Understanding and predicting the distribution and abundance of organisms is a fundamental component of ecology (Legendre and Fortin, 1989). However, identifying the abiotic and biotic factors that alter species distribution and abundance is challenging because those factors change over space and time. Anthropogenic impacts also affect the abundance and distribution of species (Jackson et al., 2001; Halpern et al., 2008). In marine ecosystems, overfishing can cause substantial reductions of target and by-catch species (Worm et al., 2006), while global warming can alter the community assemblage of fishes (Fodrie et al., 2010). However, our ability to quantify and predict the effect of humans on ecosystems is limited because multiple impacts often occur simultaneously and at different scales (Crain et al., 2008). In addition, environmental variables may interact with anthropogenic stressors. For example, natural climate cycles may mitigate or exacerbate global warming (Nye et al., 2014). Thus, studies must collect information on the focal community and relevant natural and anthropogenic factors over a meaningful scale, which often means multiple decades and 100 s of kilometers, to quantify the separate and emergent effects of humans on community dynamics.

Our understanding of the relative influence of natural and anthropogenic factors on communities is often limited by the scale of data collection. This is particularly evident for marine temperate reefs, despite supporting many economically important species throughout the world (Coleman et al., 2000; Moore et al., 2011). Temperate reef (or hard bottom) communities include habitat characteristics that range from flat hard substrate with dispersed sessile organisms to highly rugose ledges and overhangs. Temperate reef habitats are patchy and often support much greater fish abundance than surrounding unstructured substrate (Fujita et al., 1996). Depth and habitat characteristics influence temperate reef fish communities (Moore et al., 2010; Whitfield et al., 2014). Studies of the effectiveness of protected areas on tropical reef fishes have found that habitat characteristics are as, if not more, important than fishing for the fish community (Wilson et al., 2010; Claudet et al., 2011). The relative influence of habitat and fishing are likely dependent on whether the fishes are targeted by fishers (Russ et al., 2015). However, the effect and importance of fishing and habitat relative to other variables (e.g., broad-scale environmental forcing variables), are relatively unknown over broad spatio-temporal scales. Improved understanding of the natural and anthropogenic factors that structure marine communities would facilitate the prediction of how those communities will respond to a changing environment and provide information to improve conservation and management efforts.

We used fishery-independent survey data collected over broad spatial (>50,000 km²) and temporal (24 years) scales to determine what variables affect the community structure of temperate reef-associated fishes along the Southeast United States coast. These temperate reefs support active commercial and recreational fisheries that target snapper and grouper species primarily using hook and line (Coleman et al., 1999). Recreational fishing in the early 2000’s along the Southeast United States harvested about 29,000 metric tons of fish per year (Figueira and Coleman, 2010), which was approximately 38% of the total catch with commercial composing 62% (Coleman et al., 2004). This study quantified the degree to which a suite of factors (e.g., depth, temperature, climate indices, and fishing) explained variability of fishes collected in a standardized trap survey. Habitat characteristics were included in an additional analysis for a subset of years in which those data were available (2011–2013). Finally, we quantified if specific fish traits were associated with changes in predictor variables. Results from this study provide a rare and valuable assessment of natural and anthropogenic factors that affect community structure over broad spatial and temporal scales.

The influence of the Gulf Stream was evident on mean winter SST, for which temperature increased with increasing distance from the coast (Figure 1B). Bottom temperature and SST during sampling was similar except for deeper sites (Figure 1C,D). The long-term data (1990–2013) included 9,645 trap samples, with an average of 402 trap samples per year (range of 201–1196). Sampling effort was slightly higher in the middle of the survey area (Figure 1E). The mean number of individuals caught in traps tended to be higher in shallow water (Figure 1F) and the number of species caught in a trap was often higher in shallow coastal waters in the northern region of the survey (Figure 1G).

The AMO was primarily in a positive phase since 1995, while the NAO changed from positive to negative multiple times during the study (Figure 2A,B). The SST through the 24 years of the study had a slightly negative trend, indicated by the slope, for the entire survey area and within the six 0.25 × 0.25° cells that were sampled every year (Table 1 and Figure 2C). The linear trend in winter SST and bottom temperature over the 24 years of the study was not different from zero over the entire study area and for the six cells sampled every year, except for winter SST, which had a negative trend for the six cells sampled every year (Table 1 and Figure 1D,E). The slope over 24 years for the number of individuals caught in a trap was not different from zero for the six areas sampled every year, but did decrease when all trap data were included (Table 1 and Figure 2F). The number of species caught per trap decreased over the 24 years of the study (Table 1 and Figure 2G). The regression analysis did not account for potential temporal autocorrelation of traps among years, which may have resulted in biased p-values and results with p-values near 0.05 should be interpreted with caution. Seventy-one of the 119 fish taxa (most identified to species level) collected were caught in more than five traps and were included in the multivariate analyses (Supplementary Appendix S5). Trap catch was primarily composed of three taxa, Centropristis striata, Haemulon aurolineatum, and Stenotomus spp.

