Crop fields are spread globally, resulting in agricultural landscapes dominating much of Earth’s surface (Frei et al., 2018; García-Vega et al., 2024; Landis, 2017; Sirami et al., 2019). Landscapes dominated by croplands have higher spatial-temporal dynamics compared to other types of landscapes (Driscoll et al., 2013; Santos et al., 2021) and are highly influenced by anthropogenic factors, such as commodity prices and farmer’s decision making, factors that directly and indirectly influence the composition and distribution of land covers (Vanbergen et al., 2020). Agricultural landscapes comprise a variety of crops, natural vegetation, and other land cover types (Philpott, 2013), which together deliver services for farmers and society (Bommarco et al., 2013), and currently represent the main interface between the environment and people (Frei et al., 2018). Hence, characterizing and understanding the biodiversity patterns in these landscapes is crucial because agroecosystems can have a strong influence on species occurrence (Driscoll et al., 2013; Estrada-Carmona et al., 2022), and landscapes dominated by agroecosystems is where most remaining natural vegetation areas are found (Driscoll et al., 2013; Duelli and Obrist, 2003; Frei et al., 2018; Lindenmayer, 2019). Furthermore, agroecosystems may benefit from higher biodiverse landscapes that can provide more ecosystem services (Estrada-Carmona et al., 2022).

Brazilian Cerrado is a biodiversity hotspot (Myers et al., 2000) and the most species-rich savanna in the world. However, it has been rapidly replaced by intensive agriculture and in the last decades became a central area of commodities expansion in Brazil, recording deforestation rates higher than the Atlantic Forest and Amazon (Chaves et al., 2023). Landscapes in Cerrado are now dominated by intensive agriculture, mostly soybean, corn, sugarcane, and pasture for beef cattle (Alencar et al., 2020), interspaced by small remnants of different natural vegetation types that range from grasslands to forests (Santos et al., 2021).

The expansion and unsustainable intensification of crop fields and consequently the simplification of agricultural landscapes have caused a decline in biodiversity (Bommarco et al., 2013; Estrada-Carmona et al., 2022; Landis, 2017). Notwithstanding, for a long time, ecologists have been trying to uncover whether habitat fragmentation per se (Fahrig et al., 2022) could also benefit existing wildlife or only lead to unprecedented species loss (Fahrig, 2003; Fahrig et al., 2019; Fletcher et al., 2018; Galán-Acedo and Fahrig, 2024; Valente et al., 2023). Furthermore, whether a Single Large (SL) is better than Several Small (SS) areas of natural vegetation (SLOSS) to preserve biodiversity in human-altered landscapes is still contentious, a debate known as the “SLOSS dilemma” (Fahrig, 2020; Fahrig et al., 2022).

In the last decades, a mix of paradoxical results have been provided (Fahrig, 2003; Fahrig et al., 2019; Fletcher et al., 2018; Galán-Acedo and Fahrig, 2024; Valente et al., 2023). Nevertheless, they largely show that small fragments in agricultural landscapes can support an adequate level of diversity, depending on the crop type, management system, and the level of compositional and configurational heterogeneity (Fahrig et al., 2011; Landis, 2017; Priyadarshana et al., 2024; Sirami et al., 2019). Recently, Gonçalves-Souza et al. (2025) compared the diversity in continuous and fragmented landscapes across the globe and stated that fragmented landscapes have lower α-alpha and γ-gamma diversity. However, this result was mainly based on data from tropical rain forest in Brazilian Atlantic Forest in landscapes with high habitat amount and low-intensive or family farming.

In landscapes dominated by agroecosystems, the small natural vegetation remnants have a significant role in maintaining wildlife, since they are the only available resources. Thus, landscapes dominated by SS can support different mechanisms and interactions, and the organisms tend to adapt and even use areas interspersed with habitats, which are not necessarily inhospitable matrices, as predicted by the theory of island geography (MacArthur and Wilson, 1967).

