Scraping and grazing herbivorous/detritivorous fish display opposite feeding behaviours under different protection regulations

ABSTRACT Herbivorous and detritivorous fish display complex feeding behaviour, and research into how feeding behaviour changes with environmental and social variables is lacking. Such knowledge is imperative to infer how herbivory/detritivory will differ, in light of shifting resources and communities and specifically whether reefs can recover from disturbance. Fish abundance, feeding rate, body size, diet and schooling feeding of three major functional groups (scrapers, grazers and browsers) were examined across reef types under different fishing regulations. Scrapers and grazers (parrotfishes and surgeonfishes) were more abundant and displayed the highest feeding rates on reef flats. Scrapers mainly resided inside the restricted zone, while more grazers were found in the general use zone, where macroalgal abundance was highest. Browsers (rabbitfishes) were seldom observed and patchily gathered on the reef flat and reef slope in both zones. Thus, fishing protection did not appear to benefit grazers and browsers, whereas more scrapers gathered on shallow reef flats in the protection zone. Scraper and grazer feeding rates increased from an individual to a pair and increased with body size, these factors led to variations in feeding behaviours across reef types and protection regulations. Protection appears to benefit scrapers and variations in feeding rates were largely related to school size. Lastly, grazer density was associated more with macroalgal coverage than protection status. The opposite feeding behaviours of scrapers and grazers indicates not only protection status, but fishing restrictions and size limit regulations are needed to maintain coral reef fisheries and functional diversity on coral reefs in Taiwan.


Introduction
Coral reefs are changing as a direct consequence of anthropogenic activities, leading to widespread coral decline, loss of structural complexity and the decline of many functionally important species (Pratchett et al. 2011;Harborne et al. 2017;Bellwood et al. 2019). Certain functional groups have been recognized as having major roles in maintaining reef health and resilience, particularly herbivorous and detritivorous fishes (Burkepile and Hay 2008;Rasher et al. 2013;Clements et al. 2016;Eggertsen et al. 2019). Herbivorous fishes can generally be divided into bioeroders, grazers, scrapers, territorial gardeners and browsers based on their functional feeding (Green and Bellwood 2009;Hoey 2010;Rasher et al. 2013;Graham et al. 2015;Clements et al. 2016). These functional groups can control algal abundance, mediating competition between algae and corals for colonizing space (Bellwood et al. 2004;Ledlie et al. 2007;Cheal et al. 2010). Mediating such competition helps to maintain a coral-dominated state, preventing coral-algal phase shifts (Burkepile and Hay 2008).
The capacity of the herbivore assemblage to control algae is mediated by feeding behaviour. Herbivorous reef fishes display complex feeding behaviour that can vary between species (Bonaldo et al. 2014;Nash et al. 2016;Tebbett et al. 2017). Resource availability is a major driver for foraging behaviour on algal-dominated reefs as foraging behaviour is spread over large distances, whereas on healthy coral-dominated reefs this is not the case (Williams and Polunin 2001). Species composition may also affect foraging behaviour due to social facilitation leading to increased feeding rates (Nunes et al. 2013), competition (Robertson and Gaines 1986) and decreased predation risk (Madin et al. 2010a). Foraging movements vary between reefs with different benthic compositions (Nash et al. 2012); the relationship between functional groups and benthic communities is not yet clear, although research is emerging (see Robinson et al. 2019).
Given the increasing frequency and intensity of disturbances, there is a clear need to better understand herbivory and detritivory across space and time. Marine protected areas (MPAs) have been utilized to protect and conserve important reef species. Effectively managed MPAs have been shown to increase the density, body size and biomass of targeted fish species in certain regions (Cheal et al. 2010;Kopp et al. 2010;Russ et al. 2018). Although the management of certain herbivorous fish species and functional groups has been used in attempts to improve reef resilience, the protection of other herbivorous fishes, especially parrotfish (Labridae: Scarinae), has had little effect on reef resilience (reviewed in Bruno et al. 2019). Furthermore, there is growing evidence that the restoration of predatory fish communities can reduce grazing and other herbivorous feeding and alter foraging behaviour (Madin et al. 2010b;Houk and Musburger 2013). Herbivorous fish stocks cannot be managed without addressing feeding behaviour, behavioural changes and the wider ecological implications of such behaviour.
There is a growing body of literature on the drivers of the herbivore community and behaviour, with few studies exploring the relationships between the fish community, habitat-level drivers and behaviour (Nash et al. 2016;Yarlett et al. 2020). Nash et al. (2016) found the foraging movements of parrotfishes were influenced by the benthos, and densities were driven by management status. Most parrotfishes, surgeonfishes (Acanthuridae) and rabbitfishes (Siganidae) feed over relatively small areas and often form mixed species feeding schools (Welsh and Bellwood 2012b;Carlson et al. 2017). This schooling behaviour in roving herbivores can enable individuals to circumvent the territoriality of their competitors (Ogden and Buckman 1973;Robertson et al. 1976;Foster 1985) and provide protection against predators (Brandl and Bellwood 2014). There is still a gap in the literature describing the functional impact of herbivores across the protection gradient. This research into herbivore schooling behaviour, and the relationship between herbivory and the benthos and protection status is crucial for effective management. Regulations considering specific species, behaviour and functional roles can be implemented to maximize herbivory on reefs by highlighting which characteristics of foraging behaviour may be influenced by management (Nash et al. 2016). Therefore, this paper aims to explore how herbivorous/detritivorous fish associate with the benthos, while examining abundance, feeding behaviour and differences between functional groups, protection status and reef types. Here, for scrapers, grazers and browsers we investigated: (1) how management status influences fish abundance and feeding rates; (2) how feeding rates change in response to differences in fish communities, reef types and macroalgal abundance; (3) how schooling behaviour influences feeding rates; and (4) differences between fish size and gut contents across management status and reef types. We offer these findings for the basis of conserving functional groups in Taiwan. As many protected areas fail to address functional groups or feeding behaviour, there is a clear need to better understand such questions and incorporate these findings to support herbivory in further management efforts.

