Marked tree demographic variation along subtle elevation differences partially explains species' habitat associations in an Amazonian forest

It is widely recognized that large-scale topographic variation affects the distribution of tree diversity, yet the effects of topography at smaller scales are less appreciated but can be no less consequential. We evaluated how small-scale topographic variation affects tree demography and diversity in a hyperdiverse Amazonian forest where species distributions respond strongly to elevation differences as small as 22 meters. For topographically structured species distributions to arise, species should grow and survive (perform) better in the topographic habitat they are associated with (best-at-home hypothesis), and they should outperform other species that are found, but not strongly aggregated, on that habitat (resident-advantage hypothesis). Here, we tested these demographic hypotheses using data on the growth and mortality of 79,911 trees (352 species) among three topographic habitats (valleys, slopes, and ridges) in the 25-ha Amacayacu Forest Dynamics Plot. Despite the small variation in elevation, there was significant community-level variation in growth and mortality among topographic habitats: trees growing in valleys, where soil moisture is higher, had significantly higher growth and mortality rates than those growing on slopes and ridges. However, tree growth rates did not depend on, and mortality rates varied inconsistently with, species’ habitat association. Our results partially support the best-at-home or resident-advantage hypotheses for valley-associated species, which grew best in their home habitat (valleys) than elsewhere and had lower mortality there compared to slope-associated or generalist species (foreigners). For slope- and ridge- associated species our results did not support these hypotheses at the community level. Species-specific analyses revealed that 73 out of the 352 species analyzed at the community level supported either hypothesis. Synthesis. Our findings show that even small differences in elevation can lead to biologically meaningful variation in resource access that translates into significant differences in tree growth and survival. However, resource access could not fully explain the patterns of topographically driven demographic variation we observed. While certain species may still exhibit home and resident advantages in specific habitats, even when community-level averages partially reflect this pattern, alternative hypotheses are likely driving the patterns observed at the community level.

Methods
Study area
The dataset originates from the Amacayacu Forest Dynamics Plot (AFDP), a 25-hectare tropical wet forest plot in the Northwestern Amazon (3°48ʹS, 70°16ʹW). The AFDP, part of the Forest Global Earth Observatory (ForestGEO), features a unimodal rainfall regime (~3,216 mm/year), high humidity (86%), and a mean annual temperature of 25.8 °C. The plot encompasses diverse topographic habitats including ridges, slopes, and valleys, with generally poor, acidic soils dominated by kaolinite and quartz minerals.

Forest censuses
Two full censuses were conducted: the first from August 2007 to April 2009, and the second from August 2014 to November 2015. All woody plants ≥10 mm diameter at breast height (DBH; 1.3 m) were mapped, tagged, measured, and identified taxonomically, with vouchers deposited at the Herbario Amazónico Colombiano (COAH).

Topographic Habitat Classification
Three topographic habitats—ridges, slopes, and valleys—were defined using hierarchical clustering based on elevation, slope, and convexity metrics at a 20 m × 20 m quadrat scale, derived from a 5-m resolution topographic survey. These habitats account for approximately 25%, 31%, and 44% of the plot area, respectively.

Species habitat associations
Species preferences for the three habitats were tested using a Torus Translation test on 106,230 trees of 441 species. Species with fewer than 10 individuals per habitat were excluded. Associations were identified based on deviations from null model expectations; species were classified as positively associated, neutral, or repelled from each habitat.

Growth rate estimation
Tree growth rate (G) was calculated as the annualized difference in DBH between censuses, applying taper corrections for measurements taken at non-standard heights. Wood specific gravity values were assigned based on taxonomic identity. Extreme growth outliers were excluded, and growth rates were transformed using a modulus power transformation (t = 0.39) to reduce skewness before statistical analyses.

Mortality assessment
Mortality was recorded as a binary variable (dead/alive) based on the absence of live foliage and other vitality indicators. Instantaneous mortality rates (% per year) were calculated from the probability of death between censuses. The final dataset used for modeling included 352 species with 68,380 individuals for growth analyses and 79,911 individuals (12,943 dead) for mortality analyses.