Changes in the functional structures of communities are rarely examined along multiple large-scale environmental gradients. Here, we describe patterns in functional beta diversity for New Zealand marine fishes vs depth and latitude, including broad-scale delineation of functional bioregions. We derived eight functional traits related to food acquisition and locomotion and calculated complementary indices of functional beta diversity for 144 species of marine ray-finned fishes occurring along large-scale depth (50 - 1200 m) and latitudinal gradients (29° - 51° S) in the New Zealand Exclusive Economic Zone. We focused on a suite of morphological traits calculated directly from in situ Baited Remote Underwater Stereo-Video (stereo-BRUV) footage and museum specimens. We found that functional changes were primarily structured by depth followed by latitude, and that latitudinal functional turnover decreased with increasing depth. Functional turnover among cells increased with increasing depth distance, but this relationship plateaued for greater depth distances (> 750 m). In contrast, functional turnover did not change significantly with increasing latitudinal distance at 700 - 1200 m depths. Shallow functional bioregions (50 - 100 m) were distinct at different latitudes, whereas deeper bioregions extended across broad latitudinal ranges. Fishes in shallow depths had a body shape conducive to efficient propulsion, while fishes in deeper depths were more elongated, enabling slow, energy-efficient locomotion, and had large eyes to enhance vision. Environmental filtering may be a primary driver of broad-scale patterns of functional beta diversity in the deep sea. Greater environmental homogeneity may lead to greater functional homogeneity across latitudinal gradients at deeper depths (700 - 1200 m). We suggest that communities living at depth may follow a ‘functional village hypothesis’, whereby similar key functional niches in fish communities may be maintained over large spatial scales.

Fish community data

Baited Remote Underwear Stereo-Video systems (Stereo-BRUVs) were used to sample marine ray-finned fishes (Class Actinopterygii) in situ at off-shore locations across northern, eastern and southern New Zealand (see Zintzen et al. 2012; 2017 for detailed positions). The Stereo-BRUVs were deployed in a stratified random sampling design at each of seven depths (50 m, 100 m, 300 m, 500 m, 700 m, 900 m and 1200 m) within each of seven locations (from north to south): Rangitāhua, the Kermadec Islands (KER), Three Kings Islands (TKI), Great Barrier Island (GBI), Whakaari, White Island (WI), Kaikōura (KKA), Otago Peninsula (OTA) and the Auckland Islands (AUC) that spanned 21° of latitude in New Zealand waters (with n = 5 - 7 replicate deployments per depth-by-location, see Figure 1 from Zintzen et al. 2017 for a detailed map showing exact sampling locations). Video footage was obtained from a total of 329 deployments (2 hours each) across 47 depth-by-location cells (2 cells were not sampled – White Island at 1200 m and Auckland Islands at 1200 m, due to poor weather conditions). For further details regarding the sampling design, the Stereo-BRUV apparatus and deployment, calibration of measurements and associated methodologies, see Zintzen et al. (2012; 2017).

Functional traits

Fifteen raw morphological measurements were obtained from individuals of each species of fish, in situ, by reviewing footage obtained from each Stereo-BRUV deployment and using the software ‘EventMeasure’ (www.seagis.com.au; see Myers et al. (in press) and Table S1 in Supporting Information). Where possible, measurements from multiple individuals of a single species within a given depth-by-location cell were obtained. A complete set of morphological measurements were not always possible to obtain for every species observed in the video footage. For individuals that were missing no more than 3 (out of 15) measurements, the missing values were imputed using a random-forest machine-learning algorithm (Stekhoven & Bühlmann 2012), based on the other individuals of that species in the dataset having a complete set of measurements. This imputation relies on the assumption that relationships among the morphological variables remain constant within a given species. In addition, to ensure we would have a full set of measured traits for every fish species, we also took raw morphological measurements directly from two preserved museum specimens (held within the National Fish Collection at the Museum of New Zealand Te Papa Tongarewa, Wellington) for every species seen in the video footage (voucher registrations are provided in Table S2 in Supporting Information). In total, there were 144 species recorded across the 47 depth-by-location cells, and 509 species-by-cell occurrences. The original dataset comprised a complete set of 15 raw morphological measurements for 722 individuals observed in video footage (136 of these required some random-forest imputation, and missing traits were remeasured for 4 individuals), plus 291 museum specimens.

We calculated 8 trait variables, namely: eye size, oral gape position, jaw length relative to head length, elongation, eye position, caudal peduncle throttling, pectoral fin position and total body length – each as a function of the 15 raw morphological measurements (2 of the raw morphological measurements were used only for data imputation, Table S3, Supporting Information). These morphological traits focused on key aspects of locomotion, visual perception and feeding for fishes that correspond to important functional variations in the body plan and structure of fishes across large depth gradients (Myers et al. 2019).

We obtained representative trait values for every species within every cell in the study design, while taking into account the intraspecific trait variability. To do so we compiled a table of 8 unique traits (columns) for each species in each depth-by-location cell (509 rows), we randomly drew 1 individual from the list of all complete individuals for each species that were (in order of preference): (i) within that depth-by-location cell, (ii) at the same depth, (iii) from anywhere within the Stereo-BRUV study design or (iv) from a museum specimen. We replicated this random-draw procedure 100 times to generate 100 species-cell × trait (509 × 8) data tables. These data tables enabled us to build 100 multivariate functional spaces based on the 8 normalised continuous trait variables that were used to compute the Euclidean distances between species. By calculating beta diversity values for all 100 tables, then averaging these values, we were able to integrate the available individual-level (within-species) morphological variation into the study, given the logistic constraints on the number of individuals of each species we were able to measure, while also maintaining spatial variation in morphologies encountered within each species as well as possible.

Measures of functional beta diversity

We calculated the functional turnover, or functional beta diversity, by considering the functional distances between each of the species occurring within one cell, with every species occurring within another cell (Swenson 2014). We calculated the following metrics between every pair of cells: (i) mean pairwise functional distance (MPFD.beta) which corresponds in the beta context to the mean distance in functional space between all pairs of species across two cells (Swenson 2014), and (ii) mean nearest neighbor distance (MNND.beta) which corresponds in the beta context to the average of the minimum functional distance between each species in one cell, to every species in another cell (Swenson & Weiser 2014). Previously, MPFD has been defined in an alpha context as the functional analogue to average taxonomic distinctness (Clarke & Warwick 1998), and is also called mean phylogenetic pairwise distance (Swenson 2014) when used in a phylogenetic context. MNND, also called Gamma+ (Clarke et al. 2006), has been used previously in both phylogenetic (Webb et al. 2002) and functional contexts (Swenson & Weiser 2014; Pigot, Trisos & Tobias 2016) where it has been used to estimate functional originality (Mouillot et al. 2013; Leitao et al.2016), and can be considered as an indicator of differences in niche (Swenson et al. 2020). These two functional beta diversity metrics allow the full dimensionality of the functional space to be entirely maintained, which is not necessarily possible with earlier-described metrics, such as convex hulls (Villéger et al. 2013), or hypervolumes (Blonder et al. 2014; 2018). These typically require a rather drastic reduction in dimensionality, especially for species-poor communities such as those encountered in the deep sea.

We calculated the MPFD.beta and MNND.beta metrics between every pair of cells for each of the 100 species-cell by trait (509 × 8) data matrices, then computed the mean and standard deviation across the 100 tables for subsequent analyses. The result was a 47 × 47 matrix of functional dissimilarities (either MPFD.beta or MNND.beta) among all pairs of cells in our study design.

We provide everything else that is needed to know in the metadata documents for each data table.