Study sites
Kenting National Park (KNP) is Taiwan's first national park established in 1982, located on the Hengchun Peninsula in southern Taiwan (Figure 1). The coastal region, locally known as Nanwan Bay, supports one of the most diverse coral reefs in Taiwan (Meng et al. 2008;Keshavmurthy et al. 2019). The rich and diversified marine ecology attracts many local and foreign tourists annually (Lee et al. 2019). Corals in the region have suffered from six severe typhoons and two major bleaching events from 1985 to 2010, however there are signs of recovery (Kuo et al. 2012;Keshavmurthy et al. 2019). Cold-water disturbance during spring tides, typhoons, coral disease and overfishing all further damage the reefs, causing this region to be one of the five most threatened areas in Southeast Asia (Chen and Dai 2004;Meng et al. 2008). The extent to which these pressures will affect reef health in the absence of better management remains unclear.
KNP was designed to limit fishing activities and certain fishing gear. However, with illegal fishing and tourism activities increasing, KNP collaborated with the local tourism industry and announced Houbihu (Leidashi and Huayuanqu) as a marine resources protection demonstration area (MRPDA) in 2005. The MRPDA is a restricted no-take area, and stakeholders and tourists tend to report illegal activities to law enforcement (Yang et al. 2013). Higher fish diversity and large predators are often observed in the MRPDA (Jeng et al. 2015). Illegal angling and spear fishing are often observed in general use zones in KNP Wen et al. 2019).
This study was conducted between July and September 2015 in KNP (21°56 ′ N, 120°47 ′ E). To examine the influence of protection status on the function and behaviour of herbivorous/detritivorous fishes, we surveyed two sites in the general use zone and two sites in the no-take MRPDA (Figure 1). Wanlitong and Hongchaikeng are in the general use zone and open to regulated coastal angling and subject to more illegal fishing activities (i.e. drift nets and gill nets). Fish abundance and diversity are lower in the general use zone (Wen, unpubl. data). Chu-Shui-Kou and Hou-Hua-Yuan are in the MRPDA restricted zone, where the local tourism industry has acted as a regulatory body for illegal activities. Although the study sites are situated on different sides of the Maobitou Cape, these sites have similar environmental characteristics (i.e. wave action and currents). The main monsoons are northeasterly, meaning the Houbihu is sheltered by mountains. The fish communities at Houbihu, Wanlitong and Hongchaikeng are similar (Wen, unpubl. data).
We divided these fishes into three functional groups: scrapers, grazers and browsers (Bellwood et al. 2004;Green and Bellwood 2009;Rasher et al. 2013). Scrapers/small excavators scrape the algae on the surface of the reef consuming algal and detrital material while removing some components of the substratum, such as Hipposcarus spp. and Scarus spp. Although Chlorurus spp. are identified as large excavators/bioeroders (Bellwood and Choat 1990), we did not record any large Chlorurus spp., therefore we defined Scarus spp. and Chlorurus spp. as scrapers. Grazers intensively graze on epilithic algal turf/detritus; Acanthurus spp. and Ctenochaetus spp. (excluding planktivores) were classified as grazers. Browsers feed directly on large established macroalgae and S. fuscescens was the only browser recorded. Choat et al. (2004) indicated some grazers and browsers consume detrital and sedimentary materials, meaning not all individuals were classified as herbivores. Previous research has shown species composition varies with depth and reef type (Hoey and Bellwood 2008). Although depth and reef type are confounding, we chose three different reef types to investigate feeding behaviour, consisting of the reef flat (1-5 m), reef crest (5-10 m) and reef slope (10-15 m).