Here we contribute with the discussion raised in the literature about diversity patterns in SL and SS (Fahrig et al., 2019; Fahrig, 2020; Fahrig et al., 2022) testing the SLOSS prediction that a single large (SL) patch of natural vegetation has higher species richness than the sum of several small (SS) patches with total area equal to the SL, and that SS beta diversity is higher than SL. We estimate species diversity using a multitaxa dataset from a long-term ecological research (LTER) project, the LAND LTER (Agricultural Landscape Dynamics and Impacts on Biodiversity), in an intensive farming agriculture landscape in Brazilian Cerrado (Fig. 1a). The project landscape has a suitable design to test the SLOSS hypothesis, with several small remnants of natural vegetation (SS), comprising legal reserves in farms that should be preserved according to the Brazilian environmental law (Metzger et al., 2019), and two protected large areas: SL1, with 678.99 ha, and SL2, with 1744.802 ha (Fig. 1b,c).

Fig. 1.

The LAND LTER landscape. a. The Brazilian Cerrado (in yellow) in the Central-West Brazil, and the LAND LTER site location (red square). b. sites sampled in 2018. c. sites sampled in 2022. The black dots are the sampling sites. SL1, is the single large area with 678.99 ha, and SL2 is the single large area with 1744.802 ha. Habitat and matrix mapping is based on the MapBiomas land cover map database available at https://brasil.mapbiomas.org/en/ with revision and field checking of natural vegetation land cover.

We used species data from different taxa sampled in 2018 and 2022: vertebrates (birds, small non-flying mammals, and medium and large-sized mammals), Euglossini bees, and savanna and forest woody plants. Our final dataset comprised 596 species, 207 birds, 17 bees, 17 small non-flying mammals, 31 medium and large-sized mammals, 136 forest plants, and 188 savanna plants (Appendix A Tables S1 and S2).

We found significantly higher species richness in SS than SL for most taxa analysed (Figs. 3–5). For forest plants, we found significantly higher richness in SS than SL1 in 2018 (SS richness = 77.97 vs. SL1 richness = 41.02; Fig. 3A) and in 2022 (SS richness = 73.32 vs. SL1 richness = 14.29; Fig. 3B). In 2022, we found no significant difference in forest plant richness between SS and SL2 (Fig. 3C). For savannas, we found higher plant richness in SL1 than SS (Fig. 3D) in 2018. Otherwise, we found significantly higher plant richness in SS than SL1 in 2022 (SS richness = 77.89 vs. SL1 richness = 28.12; Fig. 3E).

Fig. 3.

Comparison of γ diversity between a SL (SL1) and SS patches of natural vegetation across taxa in 2018. a. birds, b. forests, c. savannas, d. Euglossini bees. * Significant difference between SL and SS.

Fig. 4.

Difference in γ diversity between SL (SL1 and SL2) and SS natural vegetation areas across taxa in 2022. A and B. birds, C. and D. small non-flying mammals, E. and F. forests, G. savannas, H. Euglossini bees. * Significant difference between SL and SS.

Fig. 5.

Difference in γ diversity between SL and SS patches of natural vegetation for medium and large-sized mammals in 2022. A. Estimated richness in SL1 and SS B. Estimated richness in SL2 and SS. * Significant difference between SL and SS.

For birds, we found no significant difference in richness between SL1 and SS in 2018 (Fig. 4A) and 2022 (Fig. 4B), and between SL2 and SS in 2022 (Fig. 4C). For bees, we found significantly higher richness in SS than SL1 in 2018 (SS richness = 13.18 vs. SL1 richness = 8.96; Fig. 4D). For 2022, we found no significant difference in richness between SS and in SL1 (Fig. 4E).

We recorded small, medium, and large-sized mammals’ only in 2022. For small mammals, we found significantly higher richness in SS than SL1 (SS richness = 36.41 vs. SL1 richness = 12.45; Fig. 5A) and SL2 (SS richness = 44.99 vs. SL1 richness = 8.07; Fig. 5B). For medium, and large-sized mammals, we found significant higher richness in SL1 (Fig. 5C) and SL2 (Fig. 5D) than in SS.

Overall, we found no significant relationship between beta diversity and its components and spatial distance for most taxa (Fig. S4). In 2018, forest woody plants’ beta diversity (regression β coefficient = 0.011, p = 0.016) and nestedness (regression β coefficient = 0.008, p = 0.009) were positively related to spatial distance (Fig S4a). Although these results indicate a clump species distribution and species loss with distance, the regression beta coefficients and the coefficients of determination were very low for both beta diversity and nestedness (Table S3). In addition, Mantel correlogram shows that the relationship with spatial distance is significant only at shorter distances (Fig. S5), which may lead to low values of regression β coefficient for global regression.