We found that the number of individuals and species of fish caught in the traps decreased over the past 2 decades. These results, consistent with the findings of Bacheler and Smart (2016), suggest regional-scale declines in the abundance and diversity of fishes within the surveyed region. An increase in temperature, resulting from global warming, has been suggested as a potential driver of fish community change within this large marine ecosystem (Shertzer et al., 2009; Whitfield et al., 2014). However, we found no indication that annual water temperature has increased along the Southeast United States coast over the 24 years of the study. To assess what may be causing the potential observed declines, we conducted multivariate analyses on the fish community with relevant natural and anthropogenic predictor variables. Climate indices (AMO and NAO) and bottom and winter temperature explained the greatest amount of variation in the fish community structure, followed by longitude, recreational landings and commercial landings. These findings indicate the long-term changes in the fish assemblage are driven by local temperature and large-scale climate regimes. The degree of correlation between the fish community structure and recreational harvest suggests that recreational fishing may be affecting or tracking changes in the fish community of this large marine ecosystem as much as or more than other human-related impacts that are often cited as causing ecosystem degradation, such as commercial fishing (Worm et al., 2006; Shelton and Mangel, 2011).

Changes in large-scale climate patterns from both anthropogenic warming and natural cycles can affect marine communities. Global warming-driven temperature changes can cause shifts in the latitude, longitude, and depth of fish populations (Perry et al., 2005; Fodrie et al., 2010; Pinsky et al., 2013) and has altered community structure of ecosystems within a similar time period as this study (Hiddink and Ter Hofstede, 2008; Stuart-Smith et al., 2010). Our findings suggest that water temperature in the study region did not increase from 1990 to 2013, which is in agreement with global predictions of warming for this region (Burrows et al., 2011) and unlike other global warming studies that have found increased temperature over a similar time period, such as by ∼0.5°C on temperate reefs around Tasmania (Stuart-Smith et al., 2010) and by ∼1.9°C in the North Sea (Hiddink and Ter Hofstede, 2008). Climate indices such as the AMO are natural cycles that affect regional-scale temperatures and abiotic and biotic factors. For example, the AMO was a leading cause of changes in fish communities in the Gulf of Mexico (Karnauskas et al., 2015). The AMO shifted to a positive phase in the mid-1990s and has a cycle that lasts 65–70 years (Nye et al., 2014). Fish abundance and richness were in decline 5 years prior to the switch in the AMO, and the AMO explained the most variation in the fish assemblage of all the predictor variables included the long-term analysis. This result suggests that the observed fish community declines could be a result of decadal changes in climate oscillations which affect mixing and vertical gradients, as well as large scale circulation such as the Gulf stream (Frajka-Williams et al., 2017). Climate indices explained the most variation in the fish assemblage with a 3 year lag, which agrees with existing studies that climate indices alter growth and survival of early life stages, and these affects are not quantified until the fishes are large enough to be caught (Nye et al., 2009, 2014; Karnauskas et al., 2015). For the Southeast United States Atlantic coast, we did not find evidence that fish have experienced changes in water temperature as would be expected from global warming, but we did find that climate indices and temperature (bottom and winter surface) measures had the greatest influence on the fish assemblage. Winter SST was positively related to fish vulnerability, suggesting that larger, longer-lived taxa are more abundant during periods with warmer winters. Thus, changes in climate indices and temperature from both natural and anthropogenic drivers will significantly affect temperate reef fishes.

In contrast to the long-term analysis, changes in the short-term (3 years) fish assemblage was driven by depth. Previous research performed over smaller scales (over 1 year and approximately 10–1000 km²) found a comparable influence of depth in shaping fish communities (Moore et al., 2010, 2011; Fitzpatrick et al., 2012). Depth-related changes in reef fish community structure could be driven by biotic factors, including changes in predation (Jordan et al., 2012) or abundance of prey availability with depth, and abiotic factors including changes in depth-specific seasonal temperature (Whitfield et al., 2014). Depth was positively related with fish vulnerability, although marginally significant, suggesting that larger, longer-lived taxa are more abundant in deeper waters. The diminished influence of depth in the long-term analysis indicated that depth plays a secondary role compared to climactic- and temperature-related variables in affecting decadal-scale changes in the fish assemblage.

To gain a holistic understanding of the factors that effect the fish assemblage on temperate reefs we have included variables that have different spatial and temporal scales. For example, climate indices only changed through time (the same for all traps each year), while other variables such as depth and bottom temperature changed through both time and space. Although our results could be affected by these differences, our conclusions are likely robust regardless of the scale of the variable for the following reasons. First, the multivariate analysis bases significance on permutations that randomly shuffle samples to determine if the variable explains more variation than would be expected by chance. Second, the variables that explained the most variation were measured at both fine and broad scale (bottom temperature and NAO), suggesting that results were not dependent or inflated by scale alone. Third, variables that only changed through time explained minimal variation in the short-term analysis, which is expected given that there was little difference in these variables over 3 years. It is also worth noting that there are other variables that were not included which likely effect variation in fish community structure including but not limited to primary production, turbidity (Floeter et al., 2001; Karnauskas et al., 2015), and landscape mosaics (Moore et al., 2011).