Fish abundance, fish size and feeding rates
The abundance and body size (cm, total length, TL) of scrapers, grazers and browsers were documented using three belt transects, among three reef types at two sites in the general use zone and two sites in the restricted zone. The start of each transect was randomly selected then scuba divers used UVC to record fishes in 20 m × 5 m (100 m 2 ) transects. The transect tapes were laid during the survey to reduce diver effects (Emslie et al. 2018). Feeding rate (bites/ minute), fish size and target macroalgae were examined using stationary point counts (Samoilys and Carlos 2000). Three divers conducted stationary point counts (3 m radius), where individual fishes were arbitrarily selected and feeding rate was recorded for 3 min. If the subject left the area before 3 min, data were excluded from further analysis. Divers recorded all fishes for the entire duration of the dives. To determine whether other variables affected feeding behaviour, factors such as time of day, schooling size and benthic composition were considered. We conducted surveys at three time periods to investigate the effects of the time of day ( Figure S1). Data were collected from the morning (8am-10am), noon (11am-1pm) and afternoon (3pm-5pm). There was no significant correlation between the feeding rates of scrapers and grazers with time periods (p > 0.05). Data collected throughout the day were combined to analyse the effects of fish abundance and reef types on feeding rates. Divers also conducted roving surveys on the reef flats to determine whether schooling behaviour affects feeding rates. Schooling groups were arbitrarily selected and group size (species and number of individuals in each group) and feeding rate (bites/ minute for individuals in the group) were recorded for 3 min. If any subjects left the area before 3 min, data were excluded from further analysis.

Macroalgal coverage
Benthic composition data were collected using the same transects mentioned above. A quadrat frame (30 × 30 cm) with an underwater camera attached 50 cm above the quadrat was used to take photographs at 1 m intervals along the transects. The 20 photographs from each transect were analysed using Coral Point Count with Excel extensions (CPCe). CPCe was used as it quickly and efficiently calculates statistical coverage over a specified area (Kohler and Gill 2006). A stratified random method was used within CPCe to apply 30 points to each frame. Benthic categories were identified to morphological functional groups (i.e. scleractinian coral, crustose coralline algae and macroalgae). Data were then converted using CPCe to provide percentages of major benthic categories.