In 2022 (Fig. S4b, Table S4), we found a positive relationship between SS beta diversity and spatial distance for birds (regression β coefficient = 0.0008; p = 0.038). For bees, nestedness (regression β coefficient = 0.007; p = 0.001) had a positive relationship with spatial distance, while turnover had a negative relationship (regression β coefficient = −0.004; p = 0.005). However, again, coefficients were very low, meaning that the explained variation was very low.

Our results provide strong evidence that SLOSS has no single solution. Our results show that in an intensive agricultural landscape, most taxa have higher species richness in SS than SL. Our findings are consistent with a set of studies that have highlighted the importance of SS to maintain biodiversity (Fahrig, 2003; Fahrig et al., 2019; Galán-Acedo and Fahrig, 2024; Lindenmayer, 2019; Riva et al., 2024; Valente et al., 2023; Wintle et al., 2019; Huber et al., 2025), and contrast to Gonçalves-Souza et al. (2025) that emphasised the higher importance of continuous over fragmented habitats for biodiversity maintenance. Nevertheless, the effect of habitat fragmentation on biodiversity is modulated by the total amount of habitat available at the landscape scale (Zhang et al., 2024). Although the concept that large habitat patches are more important for biodiversity is deep-rooted in conservation (Riva and Fahrig, 2023), our results suggest that several small patches of natural vegetation (SS) can have a higher importance in maintaining biodiversity in landscapes surrounded by low amount of habitat, as the LAND LTER landscape.

Differences in environmental conditions and vegetation structure in different biomes can lead to different importance of SL and SS remnants for biodiversity (Mucina, 2019; Olson et al., 2001). For instance, savannas are highly different from tropical rainforests, being naturally fragmented due to the high natural environmental heterogeneity (Sano et al., 2019), and harbouring more open vegetation with lower canopy and lower tree density (Silva de Miranda et al., 2018). We assessed an intensive agricultural landscape in Cerrado biome; therefore, these differences may lead to different community responses to fragmentation and habitat loss where the species are adapted to a naturally more open fragmented vegetation (de Mattos et al., 2021; Silveira et al., 2023). Especially, in this type of landscape, other mechanisms such as increasing habitat diversity, movement success, landscape complementation, matrix composition, and decreasing competition can explain a positive effect of habitat fragmentation on richness (Driscoll et al., 2013; Galán-Acedo and Fahrig, 2024; Lindenmayer, 2019; Valente et al., 2023).

Our results underscore that the contribution of SL or SS to maintain species richness cannot be generalized (Fahrig et al., 2019; Lindenmayer, 2019; Valente et al., 2023), since it depends on matrix composition, the amount of habitat at the landscape scale, regional particularities, spatial scale, taxa, and study design (Fahrig et al., 2019; Lindenmayer, 2019). For instance, our study design allowed analysing the spatiotemporal variation in species richness, showing that differences between SL and SS may vary not only across space but also over time. We found a small but significant difference in bee richness between SS and SL in 2018. However, we cannot confirm this difference in 2022. Savanna plants had a higher richness in SL1 in 2018, and an opposite pattern in 2022, but this is most likely due to the spatial extension of LAND LTER in 2022, including different SS sites (Fig. 2). In 2022, the LAND LTER study area was expanded, including new habitats of savanna and forest, specifically a new larger protected area of forest (SL2), which can also explain the significant and highest forests’ woody plant richness in SS compared to SL1 (Fig. 3). In addition, the higher savanna plant species richness in SL1 in 2018 compared to 2022 is most likely due to the highest number of sampling sites (11 in 2018 and one in 2022). Also, it reinforces the value of SS to conserve biodiversity, since we found no difference in forest richness between SL2 and SS in 2022 (Fig. 3). For birds, we found no significant difference in diversity across the years between SS and SL, independently of the single large protected area analysed (Fig. 4). Most bird species sampled in LAND LTER landscape are habitat generalists, meaning that they can use both habitat savanna and forests, and the matrix. SS can have similar contribution compared to SL to birds’ diversity due to the lower response to microhabitats and higher dispersal capacity (Huber et al., 2025).

