Abiotic and biotic factors structure species assembly in ecosystems both horizontally and vertically. However, the way community composition changes along comparable horizontal and vertical distances in complex three-dimensional habitats, and the factors driving these patterns, remains poorly understood. By sampling ant assemblages at comparable vertical and horizontal spatial scales in a tropical rain forest, we tested hypotheses that predicted differences in vertical and horizontal turnover explained by different drivers in vertical and horizontal space. These drivers included environmental filtering, such as microclimate (temperature, humidity, and photosynthetic photon flux density) and microhabitat connectivity (leaf area) which are structured differently across vertical and horizontal space. We found that both ant abundance and richness decreased significantly with increasing vertical height. Although dissimilarity between ant assemblages increased with vertical distance, indicating a clear distance-decay pattern, the dissimilarity was higher horizontally where it appeared independent of distance. The pronounced horizontal and vertical structuring of ant assemblages across short distances is likely explained by a combination of microclimate and microhabitat connectivity. Our results demonstrate the importance of considering three-dimensional spatial variation in local assemblages and reveal how highly diverse communities can be supported by complex habitats.

Ant assemblages were sampled using insecticide fogging between 0700 and 0930 on 12-24 May 2002 from within the canopy by a climber rappelling down using a Swing-Fog model SN50 (Phoenix Fogger, Dallas, TX, USA) near each of seven vertical sample transects. These vertical transects were suspended from a 130 m horizontal traverse line secured in the upper canopy and were arranged at 20-25 m intervals horizontally (Fig. 2). Each of these transects supported multiple individual circular fogging trays (1 m²) (n=86) suspended in the air with attached ethanol-filled collecting bottles spaced at approximately 5 m vertical intervals beginning at 1 m above the ground (Fig. 2). These trays collected knocked-down arthropods that were between trays at the time of fogging. A 1.6% aqueous solution of the synthetic pyrethrum (Cypermethrin) was used. Arthropods were collected into 80% ethanol 1-2 hours after fogging, and ants were separated as part of arthropod ordinal sorting (see Dial et al. 2006 for results on ordinal arthropod assemblages). The ants sampled using fogging are mainly diurnal foraging species active during the sampling period (0700 to 0930), and therefore likely present a subset of the total local ant diversity. In total, we obtained and identified ant assemblage samples for 61 out of 86 sampling points, with 14 samples having no ants, nine samples having been lost between sampling and analysis, and two samples in Transect 5 (two individual ants discovered) belonging to an emergent forest layer without horizontal positions for comparison.

To quantify microclimate and habitat structure, air temperature (˚C) and relative humidity (%) were measured at 0.5 h intervals over 24 hours using Hobo Pro RH/Temperature Data Logger (Onset Computer Corporation, Pocassest, MA, USA). Data loggers were placed at 3 m intervals along each vertical transect, starting 1 m above the ground. At the data logger locations, photosynthetic photon flux density (PPFD) was recorded using a handheld light meter (Quantum Lightmeter, Spectrum Technologies, Plainfield, 1L, USA) and normalized by dividing by maximum light value within each vertical transect to account for between-day variation in lighting. The intent was to identify the relative (not absolute) light environment of the forest canopy (Dial et al 2006). We estimated one-sided total leaf area between sampling trays as a measure of microhabitat connectivity at different sampling points for transects 1 to 6 (T1-T6; Fig 2c). The leaf area within a sampling interval was calculated by multiplying the number of leaf intersections by the size of the base area of the interval which was 1 m² (the area of the sample tray). We then used these data to estimate leaf area index (LAI) over vertical intervals (sampling methods described in Dial et al. 2004, 2006, and 2011; estimation methods in Dial et al. 2006 and 2011). Conceptually, LAI refers to the number of leaf layers above the ground surface that would be pierced by a vertical line. For example, if LAI = 7, then there are, on average seven leaf layers above a random point on the ground within that height range; or 7 m² of leaf area per m² of ground surface. We assumed (following MacArthur and Horn 1969) that for any sample point in the canopy located at height z above the ground, the foliage density was approximately equal in all directions. Following this assumption at each height z, we systematically measured horizontal distances (d_i) with a laser range finder to the nearest canopy element (foliage and stems) in 12 uniformly distributed azimuths every 2 m vertically from the ground to the height of the horizontal traverse line supporting the vertical transect. Using the n ≤ 12 distances to foliage at each sample point, we found the mean distance (d) to foliage, doubled the mean (assuming that the observer was on the average midway between foliage elements), then inverted it to find leaf intersections per vertical meter at height z as LAI_z = 1/(2d). By multiplying the LAI_z by collection area (1m²) we estimated the leaf area sampled within the interval.