Gut contents
Parrotfishes, surgeonfishes and rabbitfish were collected along the coast of the four sites from local anglers. After measuring weight (g) and fish size (cm, TL), the digestive tract was removed, gut contents extracted and refrigerated to slow digestion. To increase the number of samples, we collected specimens from the local fish market (only where suppliers could confirm catch locations in confidence). Gut contents were removed in situ and refrigerated. In the laboratory, the digestive tract was weighed (wet) with an electronic scale (minimum to 0.01 g). Gut contents were chosen at random, spread in Petri dishes and analysed using a microscope to identify macroalgae to the lowest taxonomic resolution.

Data analysis
Fish abundance and feeding rates N-mixture models (zero-inflated Poisson, zero-inflated negative binomial, Poisson and negative binomial) were used to examine the factors (protection status and reef type) influencing abundance and feeding rate. The N-mixture model approach accounts for the possible over-dispersion of data, common in ecological studies (Joseph et al. 2009). As some fishes were observed not feeding, Hurdle analysis (Poisson with logit link) was used to examine zero-bite data (i.e. observational periods where individuals did not bite) between factors, prior to the analysis of feeding rate and other factors ( Figure S2). Generalized linear model (GLM) and zero-inflated model (ZIM) with negative binomial and Poisson data distribution with multiple factors were used to create different models and to produce the Akaike information criterion (AIC) value for each model. Akaike information criterion with correction (AICc) values were compared with different mixing factors, then the Akaike weight was calculated, and the highest Akaike weight indicated the most parsimonious model (Richards 2008;Symonds and Moussalli 2011). The assumptions of models and fitness of the models were diagnosed by using residual plots and the influence of these outliers was explored using Cook's distance ( Figure S3).

Fish abundance and macroalgal coverage
Correlations between macroalgal coverage and herbivorous fish abundance were investigated using a GLM with negative binomial and Poisson distribution. The effects of herbivore abundance, protection status and reef type were tested on macroalgal coverage at the study sites. AICc was used to indicate the best explanatory factor, due to the small sample sizes. Best goodness-of-fit model indicated the potential drivers (herbivore abundance, protection status and reef type) of macroalgal coverage. Models were created using R (v3.1.0; R Development Core Team 2014) with the MASS, pscl and MuMIn packages.

Schooling feeding behaviour
We first used Hurdle analysis to examine if the zerobite feeding behaviour was related to solitary or schooling feeding among functional groups and life stages. Then we used a Bayesian model to build a probabilistic exponential growth curve to illustrate the correlation between increasing feeding rate and group size. The model was built in Python (Python Software Foundation 2010) using the PyMC3 module (Salvatier et al. 2016). We used data-driven gradientbased Markov chain Monte Carlo (MCMC) sampling algorithm, No-U-Turn Sampler (NUTS; Hoffman and Gelman 2014) in PyMC3 and Gaussian function to build the Bayesian model. We hypothesized feeding rate increased with group size from a Bayesian GLM model with normal (Gaussian) distribution, based on our preliminary observations and literature (Foster 1985;Overholtzer and Motta 2000). Exponential power regression was used to find the exponential function that fits best for increasing group size in our prior model. Models were fitted with observed bite rate and group size data. Likelihood values were used to estimate 95% confidence intervals.

Fish size and gut contents
Potential relationships between fish size (log-transformed TL) and gut content weight were analysed using GLM with AIC selection. The assumptions of above GLMs and fitness of the models were also diagnosed following the previous feeding rate examination ( Figure S4). GLM and AIC selection were processed using the glm and step function with the MASS package in R. The datasets generated and analysed during the current study are available in the Figshare repository, https://doi.org/10.6084/m9. figshare.12362726.v2.

