**3. Results**

The reef fish abundances (video transect survey) on the reef sites varied significantly by reef. The eight reef sites differed in rugosity (0.046 to 0.410), hard coral cover (0.0 to 21.4%), soft coral cover (0.0 to 27.7%), sponge cover (1.0 to 15.5%), fleshy algae cover (0.0 to 86.4%), calcareous algae cover (0.0 to 13.8%), turf algae cover (8.4 to 63.4%) and sand cover (0.0 to 42.7%). The differences in abundance (video transect survey) per site accounted for 76.9% of the variance in the use of natural structures (structure survey) and 78% of the variance in the use of artificial structures (structure survey) (Table S2). There was no significant effect of year on the use of the structures by the reef fish on either the natural (F = −1.4376, df = 1.279, *p* = 0.1517) or artificial structures (F = −1.5778, df = 1.127, *p* = 0.1171). The total number of reef fishes using the natural structures was significantly more than the total number of reef fishes using the artificial structures (F = 5.8812, df = 1.166, *p* = 0.0164) (Figure 3). There was, however, no statistically significant preference for a single type of structure, although diseased colonies had the most individuals associated with them and the soft artificial structures had the least. There were significantly more omnivore (F = 5.6278, df = 1.155, *p* = 0.0189) and invertivore individuals (F = 6.5732, df = 1.123, *p* = 0.0116) preferring natural structures over artificial structures; they preferred natural sea rod structures the most, although not significantly so. Herbivore use of the structures was not significantly influenced by the overall state of the structure (artificial versus natural) (F = 1.7337, df = 1.162, *p* = 0.1898), but utilized particular structure types significantly more than others, using diseased coral heads significantly more than any other structure type, and used natural sea rod structures the least. There was no significant effect of structure state or type on the number of piscivorous fishes (F = 0.2082, df = 1.115, *p* = 0.6491).

The data for the fish species that were most prevalent in the functional feeding guilds were analyzed to determine the reef characteristics that appeared to influence their use of habitats. Among the herbivores, the presence of bicolor damselfish (*Stegastes partitus*) was strongly correlated with low-complexity reefs (offshore) (F = 11.3251, df = 1.92, *p* = 0.0011). Striped parrotfish (*Scarus iseri*) were observed significantly more around natural and artificial structures when their abundance on the reef (video transect survey), as observed in the video transect surveys, was high (F = 8.2186, df = 1.84, *p* = 0.0052). This was also the pattern observed with the two most frequent invertivore species: blue-striped grunts (*Haemulon sciurus*) (F = 6.8165, df = 1.57, *p* = 0.0115) and white grunts (*Haemulon plumierii*) (F = 14.0461, df = 1.85, *p* = 0.0003). The most common piscivore was the schoolmaster snapper (*Lutjanus apodus*), which was significantly more numerous at natural structures (F = 6.3776, df = 1.27, *p* = 0.0117) than at artificial structures. The omnivores were mostly represented by bluehead wrasses (*Thalassoma bifasciatum*), which were also more numerous at natural structures than at artificial structures (F = 16.5063, df = 1.129, *p* < 0.0001) and at the low-complexity offshore sites (F = 4.6595, df = 1.129, *p* = 0.0327).

The principal component analysis identified three axes which best summarized reef character and substrate composition (Table 1). The first component score (PC1) accounted for 35.1% of the variation and loaded positively with both distance from the shore and depth, but negatively with physical rugosity. The axis characterized reef structural differences between nearshore and offshore reefs in the middle Keys [51–53]. The second component score (PC2) accounted for 27.0% of the variation and loaded positively with calcareous algae/sand cover and negatively with fleshy/turf algae cover. The third component score (PC3) accounted for 10.7% of the variance and loaded positively with sponge/hard coral cover and negatively with soft coral cover. The three component scores were used to evaluate whether fishes' use of structure varied with differences in the physical structure, algal substrates, or hard/soft coral substrate.

