Reefscapes of fear assess tradeoffs of risk and reward on coral reefs


 Degradation of coral reef habitats changes the abundance and community composition of fishes due in part to changes in the ecology of fear. The ecology of fear sees the predator-prey system as a dynamic game of behavioral responses to perceived risk with population and community level consequences. We measure spatial variation in predation risk as landscapes of fear. We consider changes in predation risk with habitat quality and examine the effects of fear on coral reefs in Kāne‘ohe Bay, O‘ahu, Hawai‘i. First, we associate fish and benthic communities on patch reefs with varying degradation due to invasive algae (Euchema spp. and Kappaphycus spp.). Next, we quantify the spatio-temporal variation of risk (reefscape of fear) of a common Hawaiian fish (saddle wrasse, hīnālea lau wili, Thalassoma duperrey) across reefs of varying degradation. Finally, we assess the tradeoffs in resource availability and predation risk on these reefs. At the scale of whole reefs, saddle wrasse responded to perceived risk. Intensity of patch use (measured by giving-up densities) by wrasse indicated risky reefs. Such reefs differed in benthic and fish community composition. We demonstrated the impact of an altered reefscape of fear due to habitat degradation. Habitat degradation seems to influence the tradeoff between resource availability and safety. From wrasse abundances and their patch use behavior we can classify the reefs into categories based on risk and resource availability. Allowing fish to reveal their perceptions of habitat qualities through their behaviors provides critical information for assessing and monitoring reefs.


Introduction
Habitat degradation is a major cause of biodiversity loss worldwide (Dudgeon et al. 2006;Baker et al. 2008; Newbold et al. 2015). Speci cally, coral reef habitats face critical declines from multiple sources of degradation including coral bleaching, invasive species, intense hurricanes, and destructive shing (Pratchett et al. 2014). Degradation of coral reef habitats reduces coral cover and structure (Feary et al. 2007; Graham and Nash 2013), which ultimately changes sh abundance and community composition of shes (Wilson et al. 2008;Pratchett et al. 2011;Graham et al. 2015). Fishes on reefs lacking structural complexity may experience higher predation rates due to decreased refuge availability (Almany 2004;Coker et al. 2009). While habitat degradation increases the lethal effects of predators on coral reefs, we are just beginning to understand how habitat degradation alters their non-lethal effects. These non-lethal or non-consumptive effects (NCEs) of predation may in uence the morphology, physiology, or behavior of shes living on coral reefs (Mitchell and Harborne 2020). Furthermore, habitat degradation may in uence the tradeoffs between resource availability and predation risk on coral reefs.
NCEs exert particularly strong effects in aquatic populations (Pressier et al. 2005). This may be due to altered prey behavior or habitat selection having a greater collective impact on community or population dynamics compared to the consumptive effect of predation alone ( Some known behavioral responses to predation risk include schooling (Rieucau et al. 2015), reduced foraging rates (Rasher et al. 2017;Madin 2010) and altered habitat use over space and time (Werner et al. 1983; Harvey and White 2017). These NCEs can be studied through the ecology of fear.
The ecology of fear views the predator-prey system as a dynamic game of behavioral responses that have population and community level consequences . Prey respond to acute and chronic variation in risk when predation risk is predictable and controllable (Creel 2018). However, there are tradeoffs associated with anti-predator defenses (Lima 1988). One way to minimize the cost of these defenses is to respond only when predation risk is high. Under the threat of predation, prey can spend time either avoiding risky places and activities, or by being more vigilant (Wirsing and Ripple 2011).
Foraging behavior should re ect both acute and chronic responses to predation risk. Predation risk includes properties of the landscape, and characteristics of both predator and the prey, such as: predator encounter rates, predator lethality, prey escape strategies, and the availability of refugia (Lima and Dill 1989). Prey perceptions of risk are revealed by giving-up densities at experimental food patches (Brown 1988;Brown and Kotler 2004 (Charnov 1976) and the giving-up density technique as a standardized assessment of the costs of predation risk associated with foraging across varying habitat types (Brown 1988;Brown and Kotler 2004). We consider how changes in habitat quality in uence the perceived predation risk by saddle wrasse on patch coral reefs in Kāne'ohe Bay, O'ahu, Hawai'i. This bay has a long history of coral reef degradation and recovery. Invasive carpeting algae were introduced into the bay in the 1970's and have colonized many of the bay's patch reefs (Bahr et al. 2015), leaving some reefs degraded and others algae free. Although successful conservation and restoration efforts reduced the overall abundance of invasive algae (Nelson et al. 2018), coral reefs may take decades to recover from a degraded state (Berumen and Pratchett 2006).
The goal of this study was to evaluate the tradeoffs of risk and reward on coral reefs. To do this we rst assessed sh and benthic communities on patch reefs. Next, we quanti ed the spatio-temporal variation of risk (the reefscape of fear) of a common Hawaiian coral reef sh across reef habitats. Finally, we assessed tradeoffs in resource availability and predation risk on these reefs. We expected benthic and sh communities to vary with differences in reef quality. We hypothesized that sh will perceive higher predation risk on reefs with reduced coral structure and refuge availability. Consequently, they should use resource patches less intensively and leave them at higher giving-up densities. We also hypothesized that perceived risk will vary within reefs, with greater foraging intensity and lower giving-up densities near coral refugia. Finally, we hypothesized differences in foraging behavior among patch reefs are due to tradeoffs in food availability and safety, which can serve as an integrative assessment of habitat quality from the perspective of inhabitants themselves. According to ideal free distribution and densitydependent habitat selection theory, we expect to nd higher sh density on reefs with higher resource productivity, and higher giving-up densities on reefs with higher predation risk. We propose that differences in perceived risk and resource availability between reefs will be revealed by the distribution of sh across reefs and their intensity of foraging within experimental food patches.