Fish abundance
The highest abundance of scrapers (n = 122) was recorded on the reef flat in the restricted zone ( Figure 2a; Table I). Grazers displayed a significantly higher abundance (n = 91) on the reef flat in the general use zone (Figure 2b; Table I). Browser abundance (n = 18) was influenced by protection status and reef type, however a large amount of zerocensus data resulted in less statistical power for this analysis (Figure 2c; Table I). Scraper abundance was strongly associated with reef type, and both scraper and grazer abundance decreased from the reef flat to reef slope (Figure 2a, b). Juvenile scrapers and grazers were also found gathering on the reef flat rather than other reef types, in both restricted and general use zones ( Figure S5).

Feeding rates
We examined the feeding rates of scrapers (n = 141), grazers (n = 182) and browsers (n = 11). The zero-bite data for scrapers and grazers among different factors were included in the Hurdle analysis ( Figure S2). A significant relationship was found between feeding rate and body size, as large individuals displayed more non-feeding than feeding behaviour (p < 0.05). Other factors such as protection status, reef type and life stage with grazers, and all factors with scrapers showed no significant difference between zero and non-zero data (p > 0.5). Therefore, we used a generalized additive model (GAM) with non-zero data to show the correlation between feeding rate and body size for scrapers and grazers between zones (Figure 3). Bite rate and body size were positively correlated for both scrapers (Figure 3a) and grazers (Figure 3c, d).
The model selection with lowest AIC value suggested protection status influenced scraper and grazer feeding rates. Adult scrapers had higher feeding  rates in the restricted zones (Figure 4a), while grazers fed more in the general use zones (Figure 4b).
The number of recorded herbivores differed among reef types and protection status, with limited individuals recorded at certain sites ( Figure 4). Feeding rates of subadult/adult and juvenile fish across different restricted zones and reef types show similar patterns as the above GAM (Figure 3, 4). Adult scraper feeding rates were relatively uniform across reef types in the restricted zone, and the highest feeding rate was on the reef flat and crest in the protection zone ( Figure  4a). Juvenile scraper feeding rates were highest on the reef flat of the general use zone. Adult grazer feeding rates were influenced by both protection status and reef type, opposite to scrapers. Adult grazer feeding rates were higher on the reef flat and crest of the general use zone and crest of the restricted zone ( Figure 4b; Table SI). Lastly, juvenile grazers had the highest feeding rates in the general use zone.

Fish abundance and macroalgal coverage
Functional groups were analysed to investigate protection status, reef type and benthic composition as drivers of abundance. Results from the GLM indicated macroalgal coverage was mainly driven by reef type, and the total herbivorous fish abundance was strongly associated with high macroalgal coverage (Table SII). The general use zone hosted the highest macroalgal coverage (Figure 5b), which was positively correlated with grazer abundance. Although we did not find a clear negative correlation between coral and macroalgal coverage, the highest coral coverage was found alongside the lowest macroalgal coverage, on the reef slope in the restricted zone (Figure 5a).

Schooling feeding behaviour
We observed more scrapers in the restricted zone and more grazers in the general use zone. Scrapers and grazers exercised the highest feeding rates where we recorded the highest abundance (Figure 2, 4). The above results implied schooling behaviour may increase feeding rate. Therefore, we modelled abundance and feeding rate of schooling fishes (with zero-bite data removed). Initial hypotheses were a slope regression (schooling fish increase their feeding rate) and a regression reaching a plateau (feeding rate did not increase with group size). The results suggested a high probability of a non-linear regression showing scraper feeding rates were positively correlated with schooling behaviour (Figure 6). The turning point was around 2, indicating scrapers show an increase in feeding rate from a single individual to a pair, then no further significant increase after a pair.

