1. Predicting plant community assembly is challenging in part because the influence of environmental conditions via plant functional strategies and the relevance of mechanisms of community assembly change across habitats and these changes remain poorly studied. 2. To assess how environmental conditions drive species sorting in a tropical dry forest, we used the combined RLQ and Fourth-Corner methods to analyze changes in tree species assemblages among sites with distinct atmospheric and soil drought risks. We asked how variation in potential radiation, soil water availability and evapotranspiration interact with functional traits to explain the cross-site sorting of species lying along a continuum of drought coping strategies ranging from acquisitive to conservative resource-use. 3. Environment and traits were tightly related. Opposing common expectations on the success of strategies in resource limited environments, drought-tolerant conservative species with dense tissues and tardily deciduous leaves were favored on sites with higher resource (soil water) availability. Drought-avoiding acquisitive species with water storing tissues and thin, light leaves with short retention time periods were favored in sites with drier soils. A decoupling of stressing conditions caused by soil and atmospheric aridity combined with differences in species adaptations to each of these factors can explain the apparent discrepancies. 4. Synthesis. Drought stress gradients entailed shifts in community functional composition. We show that atmospheric and soil drought risks can be decoupled and jointly determine species distribution in relation to their functional strategies. By considering atmospheric drought stress, a key yet often overlooked factor in studies of community assembly, we offer a novel explanation to a seemingly contrasting pattern in tropical dry forests where species with acquisitive, rather than conservative strategies, predominate in the most resource-limited yet less water demanding environment. More generally, our results emphasize the need for detailed studies of the multiple habitat-dependent relationships between traits and environment to advance our predictive understanding of species distributions and community assembly.

Data on woody vegetation (free standing individuals with a diameter at breast height, DBH, ≥ 2.5 cm) were obtained from 36 100 m² plots (10 × 10 m) located at low, middle and top parts of slopes in the North (N) and South (S) faces of three hills with elevations between 380 and 780 m a.s.l. (site description details on: Méndez-Toribio, Meave, Zermeño-Hernández, & Ibarra-Manríquez, 2016). Such sampling encompasses the range of floristic and environmental variability in the area and represents water availability gradients that go from sites with high soil moisture and evapotranspiration (e.g., lower N-facing slopes) to sites with low soil moisture and evapotranspiration (e.g., higher S-facing slopes).

            Air temperature (AT) above soil level, potential incoming solar radiation (Rad, including direct and diffuse radiation), potential evapotranspiration in the rainy season (ET₀, months with > 100 mm), and Topographic Wetness Index (TWI) were calculated for each sampling plot. Air temperature was recorded hourly from September 2012 to December 2013, with Hobo® Pro v2, U23-001 data loggers placed on a branch of the tree closest to the plot center at 3 m above the ground, where most of the tree foliage develops. Rad was calculated with the “area solar radiation” module of the Spatial Analyst Tools in ArcGis 9.3 ESRI (2009) using a 20 m digital elevation model (DEM) as input. The DEM was built from topographic vector maps (1:50,000 scale) of the studied landscape (~ 4.02 × 109 km²). Rad was defined as the quantity of solar energy (MJ m^-2) on each pixel (20 × 20 m) of the image and estimated monthly for the year 2014 by assuming a clearly uniform sky. Air temperature and Rad were used to estimate ET₀, a crucial indicator of the potential atmospheric drought stress suffered by a plant (van der Maarel, 2005). Potential evapotranspiration was estimated as: ET₀ = 0.0135 (KT) (Rad) × 0.408 (TD) (TC + 17.8), where TD = T_max -T_min (°C), TC = average daily air temperature (°C) and KT is an empirical coefficient for interior (i.e. non-coastal) regions equal to 0.162 (Zohrab, 2000). The TWI (Beven, 1997) was calculated from an hydrological model (TOPMODEL) via the Raster Calculator module of ArcGis 9.3. The TOPMODEL calculates the wetness index k = ln (A/tanβ). To this end, the DEM of the area is used to create a slope raster (β) and a flow direction raster. This latter raster is used to create a flow accumulation raster, which identifies the upslope contributing area (A) for each pixel.

Further methodological details can be found in the source publication.

The complement dataset pertaining to the species functional traits has been archived in the TRY database and is publicly available from https://www.try-db.org