Statistical analysis
All statistical analyses were performed using R version 4.3.2 (https://www.R-project.org/). To evaluate sampling completeness and compare the detection efficiency between bio-sampler species or genetic markers, we generated species accumulation curves using the random method with 100 permutations and estimated the expected species richness using Chao2 (Chao et al., 2014), with the vegan package version 2.6 (Oksanen et Blanchet, 2017). Following Deagle et al. (2019), dDNA metabarcoding data were summarized asĀ : (1) the Frequency of Occurrence (FOO), defined as the percentage of samples in which a fish species was present, and (2) the Relative Read Abundance (RRA), calculated as the average proportion of total reads assigned to the species.
Fish assemblages recovered through dDNA, WFD surveys, and deep inventory were compared using multidimensional scaling (MDS) based on Jaccard dissimilarity matrices. To assess potential size-related detection biases across monitoring methods, we compared the body length distributions of detected species using Wilcoxon rank-sum tests. For each species, we used the mean body length recorded from historical WFD surveys conducted in French Guiana. Additionally, to determine whether detection probability via dDNA metabarcoding was influenced by species abundance, we compared the catch abundances between species detected and not detected by metabarcoding using Wilcoxon rank-sum tests. Finally, we also tested whether species exclusively detected through metabarcoding exhibited different frequencies of occurrence compared to species detected by both metabarcoding and traditional surveys.