Estuaries and Study Design
Fish communities were surveyed in two estuaries in south-eastern Australia; Botany Bay, an ocean embayment; and Lake Macquarie, a wave-dominated estuary (Roy et al. 2001; Fig. 1). In both estuaries, three habitats types: natural reefs, artificial reefs (Reef Balls ®) (Folpp et al. 2020) and seagrass beds, were investigated using stereo baited remote underwater videos (hereafter referred to as BRUVs; Cappo et al. 2004; Harvey et al. 2007). Each estuary was monitored during the cooler, winter/spring (June-November), and warmer, summer/autumn (December- May) seasons over 3 years.
BRUVs consisted of two GoPro Hero 7 cameras in SeaGIS housings mounted to a frame, and allowed for accurate length measurements of fish (Langlois et al. 2020). Three replicate BRUVs were deployed at three sites within each habitat during warm and cool seasons, resulting in 5 seasonal trips over the study period. Deployments lasted for 30 minutes, with previous research showing this is a sufficient soak time for estimates of fish diversity and relative abundance for comparative purposes in NSW (Harasti et al. 2015). BRUVs were baited with 500 g of crushed pilchards (Sardinops sagax) secured in a mesh bag which extended ~ 1 metre horizontally from the camera housing, attracting fish into the field of view (Watson et al. 2005).
The estuarine BRUV footage was analysed using SeaGIS ‘EventMeasure’ software (www.seagis.com.au), where the relative abundance (MaxN) was determined for each species (Cappo et al. 2004; Dorman et al. 2012). All individuals were identified using published references (Kuiter 1993), classified to genus, and species where possible.
Estuarine Fish Assemblage Composition
Estuarine fish assemblages were compared using MaxN and their subsequent estimated biomass within the nine functional feeding groups used by Truong et al. (2017) and Holland et al. (2020) (Supplementary Table S1). The nine functional feeding groups included three elasmobranch groups (piscivorous sharks, non-piscivorous sharks, and invertivore rays), and six teleost groups (piscivores, invertivores, soft-bottom fish, omnivores, zooplanktivores and herbivores). These teleost groups were further classified as either ‘coastal’ for predominantly pelagic fish, ‘demersal’ for species that reside on both hard and soft substrate, or ‘reef’ as inhabiting predominantly hard benthic substrate, as this influences their diet composition (Supplementary Table S1 and S2; Truong et al. 2017). The functional feeding group biomass were compared among habitats, estuaries and between the cooler and warmer seasons.
To estimate the biomass of the functional feeding groups, for each deployment the lengths of all individuals contributing to MaxN were measured in EventMeasure (per Langlois et al., 2021). Length data was then aggregated at the estuary level to determine the mean length of each species within each estuary. The MaxN for each species within each replicate BRUV drop was converted to biomass using W = aLb, where W is the average weight of the species, L is the mean length determined above, and a and b are constants derived from Fishbase (Froese 2019) or published literature where possible (Supplementary Table S1). The average weight was then multiplied by the MaxN value to calculate the relative biomass of each species within each BRUV deployment. Functional feeding group biomass was then calculated by summing the biomass of species within each group. The overall functional feeding biomass for each variable (Estuary, Habitat, Seasons) was taken as an average from three replicate BRUV deployments.
Multidimensional scaling (MDS) plots based upon the Bray-Curtis similarity matrix (Clarke and Gorley 2006) were generated to depict patterns of functional feeding group biomass among the three habitats, two seasons and two estuaries. All estuarine data was fourth root transformed to reduce the influence of rare or highly abundant functional feeding groups (Anderson et al. 2008). Where too few unique permutations existed for a reasonable test to be run, Monte Carlo random draws were applied (Anderson and Robinson 2003). Principle coordinate ordination (PCO) was conducted on a matrix of functional feeding group biomass to determine the major groups driving dissimilarities between the three estuarine habitats. Permutational Analysis of Variance (PERMANOVA) was then run on the functional feeding group biomass similarity data comparing between Estuaries (2 levels: Botany Bay, Lake Macquarie), among Habitats (3 levels: Artificial Reef, Natural Reef, Seagrass), and between Seasons (2 levels: Cool, Warm), with all factors considered fixed and orthogonal.
Coastal Fish Assemblage Composition
Species composition and MaxN data for Port Stephens coastal rocky reefs was sourced from Harasti et al. (2018). A subset of this dataset, from Broughton (32.62°S, 152.32°E) and Fingal Islands (32.75°S, 152.19°E; Fig. 1) during 2015 and 2016, was used for both cool (July-September) and warm (February-April) seasonality. The reefs were within depths of 20–35 metres, and were analysed with the same BRUV methodology previously described for the estuarine component.
