Search for a command to run...
In most human-dominated regions, remaining forest exists in thousands of small forest patches (Riva, Koper, et al., 2024). Such small patches traditionally received lower priority for conservation than large continuous areas (Fahrig et al., 2022). Support for this goes back at least to Diamond's (1975) reserve design principles, the first of which states that protection of a large reserve should be prioritized over protection of a small reserve. This view is widely accepted because of the well-established link between forest loss and biodiversity decline (Riva, Haddad, et al., 2024). However, the second principle—protection of a single large reserve should be prioritized over protection of many small reserves of the same total area as the large reserve—has been widely debated because of a lack of empirical support (Fahrig, 2020; Quinn & Harrison, 1988; Simberloff & Abele, 1982). The main arguments for prioritizing large forest patches are that forest specialist species need the interior forest in large patches and experience high mortality, low reproductive success, or both near forest edges (e.g., Betts et al., 2019; Cabrera-Guzmán & Reynoso, 2012; Hending et al., 2024; Valdivia et al., 2011). However, multilandscape studies show that distributing a given total forest area across a larger number of smaller patches (i.e., increasing fragmentation per se [Figure 1]) usually only weakly affects biodiversity, and when such effects are significant, they are usually positive, not negative (reviewed in Fahrig [2003, 2017]). Possible explanations include higher habitat heterogeneity, between-patch movement, spreading of risk, and stability of species interactions across many smaller patches than fewer larger ones (Fahrig, 2020). Eschewing these findings, researchers and practitioners continue to recommend prioritizing protection of large patches, based mainly on minimum patch size requirements and negative edge effects. The need for a consensus on the effects of fragmentation per se is increasingly recognized (Miller-Rushing et al., 2019; Riva, Koper, et al., 2024; Valente et al., 2023) because deforestation continues and small forest patches are particularly vulnerable to destruction (Riva et al., 2022). If fragmentation per se reduces biodiversity, then policies should prioritize protection and restoration of large patches. If not, then policies should include all forest, irrespective of patch sizes (Riva & Fahrig, 2023). This would allow effective biodiversity conservation, even in human-dominated regions where no large patches remain, by protecting and restoring sufficient forest over a network of many small patches (Arroyo-Rodríguez et al., 2020). One study type often cited as support for negative fragmentation effects is the so-called “fragmentation experiment” (e.g., Laurence et al., 2011; Margules, 1992). In such experiments, forest is removed from one portion of the study area, the fragmented landscape, to create patches of different sizes. Sample plots are distributed in the patches and in a continuously forested portion of the study area, the continuous landscape. Diversity is then compared between the fragmented landscape and the continuous landscape, and among different-sized patches. Equivalently, diversity can be compared in existing forest patches versus a nearby continuous forest. Such binary continuous versus fragmented landscape studies compare diversity between only 2 landscapes (A and D in Figure 1). Because these 2 landscapes differ in their fragmentation level and the total amount of forest they contain, it is impossible to determine whether lower diversity in the fragmented than the continuous landscape is caused by forest loss, fragmentation, or their combination. In contrast, the fragmentation debate is about whether a single or a few large patches have higher or lower biodiversity than many small patches that sum to the same total area (A vs. B or C vs. D in Figure 1). Estimating effects of fragmentation per se requires sampling biodiversity in multiple landscapes across gradients in forest amount and fragmentation that are sufficiently independent to allow estimation of their separate effects. That continuous versus fragmented landscape studies cannot control for forest amount while evaluating fragmentation effects means they are not relevant to the fragmentation debate. This is underrecognized. For example, in their review of fragmentation experiments, Haddad et al. (2015) state that the studies were “designed to manipulate specific components of fragmentation … while controlling for confounding factors such as the amount of habitat lost across a landscape” (emphasis added), although this was generally not the case. Given this, we were intrigued to read Gonçalves-Souza et al. (2025). They collated data from 37 previous studies, each approximating the continuous versus fragmented landscape design (figure 1a in Gonçalves-Souza et al. [2025]). Gonçalves-Souza et al. placed their study in the context of the fragmentation debate and claim to control for forest amount while evaluating fragmentation effects. They state, “testing the so-called fragmentation per se effect by methodologically partitioning fragmentation independent of habitat amount is challenging in paired designs … To tackle this problem, we used different analytical approaches that account for habitat amount” (emphasis added) (Gonçalves-Souza et al., 2025: Supplementary Information). The studies from which Gonçalves-Souza et al. took their data were not designed to evaluate the effects of fragmentation per se. Each study had only 2 landscapes, one with high forest amount and low fragmentation and the other with low forest amount and high fragmentation; thus, they completely confounded forest amount and fragmentation (Figure 2). Gonçalves-Souza et al. attempted to resolve this by combining multiple studies, but they did not appropriately control for the effects of forest amount. As such, we consider Gonçalves-Souza et al. a good illustration of why appropriately controlling for forest amount is critical when estimating effects of fragmentation per se. We replicated the portion of their analyses for which they provided their code and reanalyzed their data with the study design shown in Figure 1 (details below). In stark contrast to their conclusions, but consistent with most empirical evidence (Fahrig, 2017), we found no effects of fragmentation per se. We conclude that properly controlling for the effect of forest amount