Spatial niche comparison: Realized niche breadth and overlap
The realized niche breadths for both taxa were quantified using Levins’ B2 metric (Levins 1968), which ranges from 0 to 1, with higher values indicating broader habitat ranges. The overlap between those ranges was described with Schoener’s D (Schoener 1968; Warren et al.2008) and Warren’s I (Warren et al. 2008). Both indices range from 0 to 1, i.e. from complete divergence and no overlap to identical niche breadth and complete overlap. Species with perfectly opposite responses to the same environmental gradient may have high niche overlap (Warren et al. 2019). Thus, we calculated the rank correlation coefficients rho (Spearman 1904) to compare correlations to the environmental factors. Both, the realized habitat breadth for each taxon and their overlap were calculated in geographic and environmental space using a sample of 100,000 points from the background, with a tolerance of 0.0001 (Warren et al. 2019). Definitions and additional information about those terms can be found in Supplementary Methods Table 3. All metrics were calculated with the R package ENMTools 1.0 (Warren et al. 2021). The realized niche breadth for each taxon was visualized by projecting their habitat suitability scores on the polygon meshes generated from the point clouds.
We tested whether the scleractinian and octocoral niches were identical with a niche identity test (Warren et al. 2008, 2010), and also performed a background test to compare the habitat overlap calculated with the empirical occurrences to the overlap expected by chance with random occurrences (Warren et al. 2019). Definitions and additional information about those terms can be found in Supplementary Methods Table 3. Identity and background tests were completed in both, geographic and environmental space with the R package ENMTools 1.0 (Warren et al. 2021).
Lastly, we calculated the distribution of suitability scores between taxa and their relative abundances and tested whether the amount of suitable habitat available for each taxon was significantly different with a Mann-Whitney U test.
RESULTS
The most important variables included for model fitting were the % coral, % octocoral, % sponge, % calcareous rock, % igneous rock, % sand, slope, and both, roughness and TEI calculated at 10-mm and 100-mm scales. Fifty-nine octocoral recruits and 32 scleractinians were used to test the performance of the SDMs. Bohl et al. tests (2019) revealed our models outperformed null models, in both geographic and environmental space (Supplementary Methods Figure 4), with AUCTest values in geographic and environmental space of 0.84 and 0.91 for octocorals and 0.77 and 0.89 for scleractinians. In addition, AUCDIFF was 0.01 in geographic space and 0.00 in environmental space, indicating low-level to no model overfitting.
The importance of each environmental predictor to the realized niche breadth of recruits and the marginal response curves for each taxon are shown in Figure 3. The distribution of recruits was not random on the reef. For octocoral recruits, reef geomorphology was more important than the presence of benthic invertebrates to characterize suitable microhabitats. The three most important predictors were roughness at 10 mm scale, the percentage of calcareous rock, and slope (importance values of 0.14 ± 0.02, 0.05 ± 0.01, and 0.04 ± 0.01, respectively; Figure 3A). Highly suitable microhabitats for octocoral recruits were located on fine-scale topographic features of calcareous rock with slopes between 15 and 80 degrees, with holes and crevices < 10 mm being more suitable than bumps and hammocks of similar dimensions (Figure 3, B8). At the scale of 100 mm, we found a positive relationship between habitat suitability and TEI, with higher suitability values on large convex regions of the reef such as boulders (Figure 3, B5). The cover of benthic invertebrates, ranked lower in importance, but also affected the suitability of microhabitats for octocoral recruits. Recruits were not found on live invertebrates, and the models projected that microhabitats in direct contact with cnidarians (i.e. areas covered > 50%) were highly unsuitable for octocorals with suitability scores ~ 0.2; Figure 2, B4 and B10). In comparison, microhabitats in direct contact with sponges were more suitable, with suitability scores ~ 0.5 (Figure 3, B6). The least important predictor for octocoral recruit distributions was sand.
As for octocorals, fine-scale roughness (10 mm-scale) was the most important predictor to characterize the realized spatial niche of scleractinian recruits (importance of 0.09 ± 0.01; Figure 3C). But in contrast to octocorals, the % cover of scleractinian corals and sponges were the most important predictors of suitable habitat (importance of 0.05 ± 0.02, and 0.05 ± 0.02, respectively; Figure 3C). Substrate type was less important. Microhabitats in direct contact with a scleractinian coral or a sponge (i.e. areas covered > 50%), were moderately suitable for scleractinian recruits (suitability scores ~ 0.6 and 0.5, respectively, Figure 3, D2 and D3). Slope and roughness at 100 mm-scale also affected habitat suitability for scleractinian recruits, but in contrast to octocorals, flat and exposed areas of the reef at 100 mm-scales were more suitable (Figure 3, D4). The estimated probability of suitable habitat for recruits of both taxa can be seen in Figure 4, and Video 1 shows the rotation of Figure 4C and 4D in 3D (the Supplementary material)