coffee, sugarcane, soybean, corn, and cattle, following the Millennium Ecosystem Assessment (2005) classification. Agricultural and livestock outputs were considered tangible provisioning ecosystem services sustained by regulating and supporting processes such as soil quality, water availability, and nutrient cycling (Burgos-Ayala et al., 2024; Meister and Marques, 2025). That inclusion allows us to assess the spatial interactions, trade-offs, and synergies between production-oriented land uses and the ecological functions that sustain them. Production data (2013) were obtained from the Brazilian Agriculture Monitoring System (SOMABRASIL) database (http://mapas.cnpm.embrapa.br/somabrasil/webgis.html; Batistella et al., 2012).

Regulating services: We used data on water balance (Wbalance), carbon stock, and soil productive capacity (soil). Water balance was calculated as precipitation minus real evapotranspiration for each municipality, using WorldClim precipitation data and Consortion for Spatial Information (CGIAR) evapotranspiration data (Hijmans et al., 2005; Trabucco and Zomer, 2010), averaged for 1950–2000. Carbon stock (2000) was obtained from Ruesch and Gibbs (2008). Soil productive capacity was estimated as the mean of three soil variables: soil water content (Trabucco and Zomer, 2010), soil organic carbon, and cation exchange capacity (Hengl et al., 2014).

Supporting services: We used the distribution of native woody species as proxies for biodiversity across the Atlantic Forest. Occurrence records from checklists (Lima et al., 2015; Zwiener et al., 2020) and speciesLink were cleaned and filtered to remove imprecise locations and reduce spatial bias, retaining 2,255 species with ≥15 records. Ecological Niche Models (ENMs) were fitted in Maxent (v.3.3.3) using the first six principal components (PCA) from 19 WorldClim bioclimatic variables (>95% of climatic variation). Suitability outputs were thresholded to obtain binary predictions and restricted to species’ accessible areas, then stacked to estimate species richness at the municipality level. Detailed procedures are provided in Supplementary Material.

All data on service magnitude at the municipal level were square-root transformed to achieve normality. The values were normalized by area to consider the municipality's size (Mouchet et al., 2014). This normalization was performed by dividing the total of services by the municipality's total area, except for services related to soil properties. The services coffee, sugarcane, soybean, corn, and cattle production were also weighted by the land use area of the municipality (Supplementary Fig. S2). All data were standardized to be comparable among services by attributing value 1 for the municipality with the highest value, and the other values were calculated relative to this maximum (Raudsepp-Hearne et al., 2010).

All the ecosystem service data were mapped and processed in ArcGIS 10.1 (Esri, 2010). The data were then exported to a matrix, with the rows representing the municipalities and the columns representing the ES for multivariate analysis. The data are available in the Zenodo repository (DOI: 10.5281/zenodo.17611824).

Municipalities were clustered to identify ecosystem service bundles in the Atlantic Forest based on the distribution of ecosystem services (Raudsepp-Hearne et al., 2010; Turner et al., 2014). We used a K-means cluster analysis to identify similar sets of ecosystem service bundles using Spatial Analysis in Macroecology (SAM) 4.0 (Rangel et al., 2010). The appropriate number of clusters was qualitatively determined by the composition of services within each cluster (Turner et al., 2014). Each cluster was mapped to visualize the spatial pattern in ArcGIS 10.1 (Esri, 2010) and starplots were constructed in R statistical software (R Core Team, 2013). The robustness of the K-means classification was evaluated through spatial subsampling and comparison with alternative clustering methods (see Supplementary Material, Table S2)

For each ecosystem service bundle, we assessed spatial trade-offs and synergies, as well as interactions between pairs of ecosystem services. We used Pearson’s pairwise correlation and to account for spatial autocorrelation, adjusted significance levels by modifying the degrees of freedom using Dutilleul’s method (Dutilleul et al., 1993). Positive relations were defined as synergies and negative as trade-offs from the correlations. To display such synergies and trade-offs, we used chord diagrams. Chord diagrams allow us to visualize numeric tables and study directional relations among a set of entities (i.e., ecosystem services) in a circular manner (Gu et al., 2014). Ecosystem services (nodes) were displayed around a circle, and their correlations were represented by arcs (links). We used the R package circlize to generate chord diagrams (Gu et al., 2014).

