The research team assembled a dataset of 3,570 occurrence points, drawing on 2,611 records from their own quantitative fieldwork and 959 records from GBIF, each one painstakingly verified against photographs of herbarium specimens to ensure that species identifications were correct. Specimens collected before 1980 were excluded because their georeferencing was considered unreliable in an era before GPS devices and accurate maps became available. Records with known coordinate problems or insufficient precision were also removed, and a spatial thinning step ensured that no two occurrences were closer than 450 metres, reducing the sampling biases that plague continental-scale biodiversity data.

What emerged from this cleaned dataset was a set of species distribution models built with three complementary statistical approaches: Generalized Linear Models, Gradient Boosting Models, and Random Forest models. Each method was trained on 70 percent of the occurrence data and tested on the remaining 30 percent, with performance evaluated using two independent accuracy metrics. The ensemble models that combined all three approaches achieved exceptionally high scores, with True Skill Statistics exceeding 0.78 and ROC values above 0.95 for every species, indicating that the predicted distribution maps captured the real geographic patterns with remarkable fidelity.

The single most powerful predictor across all eighteen species was soil base cation concentration, a measure of the availability of nutrient cations such as calcium, magnesium, and potassium that plants need for growth. This variable appeared in every single species model and was the most important predictor for many of them. Understory temperature and elevation were also included in all models but carried less weight. Cation exchange capacity, a soil property that is more readily available in global databases, proved far less informative, appearing in only half of the models. This finding carries a practical warning for the field: relying on easily accessible but ecologically imprecise soil variables can systematically underestimate the true importance of soil chemistry in shaping plant distributions.

When the researchers classified the modelled distribution patterns, six distinct types emerged, each corresponding to a specific combination of soil and climate preferences. In order of increasing range size, these were Guianan, Andean, southern, circum-Amazonian, bimodal, and widespread. The five species with the smallest ranges were confined to the Guianas, the Andean forelands, or the southern margin of the rainforest biome. Three species displayed a doughnut-shaped circum-Amazonian pattern, absent from the central-western core but present around most of the perimeter. Four species showed a bimodal distribution, occupying both the Andean foothills and the Guianan region while skipping the intervening central and southeastern Amazonia. The remaining five species spanned nearly the entire basin.

The link between these spatial patterns and soil chemistry was unambiguous. The Andean species, whose ranges hug the young, nutrient-rich sediments deposited by rivers flowing down from the mountains, showed occurrence probabilities that dropped essentially to zero on soils with base cation concentrations below 1 cmol(+)/kg. The bimodal species shared similarly high cation optima but avoided both the impoverished soils of central Amazonia and the seasonal climates of the south. In stark contrast, the five most widespread species all had the lowest base cation optima of the entire dataset, thriving on the infertile, highly weathered soils that dominate much of the Amazon basin. Because infertile soils cover such vast areas, species adapted to them can spread almost everywhere, while species requiring nutrient-rich substrates are confined to the geologically young western forelands.

Climate variables told a complementary story. No single climatic measure dominated across all species, but seasonal variability emerged as more critical than averages. Mean diurnal temperature range, temperature seasonality, and precipitation seasonality were selected in more than half of the species-specific models, while total annual precipitation was the most important climatic factor for only one species. This pattern aligns with what is known about fern physiology: these plants are highly sensitive to desiccation and reach their greatest abundance and diversity where water scarcity is not a limiting factor. The circum-Amazonian and southern species, which occupy the more seasonal margins of the biome, were the ones best adapted to these fluctuating conditions.

The forward-looking projections are sobering. Using two shared socio-economic pathways from the CMIP6 climate modelling framework, with consensus bioclimatic variables averaged across five global climate models, the researchers modelled range shifts for the 2030s, 2050s, and 2070s. Under the extreme SSP5-8.5 scenario, fourteen of the eighteen species were predicted to suffer a net loss of suitable habitat by the 2070s. Eight species would lose more than half of their currently suitable area, and five of these hardest-hit species are among the most widespread and abundant in the entire genus. Only three species were projected to gain new suitable area exceeding 20 percent of their current range, and the sole species predicted to expand by more than 50 percent was Adiantum tetraphyllum, whose distribution already extends beyond Amazonia into the drier biomes to the south.

These findings challenge an earlier assumption that widespread species are inherently more resilient to climate change. On the contrary, the very species that dominate the core rainforest understory, adapted to stable, humid, nutrient-poor conditions, appear most vulnerable as temperatures rise and rainfall patterns shift. The researchers note that their models did not explicitly account for deforestation, fragmentation, fire, or dispersal barriers, meaning the true losses will likely be even greater than projected. Ferns are generally efficient dispersers, but whether they can colonise new suitable habitat fast enough to keep pace with the deterioration of their current ranges remains an open and troubling question.

Beyond the immediate fate of eighteen fern species, the study carries a broader message for tropical biogeography. Because all plants respond to the same fundamental gradients of temperature, moisture, and nutrient availability, the six distributional types identified here are unlikely to be unique to Adiantum. Similar patterns almost certainly characterise other understory herbs, shrubs, and even some tree lineages across Amazonia. As climate change pushes the basin toward hotter and more seasonal conditions, the forests of the future may increasingly resemble the dry forests that today fringe their southern and eastern edges. Whether the species capable of surviving that transition can actually reach the new habitats in time is one of the most consequential unknowns facing Amazonian biodiversity in the coming decades.

Subject of Research: Amazon-wide distribution patterns and climate change vulnerability of understory fern species

Article Title: Amazon-wide distribution patterns of understory plants: Effects of soils and changing climate

Article References: Gupta, R., Zuquim, G., Moulatlet, G. M., Lehtonen, S., Sirén, A., Alonso, A., Cárdenas, G. G., & Tuomisto, H. (2026). Amazon-wide distribution patterns of understory plants: Effects of soils and changing climate. Frontiers of Biogeography, 19, Article e188764. https://doi.org/10.21425/fob.19.188764

Image Credits: AI Generated

DOI: 10.21425/fob.19.188764

Keywords: Amazonia, Adiantum, ferns, species distribution models, soil base cation concentration, edaphic niche, climate change, understory plants, biogeography, SSP5-8.5, habitat loss, Frontiers of Biogeography