Fish size and gut contents
The body size of scrapers and grazers was examined across protection status and reef types using a GLM (Table SIII). The largest individuals of scrapers were found in the restricted zone and the largest grazers in the general use zone ( Figure S6). In addition, while the scrapers on the reef flat in the general use zone were much smaller than elsewhere, grazers had relatively similar size distribution among reef types for both zones. Simple linear regression was used to study the relationship between fish size and gut contents. Scraper size and browser size were both positively correlated with gut content weight (Figure 7a, 7c), whereas no relationship was found with grazer size (Figure 7b). The consumed algae belonged to the genera Gracilaria, Dictyopteris, Laurencia, Ulva, Sargassum and Dictyota ( Figure S7). The majority of the consumed algae types were most dominant on the reef flat in the general use zone. Less common algae types found in the gut contents (including turf algae) were more abundant in the restricted zone.

Discussion
Understanding fish density, body size and foraging behaviour is vital for assessing the contribution of herbivorous and detritivorous fishes to coral reef ecosystem function. Research investigating specific variabilities of herbivory on reefs is now relatively abundant (Choat and Clements 1993;Russ 2003;Fox and Bellwood 2007;Bejarano et al. 2017;Eggertsen et al. 2019;Streit et al. 2019). However, research focusing on which characteristics of foraging behaviour, or the community overall, may be influenced by management is lacking (Nash et al. 2016). Such knowledge can improve management efforts by implementing strategies that promote feeding behaviour or drive herbivore densities. In this study, the highest abundance of scrapers and grazers and the highest feeding rate were found on the reef flats. This may well be due to the increased macroalgal abundance on reef flats compared with other reef types (Russ 2003). This pattern in algal abundance on the reef flat may be due to light availability, which may be why herbivorous fishes congregate here and subject the reef to higher levels of grazing (Fox and Bellwood 2007).
We found scraper and grazer abundance was associated with macroalgal coverage, however there are many factors affecting reef fish densities and macroalgal distributions. Jessen and Wild (2013) found few herbivorous fish species on a flat reef in the Red Sea, when compared with studies at greater depths and in greater study areas, indicating larger study areas and greater depths may host more feeding opportunities and refugia. We found more scrapers in the restricted zone with low turf algae cover, and more grazers in the general use zone with high turf algae cover. Hoey and Bellwood (2011) have reported grazers, scrapers and browsers tend to avoid high abundances of macroalgae (Sargassum) and focus on medium-and low-density patches. This may be the reason why grazers in our study gathered at the general use zone hosting 20% macroalgal coverage, which is closely related to the medium-density algal coverage found by Hoey and Bellwood (2011), although comparisons of specific algae types are needed to confirm this. Although we did not record damselfishes (Pomacentridae), we noticed more territorial damselfishes on the reef crest and slope in the general use zone. A high abundance of territorial damselfishes will likely affect feeding behaviour of large roving herbivores (Foster 1985). Scrapers in this study were shown to be the main driver for top-down regulation on coral reefs, as found in other literature (Smith et al. 2010). The prevalence of top-down or bottom-up regulation on coral reefs in Taiwan requires further research.
It appears that scrapers benefit from protection status whereas grazers were more influenced by benthic composition, similar to the findings of Robinson et al.  research into the foraging movements of two parrotfish species, and found movements were largely influenced by benthic composition. Densities were associated with management status, and grazing rate was influenced by management status and species. This study partially supports our main findings, highlighting that scraper densities were strongly associated with protection status. Our findings suggest scraper abundance is reducing macroalgal coverage in the restricted zones. This may not be the case, as Russ et al. (2015) found compelling evidence that benthic composition exerted strong bottom-up control on parrotfish (both scrapers and excavators) density. Furthermore, Taylor et al. (2019) provided additional evidence for this nuanced ecological feedback systemone where disturbance plays a key role in mediating parrotfish-benthos interactions. By influencing the biology of assemblages, disturbance can thereby stimulate change in parrotfish grazing intensity. This suggests benthic composition (which may be initially driven by disturbance) drives parrotfish density and behaviour and not the other way around. Russ et al. (2018) investigated the decadal-scale response of surgeonfishes to no-take marine reserves and changes in benthic composition and found density of surgeonfish species was influenced more by changes in benthic composition than marine protection. This is possibly why we observed more grazers in the general use zone, despite this also being the area subject to fishing. This highlights the greater importance of bottom-up control than top-down control for surgeonfishes.
