We used r (R Development Core Team 2017 ) for statistical analyses, with all recorded fish species included. We used the findCorrelation function from the caret package to identify a set of 17 predictors that were not strongly correlated with each other (based on Spearman’s correlation coefficient <0.7; see Supporting Information Table S2 for list of all variables measured). To determine at what spatial scales fish–habitat associations are the strongest (Question 1), we used the BIOENV procedure (Clarke & Ainsworth, 1993 ), which is a dissimilarity-based method that can be used to identify the subset of explanatory variables whose Euclidean distance matrix has the maximum correlation with community dissimilarities, in our case, based on Bray–Curtis dissimilarity. BIOENV was implemented with functions from the vegan and sinkr packages. We extracted the rho value for the best model at each spatial scale as a measure of the strength of fish–habitat associations, with a higher rho value indicating a stronger association between fish and habitat variables.
We hence computed the potency of fish–environment connectivity that could be asked built purely for the peak away from replication at each and every size in the lack of one seafood–habitat relationships, immediately after which examined if our BIOENV performance was more powerful than so it null expectation
To do so, we at random resampled the first 39 BRUV samples of complimentary seafood–habitat investigation collected on a hundred-m size, generate an entire meant amazing dataset (we.age., 72 trials). So it dataset was split into a couple of independent matrices, one to who has new seafood and something the fresh new habitat data, and the rows was basically randomly shuffled to eradicate seafood–environment connectivity from the 39 rows regarding brand spanking new analysis. Matrices was in fact following inserted together with research aggregated of the summing all step 3, six and you can twelve rows of your simulated one hundred yards dataset so you can create new null Fitness dating distributions of your own 300-yards, 600-yards and you may step 1,200-meters balances. This course of action was regular to generate 999 BIOENV habits for every spatial size, on indicate and 95% believe menstruation of the greatest design rho at each and every measure computed round the all simulations. I made use of a-one shot t attempt examine whether your rho to discover the best model predicated on the empirical data was somewhat distinct from the newest rho values questioned at each and every spatial scale in accordance with the artificial study. If the observed rho is actually higher, it can mean that fish–environment contacts try more powerful than was requested by chance, after bookkeeping having differences in testing efforts. We also-ran a power study for every single spatial size playing with the new pwr.t.attempt setting and removed the end result proportions (Cohen’s d), which allows us to take a look at of which spatial level the real difference between seen and you may empirical rho values try finest. We also ran BIOENV patterns toward 300-m and you will step one,200-meters spatial bills using the UVC studies. So it assessment is incorporated to look at surface amongst the UVC and you can BRUV sampling techniques within these types of bills.
We plus compared the fresh parameters identified as becoming most influential when you look at the the new BIOENV data for each and every spatial scale considering the observed BRUV studies, in which we had four spatial balances examine
To assess if environmental predictors of fish are scale-dependent (Question 2), we calculated Pearson’s correlations between the abundance of each fish species and each habitat variable at each scale. We then converted all these correlations to absolute values (i.e., all negative correlations were multiplied by ?1). We compared how the rank order of habitat variables varied between spatial scales based on this absolute Pearson’s correlation coefficient by calculating Kendall’s tau for all pair-west correlations. Kendall’s tau is used to measure ordinal associations between two measured variables (in our case a pair of spatial scales), with a value of 1 when observations (in our case Pearson’s correlation coefficients describing fish abundance–habitat correlations) have identical ranks, and ?1 when the ranks are fully different. Statistically significant (p < 0.05) values indicate that ranks are not different between comparisons.