is critical in studies of fragmentation per se, particularly given the conservation context discussed above. For each of the 37 studies, Gonçalves-Souza et al. calculated mean diversity metrics across pairs of sample plots in the continuous landscape and across pairs of sample plots in the fragmented landscape (Figure 2). They considered landscape type (continuous or fragmented) as their fragmentation variable. They did not measure forest amount or fragmentation in the categorical continuous and fragmented landscape types. Instead, they measured mean forest amount in smaller landscapes surrounding the sample plots within the much larger continuous and fragmented landscapes (pink circles in Figure 2), which they included in their models. They considered various small landscape sizes, but presented results only for 2-km-radius landscapes. They then inferred that the effects of landscape type (continuous or fragmented) on diversity were evidence for effects of fragmentation controlled for forest amount (i.e., fragmentation per se). Gonçalves-Souza et al. treated landscape type as a fixed factor and study and mean forest amount as random factors, with a random slope and intercept for forest amount: glmmTMB [diversity ∼ landscape type + (forest amount | study)] (Gonçalves-Souza et al. Model 2). Despite their claims, they did not assess effects of fragmentation per se with their model because it did not control for the difference in forest amount between the 2 large landscapes (continuous vs. fragmented) in each of the 37 studies (Figure 2). Mean forest amount in the small 2-km-radius landscapes (pink circles in Figure 2) is not equivalent to forest amount in the large continuous and fragmented landscapes. Perspicaciously, Gonçalves-Souza et al. included with their data the median number of patches across the 2-km-radius landscapes within each large continuous or fragmented landscape, although they did not use this variable. This allowed us to reanalyze their data, abandoning the categorical landscape type variable (continuous vs. fragmented). Instead, we directly estimated the effects of fragmentation per se on diversity by modeling effects of mean forest amount and median number of patches (fragmentation), both in the 2-km-radius landscapes. Number of patches is a common and appropriate measure of fragmentation because an increase in fragmentation implies more, smaller patches. Our model was glmmTMB [diversity ∼ number of patches + forest amount + (1 | study)]. We replaced landscape type (continuous or fragmented) in Gonçalves-Souza et al.’s analyses with fragmentation measured at the same scale as their forest amount measure. Conceptually, this shifts the study design from that in Figure 2 to that in Figure 1, where the landscapes of Figure 1 are the 2-km-radius landscapes around the sample plots (pink circles in Figure 2). We also replicated their analyses by using the code they provided, in order to allow direct comparison with our results (Figure 3). We analyzed the same responses as Gonçalves-Souza et al. (alpha, beta, and gamma diversity), weighted species rarity in 3 ways, and used their 2 methods for sample plot selection (all possible pairs of plots and pairs of nearest-neighbor plots only). We included forest amount as a fixed effect rather than a random effect so that we could directly estimate and control for its effects when assessing the effects of forest fragmentation (Smith et al., 2009). This also avoided model convergence problems. As in Gonçalves-Souza et al., we log-transformed response variables if model assumptions were not satisfied. Our code is available from https://osf.io/hu3xw/?view_only=7c6fe118f8514e03a1b37eb8bd906dd8. Our replication of Gonçalves-Souza et al.’s analyses produced lower alpha and gamma diversity and higher beta diversity in the fragmented landscapes than in the continuous landscapes (Figure 3a), as reported in their study. In contrast, directly measuring both mean forest amount and median number of patches (fragmentation) in the smaller 2-km-radius landscapes, we found no evidence for effects of fragmentation per se on any diversity response (Figure 3b) and positive effects of forest amount on several alpha and gamma diversity responses (not shown). Given this, we surmise that the cause of the categorical landscape type effect in Gonçalves-Souza et al. is the difference in forest amount between the continuous and fragmented landscapes. We could not directly test this because Gonçalves-Souza et al. did not measure forest amount or fragmentation at the larger scale of the continuous and fragmented landscapes. However, our interpretation is consistent with previous evidence showing strong forest amount effects and generally weak fragmentation effects (reviewed in Fahrig [2017]). We do not suggest that binary continuous versus fragmented landscape studies are irrelevant to biodiversity conservation. Hundreds of articles from such studies have provided information about patch size effects, edge effects, matrix effects, patch isolation effects, and forest loss effects (summarized in Haddad et al. [2015]). However, such studies cannot provide estimates of the effects of fragmentation per se because they completely confound fragmentation effects with forest amount effects. Gonçalves-Souza et al. attempted to resolve this by combining multiple such studies, but they did not appropriately control for forest amount. Claims of negative fragmentation effects based on study designs and analyses that do not control for habitat amount can hinder conservation efforts, particularly in human-dominated landscapes, where most patches are small and underappreciated for their contributions to biodiversity conservation. Our reanalysis of Gonçalves-Souza et al. illustrates why appropriate control of the effects of forest amount is critical for resolving the fragmentation debate. The debate is not about the effects of forest loss, on which there is general consensus (Riva, Haddad, et al., 2024). Rather, it is about the effects of fragmentation independent of forest amount. Most multilandscape studies of fragmentation per se show no effects (reviewed in Fahrig [2003, 2017]), just as we found here. The positive effects of forest amount and the lack of fragmentation effects mean that to protect biodiversity, as much forest as possible must be protected, irrespective of patch sizes. We are grateful for helpful comments from D. Deane and 3 anonymous reviewers and to the GLEL Friday Discussion group for catalyzing this study.