Other coral reef herbivores such as sea urchins may play an important role in driving macroalgal abundance. Sea urchins have a small habitat range (scale of m 2 ), and when herbivorous fish density decreases (e.g. due to overfishing), sea urchins still create patchy areas of low algal biomass around individuals (Nozawa et al. 2020). Although all the scrapers in our study were parrotfish species, analyses were based on pooled bite rates; further comparisons of species level bite rate data are required to investigate this further. There is a great deal of variation in feeding behaviour within reef fish families Manning et al. 2019). To expand on this, with species-specific analyses we would probably find individual species preferring different benthic types. For instance, four species in the genus Ctenochaetus have a distinct dentition adapted for detritivory primarily associated with rubble (see Russ et al. 2018).
Our results show that the time of day did not affect coral reef fish feeding, although these findings do not reflect the general consensus on this. A strong influence of the time of day on herbivorous fish feeding has been shown across many species in many regions, displaying a pattern of algal nutrients increasing to a midday peak and remaining constant through the afternoon Zemke-White et al. 2002;Yarlett et al. 2018).
Although grazer abundance was highest on the reef flat, there was no clear difference in feeding rates on the reef flat and crest. We speculate the low feeding rates of grazers on the reef flat in the general use zone was a result of anthropogenic interference. Huang et al. (2017) found reef fish abundance and diversity were significantly reduced in the presence of many tourists without management. The reef flats of the general use zones in KNP are subject to snorkelling, swimming, scuba diving and fishing. Although fish generally congregate on reef flats, with so much external disturbance, fish are less likely to display normal behaviour, perhaps leading to reef crest and slope feeding.
Certain reef fish species may prefer the reef crests, but generally the energetic costs of living on the reef crest are greater due to higher hydrodynamic activity. Traditionally recognized herbivorous functional groups differ in swimming performance, and in their capacity to feed consistently across levels of wave exposure. Species within the same feeding functional group are known to have contrasting responses to wave exposure, emphasizing the distinctness of their ecological niche and functional complementarity (Bejarano et al. 2017). Our sampling sites were located on different sides of Maobitou Cape, which face different monsoon influences between summer and winter. Although we conducted the study between monsoon seasons with no clear wave impact, the long-term influence of wave exposure may contribute to different behaviours of herbivorous and detritivorous fishes between sites. Due to our categorization of functional groups, and not conducting species-level analyses, it is difficult to investigate how hydrodynamic activity may influence our findings. A change in experimental design is required to incorporate differences in swimming performance and how these differences affect feeding behaviour.
Scraper and grazer feeding rates were correlated with body size, but dependent on protection status. Scraper feeding rates were correlated with body size in the restricted zone, while grazer feeding rates were correlated with body size in the general use zone. Large fish inside protected areas are known to exhibit fewer but more efficient bites (Kopp et al. 2010). Large parrotfish (terminal phase) species usually spend more time patrolling feeding and breeding territories, explaining why large individuals exhibit slower feeding rates than juveniles (initial phase; Bonaldo et al. 2006). Our results contrasting with the above research may be due to differences in overall fish assemblages. Most individuals observed on the sites in this study with high abundance also exhibited schooling behaviour. Furthermore, research is emerging showing how relationships between feeding rate and size can also be different based on seasonality (Satterfield and Steele 2019).
