A synthesis of 1,905 seed traps across 52 fragments and 1.3 million seeds found species richness tracking landscape cover rather than how large any single patch is.

Only 23 to 24 per cent of Brazil’s Atlantic Forest remains, scattered across thousands of fragments in one of the most densely populated parts of South America. Conservation planning in that setting is largely a question of which fragments to protect, and the usual instinct is to favour the big ones.

A study in the Journal of Ecology suggests the instinct is pointed at the wrong variable. Across 52 fragments spanning the length of the biome, the diversity of seeds arriving on the forest floor tracked how much forest surrounded a fragment, not how large the fragment itself was.

The first author is Luís Felipe Daibes, who did the work as a postdoctoral fellow at São Paulo State University’s Institute of Biosciences in Rio Claro, with an internship at Virginia Tech, and is now a researcher at the Rio de Janeiro Botanical Garden. More than 60 authors from several institutions contributed.

The dataset

The scale is what makes the result worth attending to. The synthesis draws on 1,905 seed traps, between 20 centimetres and a metre in diameter, placed in the forest understory and checked monthly, usually across a year. They collected 1.3 million seeds, from which 1,029 taxa were identified at least to family.

The fragments ranged from about 4 hectares to several thousand contiguous hectares, and the underlying field studies were conducted between 1987 and 2021, from the Northeast of Brazil to the South.

“No study had ever been conducted covering the entire extent of the biome, from the Northeast to the South of the country. This lends considerable robustness to our results,” Daibes said.

Seed rain is a well-chosen measure for this question. It is the input to forest regeneration — what arrives and could germinate — and because most of those seeds were carried in by animals, the composition of the seed rain also reports indirectly on whether the animal community that moves them is intact.

Cover, not size

The central finding is that species richness in the seed rain rose with surrounding forest cover and with precipitation, regardless of fragment size.

“Forest cover proved to be a key metric for assessing landscape composition, or the amount of habitat. It’s the factor that most influences the parameters of seed fall in tropical forests,” Daibes said.

Wetter areas had a higher proportion of zoochorous seeds — those dispersed by animals — while fragments in drier landscapes with little surrounding cover were dominated by pioneer species.

The counterintuitive part concerns quantity. Highly fragmented areas can actually produce more seeds, which sounds like good news and is not.

“The interesting thing is that highly fragmented areas can actually produce a greater quantity of seeds. However, these landscapes tend to be dominated by a few generalist species with many small seeds,” Daibes said.

That pattern has a clear ecological reading. Small-seeded generalists are wind-dispersed or carried by the hardiest animals, they colonise disturbed edges readily, and a fragmented landscape is mostly edge. Large-seeded trees depend on large-bodied frugivores — the birds and mammals most likely to have been lost from a small isolated patch — and large seeds are disproportionately the ones that grow into canopy trees. A high seed count can therefore describe a seed rain that cannot rebuild a forest.

What pioneers can and cannot do

Daibes was careful not to dismiss the pioneer species, which have a real function.

“Pioneer species, which grow rapidly and don’t necessarily depend on wildlife for dispersal, are important for forest regeneration, providing shade for slow-growing species. However, a functional ecosystem can’t be built with them alone,” he said.

This is the standard succession sequence, and the point is that it stalls. Pioneers establish quickly and create the shaded, humid conditions that late-successional species need, but the late-successional seeds then have to arrive from somewhere. In a fragment cut off from the rest of the forest, with the large frugivores gone, they do not — and the regeneration halts at a permanent early stage that looks like recovering forest from a distance.

A number planners can use

The study’s practical output is a threshold, which is more than most landscape ecology papers offer.

“If only a single fragment can be conserved within an agricultural area, it’s important that the region have some surrounding forest cover. If there are several other fragments within a five-kilometre radius with more than 60 per cent forest cover, there’s a greater chance of seed exchange between them, thereby maintaining diversity,” Daibes said.

Framed as guidance, that shifts the unit of decision from the patch to the neighbourhood: the question is not how big a fragment is but what is within five kilometres of it. For a region at 23 per cent remaining cover, it also implies that much of what survives sits below the threshold at which exchange is likely — which is an argument for restoration placed to connect existing fragments rather than restoration placed wherever land is cheapest.

The limits of a synthesis

The dataset’s strength is also where its caveats live, and the release does not raise them.

Field studies conducted between 1987 and 2021 were designed to answer different questions, with traps of varying size — a 20-centimetre trap and a one-metre trap do not sample the same way — and sampling periods described as usually a year rather than uniformly so. Harmonising that into one analysis is necessary and introduces heterogeneity that a purpose-built survey would not have.

A span of three decades also means landscape conditions were not measured at one moment. Forest cover around a fragment sampled in 1989 was not what it was in 2019, and the correlation between cover and seed richness is assembled across that drift.

Seed rain is an input, not an outcome. Seeds arriving are not seedlings surviving, and germination, herbivory and competition sit between the two. The study measures what reaches the forest floor, which is the right thing to measure for dispersal and is one step removed from whether the forest actually regenerates.

The work was supported by FAPESP, which funded Daibes’s postdoctoral fellowship and a Thematic Project, along with a series of grants from the 2000s onward that generated the underlying data. The paper was published on 6 May 2026.

---

Notes

1. Daibes, L.F., et al. “Landscape features predict broad-scale seed rain patterns across fragments of the Brazilian Atlantic Forest.” Journal of Ecology, 2026. DOI: 10.1111/1365-2745.70328.
2. Fundação de Amparo à Pesquisa do Estado de São Paulo. “In Atlantic Forest restoration, connection between fragments is more important than fragment size.” EurekAlert!, 4 August 2026. https://www.eurekalert.org/news-releases/1138771.

Luis Felipe Daibes/Rio de Janeiro Botanical Garden