Ecosystem services categories are spatially structured in the Atlantic Forest (Fig. 1). Provisioning services (the average of coffee, sugarcane, soybean, corn, and cattle production) are distributed mainly in the continental region of the Atlantic Forest (Fig. 1a), on the limits of the biome with the Brazilian Cerrado. Regulation services (Wbalance, carbon stock and soil_capacity) are distributed mainly in the coastal region of the northern and southern portions of the Atlantic Forest and inner areas of south Brazil (Fig. 1b), coinciding with the more significant remnants of the biome (Supplementary Fig. S1). Supporting service (woody plant species diversity) is also distributed in the coastal region (Fig. 1c), in a more extensive and connected region, compared to regulation services, incorporating large and medium remnants of the Atlantic Forest (Supplementary Fig. S1).

Fig. 1.

Distribution of three ecosystem service categories across the study area: (a) provisioning; (b) regulating; (c) supporting. The gradient in the heat maps represents municipality averages for each ecosystem service category. Provisioning (n = 5 ecosystem services: coffee, sugarcane, soybean, corn and cattle production); Regulating (n = 3: Wbalance, carbon stock, and soil); Supporting (n = 1: (woody plant species biodiversity).

We found three main groups of bundles of ecosystem services across the 3,039 municipalities of the Atlantic Forest from the cluster analysis of ecosystem services (Fig. 2). The Continental mosaic cropland - sugarcane bundle (Fig. 2), comprises 761 municipalities, and is characterized by high levels of one type of provisioning services (sugarcane) and one type of regulating services (soil capacity). The municipalities in this group were mainly located in the continental portion of the Atlantic Forest, along its borders with the Cerrado biome. The Mosaic cropland – soybean and coffee bundle (Fig. 2) comprises most municipalities (1,282), and is characterized by intensive agriculture, which was reflected by the high provision of provisioning services (soybean, coffee, and corn) and low levels of regulating services. The Coastal mosaic forest bundle (Fig. 2) comprises 996 municipalities, mainly in the Atlantic Forest coastal region. These areas corresponded to the region with more Atlantic Forest remnants, which was reflected by the high provision of high values for many services, except for coffee and sugarcane.

Fig. 2.

Distribution of the three ecosystem service bundles across the Brazilian Atlantic Forest. Starplots show the normalized (0-1) average values of ecosystem services for each bundle.

The Continental mosaic cropland—sugarcane bundle is characterized by synergies among provisioning services and trade-offs between regulating services, provisioning services, and supporting services (Figs. 3 and 4a; 4a). For example, provision services such as sugarcane and corn (r = 0.48) and soybean and corn (r = 0.34) were highly and positively correlated. On the other hand, biodiversity (supporting) with carbon stock and water balance (regulating) were the most negatively correlated (r = −0.39).

Fig. 4.

Pairwise Pearson's correlations (r) among ecosystem services in each bundle are categorized into positive, negative, and no correlation patterns. Red and blue indicate negative and positive correlations, respectively. (a) Continental mosaic cropland - sugarcane bundle; (b) Mosaic cropland– soybean and coffee bundle; (c) Coastal mosaic forest bundle (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article).

Fig. 3.

Chord diagram of relationships among ecosystem services in each bundle. (a) Continental mosaic cropland - sugarcane bundle; (b) Mosaic cropland – soybean and coffee bundle; (c) Coastal mosaic forest bundle. The width of each chord represents the number of synergy (positive) or tradeoff (negative).

