Joint species distribution models in response to a distance gradient
The SWARM pipeline recovered a total of 196 MOTUs. Among these MOTUs, 139 could be attributed to 50 families, 103 could be attributed to 70 genera, and 44 MOTUs were assigned to species. The most common families were Labridae (n = 12), Pomacentridae (n = 10), Myctophidae (n=8) and 46% (23/50) families were represented by 1 MOTU. HMSC requires we subset this full set of data (see Methods) leaving 79 MOTUs from 34 genera in 26 families remaining in our final analyses. HMSC applied to the MOTUs revealed relatively consistent responses across species and families to the occupancy of the two coastal areas (Figure 4a). Most MOTUs showed a positive response to Willemstad (0) vs. Valentijnbaai (1) (87%, n = 66 of 79 MOTUs), with 14 (17%) MOTUs with positive responses with 90% posterior estimate support in contrast to only 1 (1.2%) MOTU with a negative response at this confidence level (Figure 4a). eDNA detected strong positive responses of two cryptic cardinal fish species (Phaeoptyx pigmentaria and Phaeoptyx conklini ) and two pelagic top predator species (Acanthocybium solandri ‘wahoo’ andKatsuwonus pelamis ‘skipjack tuna’) to Willemstad vs. Valentijnsbaai, species absent from visual surveys. Among families with more than 3 representative MOTUs, Myctophidae (mean β = 0.5 [2.5% CI = -0.10, 97.5% CI = 1.13]) and Apogonidae (β = 0.45 [-0.13, 1.07]) show consistent positive responses with > 90% posterior estimate support. Lutjanidae (β = 0.40 [-0.23, 1.04]) and Scombridae (β = 0.32 [-0.28, 0.91]) have a positive response with > 80% posterior estimate support, and Belonidae (β = 0.28 [-0.34, 0.91]), Muraenidae (β = 0.26 [-0.33, 0.87]), Clupeidae (β = 0.25 [-0.33, 0.83]) and Mugilidae (β = 0.21 [-0.39, 0.84]) with > 70% estimate support. In contrast to the MOTUs, joint species distribution models applied to UVC revealed more balanced but weaker occupancy responses, of fewer species (n = 35), to the two contrasting coastal areas (Figure 4b). When comparing consistent sampling effort between eDNA metabarcoding and UVCs diversity estimates (19 samples each with 30 minutes survey time), we revealed weaker discrimination of species occurrence between areas using UVC compared to eDNA metabarcoding: the species-specific standard deviation of β estimates was 1.25 times higher for species from UVC compared with MOTUs from eDNA metabarcoding (mean eDNA = 0.37, mean UVC = 0.46, t = 12.31, p < 0.001). We found a significant phylogenetic signal with λ = 0.69 ± 0.06 (p = 0.025 ± 0.008) in the species-specific estimated coefficients, with for instance low β parameter values for Labridae especially for the Halichoeres genus, intermediate values for Muranidae and high values for Apogonidae.
From the models applied to the UVC data, around half of the detected species show positive and negative responses to Willemstad vs. Valentijnsbaai (48% vs. 52% of 35 species). Only 3 (8%) species showed positive responses with 90% posterior estimate support, but 7 (20%) species showed a negative response at this confidence level. Among families with more than three species detected in UVCs, no families had consistent responses with > 90% posterior estimate support. Only Lutjanidae (mean β = -0.61 [-1.53, 0.17]) had a consistent negative response with > 80% posterior estimate support. Even though the full set of UVC data were available to use in our analysis, we found only a marginal reduction in the standard deviation of β parameters using the full data set (mean full-UVC = 0.34, mean eDNA = 0.37, t = 5.99, p < 0.001), which is equivalent to a 1.07x increase in parameter certainty despite an additional 1.46x increase in sampling units (UVC) and ~ 390 minutes of UVC dive time (Figure 4c). In contrast to eDNA, we found no clear phylogenetic signal for UVC transect β parameters (λ = 6.3 x 10-3 ± 5.4x 10-6 0.06; p > 0.05).