MaxN was averaged across the 8 replicate BRUV deployments at each site during each Season. PERMANOVA was conducted on the Site (2 levels: Broughton Island, Fingal Island), Season (2 levels: Cool, Warm) and Year (2 levels: 2015, 2016), with all factors considered fixed and orthogonal. This analysis allowed the examination of possible among year-effects on fish community composition. Species abundance data was converted to biomass using the same methods as used with the estuarine dataset, with species lengths measured in EventMeasure. Because Harasti et al. (2018) only measured lengths of recreationally and commercially important species, additional length data for all other species was obtained by reanalysing videos from selected deployments which were known to contain high abundances of species previously not measured. In a few cases, the Port Stephens BRUV footage was too poor to calculate reliable length data (e.g. poor visibility or complete length of fish obscured due to other fish or camera angle), so species lengths were taken from a nearby study (Truong et al. 2017).
Species from the Port Stephens dataset were allocated to functional feeding groups as per the estuarine habitats (Supplementary Table S1). The biomass of the species was summed across the functional feeding groups to calculate a group biomass. These were then converted to relative proportional biomass of each functional feeding group, per Site and Season, across the fish community, to be comparable with the estuarine data.
The mean energy source results from Truong et al. (2017) were included as an additional example of a typical nearby coastal reef location, to provide more context for the patterns observed at other locations. These results were derived from 14 sites collected by the Reef Life Survey (RLS; Edgar et al. 2020) and three additional sites surveyed using visual census and remote underwater video (RUV).
Baited versus Unbaited Methods in Coastal Habitats
As some functional feeding groups, such as herbivores, may be under-represented using BRUV, underwater visual census data was also obtained from the RLS program for the Port Stephens coastal rocky reefs to explore how survey method influenced the observed fish assemblage composition. Three RLS survey sites closest to the BRUV survey positions around Fingal Island were used: Fingal Island North East (32.74°S, 152.20°E), Fingal Island Sanctuary (32.75°S, 152.19°E) and Fingal Sponges Fingal Island (32.74°S, 152.21°E) (Supplementary Figure S2).
Species counts from RLS surveys were averaged across the two replicate surveys conducted in each site during both warmer (November-April) and cooler (June-September) seasons, and fish lengths were converted to biomass. The underwater visual census estimates were then converted to relative functional feeding group biomass and compared to the coastal BRUV data with ANOVA with Tukey post-hoc tests used to determine the direction significant effects. Finally, the BRUV methods used by Harasti et al. (2018) at Port Stephens, the RUV methods used by Truong et al. (2017) in Sydney and the RLS data sourced from both locations were also compared using ANOVA. The Sydney data also contained diver surveys adopting a similar method to RLS at each of the RUV locations.
Food Web Analysis and Energy Sources
A diet matrix, developed by Truong et al. (2017) was used to define local food webs and identify basal energy sources supporting fish biomass (see Supplementary Table S2). The matrix identified links between predators and prey connected through fish and non-fish functional feeding groups to the basal energy sources: phytoplankton, macrophytes and detritus. For some analyses, zooplankton was used a basal food source in place of phytoplankton due to the direct links between zooplankton and higher trophic groups. Additionally, zooplankton are the main consumers of phytoplankton, meaning the basal support provided by phytoplankton to the fish assemblage would only be marginally higher than zooplankton. This single diet matrix was used to define the food web for all surveyed habitats, meaning that any difference in calculated energy sources between systems and habitats is due solely to differences in species composition. For example, if a coastal site has a greater proportion of zooplanktivorous fish than an estuarine site, zooplankton will support a greater proportion of the fish assemblage. Deriving diet matrices for each system or habitat would probably lead to more accurate basal energy source estimates, but this would require extensive high-resolution diet data which is typically not available. Estuarine visual surveys would improve the confidence in using BRUV methods for food web analysis, but all such visual methods are limited by daylight and water visibility.
Once the diet matrix was specified, the importance of the trophic pathways to the fish assemblage was measured by weighting these pathways using the observed relative biomass of the functional feeding groups (Truong et al. 2017). Firstly, the basal energy sources were estimated for each functional feeding group by summing the product of every possible unique pathway between the fish group and a basal energy source. Secondly, the group-specific estimates were multiplied by the relative biomass of each group and then summed to create an assemblage-wide estimate of support for each basal energy source. See Supplementary Figure S1 for a worked example of estimating the proportion of basal food sources supporting a functional feeding group. There are typically many pathways in a food web (hundreds of thousands), so these calculations were automated using the algorithm of Truong et al. (2017), run in R Version 4.0.3 (R Core Team 2020).
Sensitivity Analysis
Fish diets are inherently variable due to food availability, seasonality, and other ecological processes (Becker and Laurenson 2007). To explore the influence of this variability and uncertainty on basal energy source calculations, we conducted a sensitivity analysis which manipulated the proportion of prey items consumed by each functional feeding group. The dietary proportion of individual prey items were varied by \(\pm 10\%\), with energy sources re-calculated each time. All of the other dietary items for the targeted functional feeding group were adjusted to balance the relative diet matrix. This was repeated 50 times per functional feeding group, with the prey item and 10% increase or decrease selected at random. This provides a mean value for each consumer group and energy source, representing the influence of variation in that consumer’s diet on the estimated energy sources values. The complete output of this analysis is reported in Supplementary Table S3 for each of the eight survey sites and habitats. All analyses in this study were done using Primer v6+ (Anderson et al. 2008) or R (R Core Team 2020).