The Mosaic cropland – soybean and coffee bundle is dominated by trade-offs among provisioning services and regulating services (Fig. 3b; 4b). For example, coffee and carbon (r = −0.26) and sugarcane and soil (r = −0.19) were negatively correlated. Biodiversity (supporting) was negatively correlated with all ecosystem services, and the most negative pair was with soil (r = −0.58). Synergies were predominantly among provisioning services (Fig. 3b). For example, provision services such as soybean and corn (r = 0.54), sugarcane and corn, and soy (r = 0.33).

The Coastal mosaic forest bundle is characterized by trade-offs in provisioning and regulating services (Fig. 3c; 4c). For example, carbon and all provisioning services were negatively correlated. Biodiversity (supporting) was negatively correlated with soil (r = −0.63) and water balance (r = −0.33). Synergies were predominant among provisioning services (Fig. 3c; 4c). For example, provision services such as soybean with cane (r = 0.68), coffee (r = 0.37), and corn (r = 0.36).

Socioeconomic, ecological, and historical patterns are key drivers of the distribution of ecosystem services (ES) in the Atlantic Forest. Provisioning services (sugarcane, soybean, corn, and cattle) were more prominent in the western region (São Paulo and Paraná), where favorable soils, climate, and infrastructure support agricultural development (Buainain et al., 2014). This region experienced severe deforestation (Dean, 1996), with remaining forests largely restricted to inaccessible areas such as Serra do Mar, spanning SP, PR, and SC (Vancine et al., 2024). These remnants sustain high levels of regulating (carbon, soil, and Wbalance) and supporting (biodiversity) services. Protected areas play a central role, concentrating 66% of carbon supply, 60% of soil quality provision, and key biodiversity refuges (Pires et al., 2021; Zwiener et al., 2017).

Agricultural modernization, through mechanization and fertilizer use, expanded production into less fertile soils (Buainain et al., 2014). This process likely influenced the spatial patterns of soil productivity and regulating services, which were more prominent in the southern region, in contrast to crop production that was concentrated in the western region, characterized by poorer soils. Similar topographic influences on agricultural suitability have been observed in other regions, such as Beijing and Changzhi, China (Chen et al., 2021; Yuan et al., 2023).

Understanding spatial interactions among ecosystem service bundles is critical for achieving sustainable development goals and ecosystem-based management (Deng et al., 2016). The bundles Continental mosaic cropland - sugarcane and Mosaic cropland – soybean and coffee were dominated by trade-offs with provisioning and regulating services. These regions combine favorable geographic and climatic conditions with extensive infrastructure that supports agricultural expansion (Buainain et al., 2014; Kong et al., 2018). However, prioritizing provisioning services has reduced key regulating services, such as water balance, leading to economic and social losses across the biome (Nobre et al., 2016).

In areas where land-use competition threatens biodiversity, expanding and strengthening protected areas—especially by integrating climate change into planning—can help conserve forest fragments and limit agricultural expansion (Medeiros and Araujo, 2011; Malecha et al., 2023). Regulating services such as water balance, carbon stock, and soil productivity are largely concentrated in protected areas (Pires et al., 2021), highlighting their crucial role in maintaining ecosystem services.

The Coastal mosaic forest bundle showed strong synergies among regulating services (carbon stock, Wbalance, soil_capacity), with biodiversity (supporting services) positively correlated to carbon stock. These synergies coincide with regions containing the largest Atlantic Forest remnants, mainly along the Serra do Mar in Paraná, São Paulo, and Rio de Janeiro, and in parts of Bahia. The Serra do Mar retains 36.5% of its original vegetation (1,109,546 ha), while Bahia preserves about 17.7% (Ribeiro et al., 2009; Vancine et al., 2024).