The relatively limited amount of variance in reef fish community structure explained by habitat characteristics was surprising given that habitat characteristics are often related to abundance and diversity of reef fish (Lindberg et al., 2006; Schobernd and Sedberry, 2009). In the present study, the three habitat variables (percent hard substrate, substrate relief, and biota height) each explained less variance than most other variables. The minimal variance explained by habitat variables in this study may have resulted from a scale mismatch (Kendall et al., 2011), with fish responding to habitat characteristics over a broader spatial scale (e.g., hundreds of m²) than the scale at which habitat characteristics were assessed from the video data (tens of m²). However, other studies indicate that habitat characteristics play a minimal role in shaping the temperate, reef fish assemblage (Paxton et al., 2017).

Many studies have found that anthropogenic impacts, particularly commercial fishing, can alter marine food webs and cause regime shifts over similar time spans as this survey (Mollmann et al., 2009; Link et al., 2010). For instance, Link et al. (2010) found that commercial fishing had a greater impact than climatic indices on soft bottom and pelagic marine ecosystems within the Northern Hemisphere. Our findings indicate the opposite, with climate indices being more related to changes in fish community structure than fishery landings, and with recreational landings being slightly more correlated with community change than commercial landings. In general, the contribution of recreational fishing to stock declines and ecosystem alterations has been underappreciated compared to commercial fishing (Post et al., 2002; Cooke and Cowx, 2006), especially in the marine environment (Lewin et al., 2006). This is surprising given that recreational fishing can alter ecosystems in similar ways as commercial fishing, which includes altering age and size structure, habitat modifications, and trophic interactions (Cooke and Cowx, 2006; Lewin et al., 2006; Altieri et al., 2012). In addition, recreational landings exceed commercial landings on temperature reefs in areas around the globe, including the Gulf of Mexico (Coleman et al., 2004), the California coast (Schroeder and Love, 2002), and Australia (McPhee et al., 2002). To our knowledge this is the first large-scale marine study that combined both recreational and commercial landings with an ecological survey to compare the relative association of these two fisheries on fish community structure.

Our analysis was based on correlation and given the dynamic nature of fisheries (altering targets, effort and location to maximize catch) it cannot determine causation, but discussing potential mechanisms can facilitate future research. For the years covered in this study, total annual commercial landings were greater than recreational landings, but recreational landings were greater for a majority of the individual species (13 of the 21 species quantified in both fisheries; Supplementary Appendix S3). It is important to understand that our analysis only considered total annual catch of the individual fisheries and not the catch of individual species which was included in the analysis for the trap survey. Although two species caught by commercial fishers, M. phenax and C. nodosus, did have a negative relationship with this fishery there were fewer species with negative relationships with commercial compared to recreational harvest probably because the commercial fishery primarily targets snapper and grouper (Supplementary Appendix S3; Coleman et al., 1999) and these commercially target fishes do not make up a large percentage of the trap catch, with the exception of R. aurorubens. In addition, the greater correlation between recreational landings and changes in the fish community could result from recreational fishing targeting a broader range of species than the commercial fishery because commercial fishers have more incentive to target species with the greatest economic return. Finally, species caught in the survey had more positive relationships with recreational harvest compared to commercial harvest and the majority of these are not targeted by fishers. These species were primarily smaller species and omnivores who could benefit both from the reduced predation and competition of species that are targeted by harvesters. The findings of the multivariate analysis on fisheries likely result from the following coexistent factors, harvest reduces fish abundance (direct negative relationship), harvest is greatest in areas of highest fish abundance (direct positive relationship), and release of non-target fishes from predation and competition (indirect positive relationship). Although not significant, recreational harvest had the second greatest relationship with fish vulnerability. This may suggest that recreational fishing affects larger, longer-lived fish taxa that are trapped in the survey. Our findings highlight that overall fishery landings summed across all species may not strongly correlate with changes in communities (Arlinghaus and Cooke, 2005) and that recreational fishing could be affecting communities as much as commercial fishing, given that associations between commercial fishing and fishes has altered marine ecosystems (Mollmann et al., 2009; Link et al., 2010).

Data generated from large-scale, long-term ecological surveys can be particularly useful for elucidating which factors affect community structure, and the relative importance of those factors (Hughes et al., 2005). Our study provides the first quantification of the factors, both natural and anthropogenic, that affect the community structure of temperate reef fish over broad spatial and temporal scales. We found that long-term fish community structure in this large marine ecosystem is primarily driven by climate indices, temperature, and fishing. The importance of climate indices highlights the need for long-term data and inclusion of these variables when assessing changes in communities.

The study was conducted following NOAA NMFS animal use protocols.

NG, GK, and NB conceived the study. NG compiled and analyzed the data. NG wrote the initial draft. GK and NB had substantial input in editing the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.