Several herbivorous reef fish species have been described moving in large schools (Ogden and Buckman 1973;Robertson et al. 1976;Lawson et al. 1999;Choat and Bellwood 1985;Fox and Bellwood 2007). We found that feeding rates increased with school size, but our model revealed that feeding rates increased from a single fish to paired feeding, then remained constant as school size increased. Welsh and Bellwood (2012a) found the average feeding rate in a school was greater than that of a single individual. They highlighted four rabbitfish species that displayed increases of 50-100% bite rates while feeding in pairs. Other studies have found the larger the group size the higher the feeding rate (Kopp et al. 2010). Foster (1985) found individual bite rate in surgeonfish was positively correlated with group size because individuals in the group suffered attacks from territorial fish less frequently. Schooling reduces predation risk, provides an increase in relative foraging time and allows access to resources which would otherwise be unavailable (Wolf 1987;Brandl and Bellwood 2014).
The gut content weight of scrapers was consistent with their body size, comparable to Elliott and Persson (1978), who suggested fish gut contents linearly increase with weight. While grazer body size and gut content weight were not correlated, it was possibly affected by the range of data collected. The anglers who sourced many of our samples mainly targeted grazers between 15-20 cm TL, unfortunately heavily influencing regression analyses. The gut content analysis showed species belonging to the genera Gracilaria, Dictyopteris, Laurencia, Ulva, Sargassum and Dictyota. Consumed macroalgae were consistent with other findings documenting herbivorous fish diet (Tolentino-Pablico et al. 2008). Although contrarily, research has shown herbivorous fish avoided brown macroalgae (i.e. Padina) instead favouring turf algae on the Great Barrier Reef (Bellwood and Choat 1990) and in the Caribbean (Kopp et al. 2010). Most parrotfishes are microphages that target cyanobacteria and other protein-rich autotrophic microorganisms . We acknowledge, without the aid of biochemical analyses of diet, more in-depth investigation into dietary choices poses problems.
Although factors affecting food choices in herbivorous fishes in Taiwan are largely unknown, the authors predict a key factor to be food availability. Food availability is likely to affect feeding behaviour as Gracilaria spp. and Laurencia spp. are abundant in southern Taiwan (Tsai et al. 2005). To further investigate food choices, the analysis of functional roles and algal removal rates for specific species would provide more insight. For instance, Tebbett et al. (2017) compared the gut contents of two surgeonfishes finding A. nigrofuscus predominantly ingests algae while Ctenochaetus striatus (Quoy and Gaimard, 1825) ingests detritus and sediment. In a dietary analysis conducted by Choat et al. (2002), they found a wide range of dietary items were consumed, and dietary groupings did not reflect taxonomic relationships. Finally, Kelly et al. (2016) concluded all surveyed species preferred turf algae, but overall, there were variable foraging portfolios across species. Estimates of algal consumption, grazing rates, additional fish counts and gut content analyses employing a standardized method incorporating DNA analysis are required to further investigate food choices. This crucial knowledge of herbivorous fishes and their food items is important when designing and managing MPAs and would be the next step to continue the research presented in this paper.
This research could be the crucial first step investigating the role that functionally important detritivorous fish may play in the recovery of degraded reefs in southern Taiwan. The restricted zone of KNP is relatively small and management strategies for the protection of functional groups are essential to maintain diversity and coral reef structure. Scientists have long advocated for local interventions, such as creating MPAs of varying management levels and implementing fishery restrictions as methods to mitigate local stressors on reef-building corals. However, Bruno et al. (2019) found very little evidence to support the notion of managed reef resilience, primarily because the impacts of local stressors are often minor compared with the much greater effects of ocean warming. We propose these findings are used to inform fishing regulations, contribute towards protecting certain areas and inform decisions based on feeding behaviour and functional groups, thus maintaining coral reef health in KNP. If reefs in KNP cannot be managed properly and saved by local action alone, then we must face reef degradation by addressing anthropogenic climate change.