Mechanisms underlying synergies among regulating services rely on a high proportion of vegetation cover in forested areas. In this respect, ecological restoration projects can simultaneously promote these regulatory ES in the Atlantic Forest. Beyond reinforcing the known role of vegetation cover, the identification of bundles provides a practical framework for prioritizing restoration and management actions. For instance, areas with strong trade-offs could be great potential for agroforestry or incentives such as Payment for Ecosystem Services (PES, a mechanism for transferring resources between social actors), whereas regions showing synergies could serve as core areas for biodiversity conservation and natural regeneration (Ruggiero et al., 2019; Pires et al., 2021; Urruth et al., 2022). Considering that the United Nations declared 2021–2030 the Decade on Ecosystem Restoration (UN, 2024), understanding bundle dynamics can guide more balanced interventions that reconcile production needs and ecological integrity.

The Atlantic Forest, home to over half of Brazil’s population and its main economic centers (IBGE, 2023), faces intensifying trade-offs between provisioning and regulating services due to agricultural and urban expansion. In 2023 alone, 12,094 ha of native vegetation were lost (Mapbiomas, 2024). Our findings provide actionable insights for designing differentiated management strategies that secure multiple ES.

ES bundles dominated by provisioning services indicate where restoration and incentive policies are most needed. Instruments such as PES, agroforestry systems, and ecological restoration programs can enhance multifunctionality and reduce trade-offs, particularly in private lands that host nearly 90% of remaining Atlantic Forest (Ruggiero et al., 2019; Pires et al., 2021; Resende et al., 2024). In contrast, bundles dominated by regulating and supporting services — especially the Coastal mosaic forest — represent priority areas for conservation and carbon storage. These zones could anchor connectivity initiatives and biodiversity corridors under climate change scenarios that demand more robust and representative protected area networks (Malecha et al., 2023). Strengthening these protected areas through initiatives such as National Policy for the Restoration of Native Vegetation (Planaveg) and a collective initiative Pact for the Restoration of the Atlantic Forest would enhance synergies among ES and provide co-benefits for local communities (Pinto et al., 2023).

Integrating the spatial structure of ES bundles with Brazil’s environmental legislation — including the Atlantic Forest Law (Federal Law 11.428/2006), and Native Vegetation Protection Law (Federal Law 12.651/2012) — opens new opportunities for targeted conservation. Regions dominated by regulating services could be prioritized for environmental compensation mechanisms, while agricultural regions could integrate restoration targets into rural property management (Pinto et al., 2023; Resende et al., 2024). Translating these frameworks into practice through local governance and economic incentives remains a central challenge.

Overall, the spatially explicit analysis of ES bundles offers a valuable tool for integrating ecological and socio-economic dimensions in decision-making. It supports balanced landscape planning that reconciles production, restoration, and conservation goals, in line with Brazil’s zero-deforestation commitment and the global ecosystem restoration agenda.

We acknowledge that the datasets used in this study span the period from 2000 to 2017. Although the datasets are not fully recent, they were carefully selected to ensure spatial completeness, methodological consistency, and comparability across all municipalities of the Atlantic Forest biome. This selection allowed the development of a coherent framework for integrating multiple ecosystem service and biodiversity indicators across a large and socioecologically diverse region.

Recent studies show that the spatial configuration of land-use and associated ecosystem services in the Atlantic Forest has remained relatively stable over the past two decades. Vancine et al. (2024) report only minor net changes in native vegetation between 1986 and 2020 (a loss of 2.4–3.6%), with evidence of stabilization and partial recovery after 2005 due to environmental policies and natural regeneration. Similarly, Meister and Marques (2025) indicate that in the southern part of the Atlantic Forest, major trade-offs and synergies among ecosystem services—particularly between provisioning and regulating services —are strongly associated with persistent land-use patterns.

Therefore, despite the period of datasets, they provide a robust and reliable basis for analyzing large-scale spatial patterns and interactions among ecosystem services, and the results remain fully relevant for current decision-making. Nonetheless, future studies could examine temporal dynamics and further refine ecosystem management strategies.