**Figure 3.** Log n (ln) mean per structure of the counts for all reef fishes combined (gray) and for each functional group herbivores (green), omnivores (orange), predators (blue), and invertivores (yellow)—obtained during the time-lapse photography observations on the different natural and artificial corals. Error bars are ±1 SEM. Differences in the letter located over each bar indicate when values were shown to be statistically different from one another by Tukey's post hoc comparisons.


**Table 1.** Principal component score correlations for reef substrate measures. Bold indicates loadings with *p* < 0.05.

We included four covariates in the ANCOVA model for fish use on individual structures including fish abundance (as estimated by video transect surveys), PC1 topography, PC2 algal cover, and PC3 coral cover (Figure 4). The fully fitted ANCOVA for all reef fishes combined was significant (F = 3.04, df = 12,164, *p* = 0.0007) with an adjusted r2 = 0.122 (Table 2). Four factors significantly contributed to the model: reef fish abundance on the video transects (F = 5.62, df = 1, *p* = 0.0189, estimate = 0.531), structure state (whether artificial or natural) (F = 10.24, df = 1, *p* = 0.0015, estimate (artificial) = −0.3601), topography (PC1) (F = 7.182, df = 1, *p* = 0.0058, estimate = 0.2793), and the structure type by rugosity interaction (F = 6.495, df = 1, *p* = 0.0117, estimate = 0.2427).

**Table 2.** Nested analysis of covariance with fish abundance (from video transect surveys), topography, algal cover, and coral cover as covariates and structure type (control, soft, barrel, elkhorn, sea rod, dead coral, diseased coral, and healthy coral) nested within structure state (natural versus artificial). Bolded values indicate significant values.


**Figure 4.** Linear regressions of mean numbers of fish observed (log-transformed) for the structure surveys within a reef site versus the three component scores that best describe the reef characteristics. The three component scores, from left to right, are topography (PC1), algae (PC2), and coral cover (PC3). The specific loadings of the reef measures for each component score are in Table 1. Graphs include all reef fish combined (gray), herbivores (green), omnivores (orange), invertivores (yellow), and predators (blue).

Nested ANCOVAs for each separate feeding guild of fish revealed differences in the use of the structures among fish guilds. For herbivores, the fully fitted ANCOVA was significant (F = 4.55, df = 11,162, *p* < 0.0001) with an adjusted r2 = 0.184 (Table 2). Three factors significantly contributed to the model: herbivore abundance in the video transect surveys (F = 8.41, df = 1, *p* = 0.0042, estimate = 0.5265), topography (F = 17.25, df = 1, *p* < 0.0001, estimate = 0.2217), and the type of natural structure (F = 2.52, df = 1, *p* = 0.0232) (Table 2). Herbivore use of the structures demonstrated a strong negative relationship with reef rugosity (or a positive relationship with deeper, offshore reefs) but no relationship with algal cover (Figure 4). For the omnivores, there were two factors that significantly contributed to the model of structure use: depth/rugosity (F = 28.69, df = 1, *p* < 0.0001, estimate = 0.3757) and structure state (F = 5.62, df = 1, *p* = 0.0189, estimate (artificial) = −0.3195) (Table 2). Omnivores demonstrated a strong negative relationship with reef rugosity (or a positive relationship with deeper, offshore reefs) (Figure 4). For invertivores, three factors significantly contributed to the model of structure use: invertivore abundance observed in the video transect surveys (F = 8.16, df = 1, *p* = 0.0050, estimate = 0.3742), topography (F = 11.78, df =1, *p* = 0.0008, estimate = −0.3693), and structure state (artificial or natural) (F = 6.57, df = 1, *p* = 0.0116, estimate (artificial) = −0.4964) (Table 2). In contrast to the findings for herbivores and omnivores, invertivores' use of the structures demonstrated a strong positive relationship with reef rugosity (or a negative relationship with deeper, offshore reefs) (Figure 4). The abundance of piscivores across the video transect surveys was the only factor that significantly contributed to the model of structure use (F = 5.88, df = 1, *p* = 0.0170, estimate = 0.2663) (Table 2 and Figure 4).
