3.1 Distribution pattern of microclimate and microhabitat structure
We found high horizontal and vertical variation in microclimate (air temperature: R2 = 0.92,F 7, 59 = 47.72, P < 0.01; relative humidity: R2 = 0.91,F 7, 59 = 81.96, P < 0.01) and microhabitat structure in our study site (Fig. 2). In general, air temperature increased and relative humidity decreased with vertical height based on linear regression model (Fig. 2). In addition, we found significant effects of horizontal position on temperature and humidity (Table S2). The estimated vertical change in air temperature was 0.13˚C per 10 m, while estimated horizontal change was 0.36 ˚C per 10 m (Table S2). The relative humidity was estimated to decrease 1.4% per 10 m vertical increase, and the horizontal variation was also 1.4% per 10 m horizontal distance. The photosynthetic photon flux density (PPFD) increased with vertical height (as PPFD was normalized within each vertical transect, we could not compare this across transects,R2 = 0.35, F 1, 62 = 32.84, P < 0.01; Table S2, Fig. 2). The total leaf area generally decreased towards the upper canopy but did not show significant horizontal variation (note that leaf area data was missing for Transect 7; R2 = 0.22,F 6, 48 = 2.32, P = 0.05; Table S2, Fig. 2).
Distribution of ant assemblages along vertical and horizontal gradients
In total, 35,710 individual ants from 138 species in 31 genera were sampled. Both ant abundance (ln (X+1) tranformed,R2 = 0.67, F 7, 59 = 2.32, P < 0.01) and richness (R2 = 0.66, F 7, 59 = 16.65, P < 0.01) significantly decreased with vertical height, and varied across transects (horizontal positions) (Table S3, Fig. 3). Both NMDS plots and adonis analysis showed clustering of ant assemblages both vertically and horizontally (Fig. S2; adonis analysis (abundance based), vertical stratification: F = 1.8, R2 = 0.09, P = 0.01; across transects: F= 6.4, R2 = 0.39, P = 0.001). Dissimilarity partitioning revealed that the pairwise assemblage dissimilarity between sampling points was largely composed by assemblage turnover, meaning the pairwise dissimilarity was mostly explained by replacement of assemblage composition between sampling points. Using either straight line distance or maximum surface travel distance calculated, we detected a significant increase in total (unpartitioned) pairwise dissimilarity with increasing distance between sampling points (MRM analyses; Table 1, Fig. 4). While no distance effects were detected in turnover, there were positive effects of distance on the nestedness component of pairwise dissimilarity. Both vertical and horizontal distance showed significant positive effects on total pairwise dissimilarity between sampling points. The effect size of vertical distance on pairwise dissimilarity of ant assemblages was slightly higher than that of horizontal distance, but the explanatory power of both predictors was low (Table S4).
When pooling the data of sampling points within the same transect or vertical stratum to examine distance-decay patterns between transects/strata, we observed strong effects of vertical distance on pairwise dissimilarity indexes between vertical strata. In contrast, we found no effects of horizontal distance on pairwise assemblage dissimilarity between transects (MRM analyses, Table 2, Fig. 5). When comparing across the same distances vertically and horizontally, pairwise dissimilarity was consistently higher horizontally than vertically (Table 2, Fig. 5).
By comparing the effects of vertical distance on assemblage dissimilarity within vertical transect and effects of horizontal distance within vertical strata, we found greater small-scale pairwise dissimilarity (higher intercepts) over horizontal distance within the same vertical strata than over vertical distance within the same vertical transect (linear regression: R2 = 0.60, F 1, 17 = 25.03, P <0.01, Fig. 6C). However, the effects of vertical distance on pairwise dissimilarity was more apparent and much stronger than that of horizontal distance, presenting a distance-decay pattern within each transect (linear regression: R2 = 0.73,F 1, 17 = 46.45, P <0.01, Fig. 6D). We did not find differing horizontal turnover of ant assemblages in different vertical strata (Fig. 6A, Table 3). We also did not detect significant interactive effects between horizontal distance and vertical height on horizontal pairwise dissimilarity of ant assemblages within each vertical stratum (linear regression: R2 = 0.005, F3 , 127 = 0.22, P<0.89; coefficient of horizontal distance: p = 0.77; coefficient of vertical height: p = 0.95; horizontal distance * vertical height: p = 0.73). Hence, change in horizontal dissimilarity of assemblages did not depend on height, as would be expected if ant mosaics were driving more rapid increase in disimilarity at greater heights in the canopy.
3.3 Explanatory factors for ant assemblage distribution patterns
Results from Canonical Correspondence Analysis (CCA) and backward model selection suggested both air temperature and relative humidity significantly associated with ant assemblage composition, while PPFD and total leaf area did not show significant effects (Table 4, Figure S5).