For small mammals, we found a significantly highest richness in SS than in SL1 and SL2 in 2022. Other studies have already reported a positive effect of habitat fragmentation on small mammal diversity (Palmeirim et al., 2019; Rocha et al., 2018) as we found here. Moreover, for medium and large-sized mammals (Fig. 5A and B), we found smaller richness in SS compared to both SL1 and SL2 in 2022, agreeing to other studies (Rios et al., 2022, 2021; Rocha et al., 2025). For this group, our results highlight the importance of large natural vegetation remnants for maintaining large-sized mammal communities in human-altered landscapes. However, it is important to note that SS and SL have the same importance in keeping primates’ diversity across different regions of the globe (Beard et al., 2025).

Overall, our findings highlight that SLOSS has no unique answer. Here, in an experimental area designed to test SLOSS predictions, we showed that for two SL with different sizes, SS had higher species richness than SL for most taxa. Thus, our results also reveal the importance of SS to the cumulative species richness across human-made landscapes (Riva and Fahrig, 2023). Moreover, we show that for the same taxon, SLOSS can have different solutions depending on the sampling design. Thereby, the lack of temporal, spatial, and taxonomic pattern emphasizes the need of recording wildlife in the same area over different years to avoid stochastic differences and hasty outcomes interpretation.

We also found no relationship of beta diversity with spatial distance between SS for most taxa. In fact, the high heterogeneity of the Cerrado biome is acknowledged to drive the high beta diversity in several taxa such as plants (Bridgewater et al., 2004), small non-flying mammals (Ribeiro et al., 2020), birds (Jesus et al., 2017), medium and large-sized mammals (Bogoni et al., 2025), and bees (Aguiar et al., 2018). Therefore, if micro-habitat suitable conditions are not clumped, fragmentation may not lead to SS > SL (Fahrig, 2020). Deforestation can be spatially autocorrelated in intensive agriculture landscapes, because flat areas with deeper soils are usually used for intensive agriculture due to mechanization, letting the hills with higher amount of habitat, leading to SS > SL (Fahrig, 2020). For instance, in the Atlantic Forest, most remaining habitats are in hilly regions (Bicudo da Silva et al., 2020) that are unsuitable for intensive farming. However, in LAND experimental area, hilly areas are usually occupied by pasture or other crops, which are adapted to this relief type or require little mechanization.

Taken together, our findings point that, particularly in Cerrado, overlooking the role of SS towards SL can be especially critical, because it can give support for the agribusiness sector’s demand to reduce or remove the formal protection of natural vegetation (riparian forests and legal reserves) in private lands, since these habitats will not offset biodiversity loss. Most of the small areas of natural vegetation in the Cerrado are private legal reserves (De Marco et al., 2023), in which the Brazilian government has little governance power. Legal reserves are the most abundant natural vegetation areas in the Brazilian territory, with huge importance for biodiversity (Metzger et al., 2019). We believe that upholding that SL is more important than SS in a world where swath of human-made structure dominates over natural vegetation (Elhacham et al., 2020) is a menace to biodiversity conservation (Riva and Fahrig, 2023). Additional specific policies and a reinforcement of environmental laws are necessary for safeguarding both large and small natural vegetation areas in the Anthropocene landscapes.

Undoubtedly, SL are important everywhere. However, SS remnants of natural vegetation are spread globally, and it is important to recognize their value for biodiversity conservation, especially in intensive agricultural landscapes. More importantly, our results point to the importance of intensive farming landscapes for the SLOSS debate.

Together, SL and SS play a leading role in different landscape types in different regions of the world. In Brazil, for instance, the agribusiness lobby has been trying to pass bills to easy environmental laws about protection of natural vegetation in farms (legal reserves) and environmental licenses for deforestation. Therefore, showing the importance of keeping SS will be essential in landscapes dominated by agroecosystems, since they are the only natural vegetation remnants in most of these environments, and can safeguard species richness as we show here.

RGC and JSS conceptualized the work. RGC funded and supervised the work. RGC and JSS wrote the manuscript with inputs from ELL and PS. ELL and PS analysed the data. BEL, KBA, WT, FGS, ALR, and CMSN collected the data. BEL and CMSN identified plant species. ALT identified bird species. FGS identified Euglossini bees. KBA identified small non-flying mammals. ALR identified medium and large-sized mammals. RGC, JSS and ELL curated the data. All authors read and approved the final version of the manuscript.

The datasets used in this paper and results are provided in supplementary material Appendix A, and the codes used in this paper are included in Appendix B.