Benthic Community
A 10m × 20m grid was randomly selected along the leeward side of each reef, with experimental stations placed 5 meters apart. Photos of the benthos were taken at each experimental station (15 stations per reef). Images were analyzed using the program Coral Point Count with Excel extensions (CPCe) (Kohler and Gil 2006). Sixteen randomized points were projected onto each image and benthic cover type was identi ed for each point (see S.I. Table S1 for list of benthic variables). Benthos data were used to calculate benthic community composition, benthic diversity, and coral cover at each station.
Three-dimensional (3D) reconstructions of coral reefs were used to quantify coral rugosity using structure-from-motion (sfm) techniques ( Fish Community and T. duperrey Population: Belt transect surveys (25m long x 4m swath) were conducted on each patch reef during daylight hours to assess T. duperrey populations and the sh community composition. A total of four transects were surveyed per reef with transects haphazardly placed across 2 leeward sites and 2 windward sites. All shes within the belt transect were counted and identi ed to species. Belt transect data were not used to estimate predator populations because inherent biases in this type of data may lead to an overestimation of transitory predators and an underestimate of cryptic benthic predators (Caldwell et al. 2016). Therefore, we relied on the giving-up density approach to quantify perceived predation risk (Brown 1988).
Measuring Giving-up Densities and Reefscapes of Fear: The giving-up density (GUD) technique is commonly used to quantify the costs of predation risk associated with foraging at an experimental food patch with diminishing returns (Brown 1988). In this study, we used krill "burritos", which consisted of 4 krill placed within a substrate and wrapped in polyurethane coated mesh. Krill burritos were used to measure the GUDs of saddle wrasse at 15 stations (each 5 meters apart) forming a 10m × 20m experimental array on each of the 5 study reefs. Krill burritos were placed at each station and T. duperrey were allowed to forage from the patches for 4 daylight hours over a minimum of 10 experimental days (with the exception of 4 stations which were measured over 4 days (n=3), and 9 days (n=1)). After the 4hour period, the krill burrito was removed and the GUD (the remaining krill within the food patch) was quanti ed to nearest quarter krill. Food removal from the krill burrito generates diminishing returns: as more food is removed, the harder it becomes to access the remaining food items. Depleting a krill burrito was akin to a continuous process as the wrasse incrementally bite pieces from a given krill.

Statistical Analyses
We ran all statistical analyses using R Project for Statistical Computing version 3.3.3 (R Development Core Team 2018; package citations in Appendix S1).
A permutational multivariate analysis of variance (PERMANOVA) was used to test for differences in benthic communities between patch reefs by partitioning the Euclidean distance matrix by species with adonis in the vegan package. Finally, we tested for homogeneity of multivariate dispersions using betadisper and permutest functions in the vegan package. To visualize differences in benthic communities on patch reefs we ran a principal component analysis (PCA) on Hellinger-transformed abundance data using decostand in the vegan package, and prcomp in the ggord package. We assessed differences in coral cover and rugosity between reefs with generalized linear models (GLMs) with Gaussian error structure using the glm function in the base R package. We tested for signi cance between our GLM and the null model using the anova function in the base R package.
We calculated sh diversity (inverse Simpson's index) using the diversity function in the vegan package. Fish diversity and species richness were compared with a GLM using the glm function in the base R package. We used sh species abundance data to assess patch reef sh communities using the same multivariate analyses as described for the benthic communities above. We calculated the 10 most common species in our belt transect surveys (species present in highest numbers for combined data). We then calculated relative abundances of the 10 most common species, with all other species classi ed as "other". We investigated common standardization approaches for T. duperrey abundance data using the decostand function in the vegan package. Since no transformations were necessary (see S.I.), we assessed differences in untransformed T. duperrey density (individuals/100m 2 ) using GLM with Gaussian error structure using the glm function in the base R package. We tested for signi cance between our GLM and the null model using the anova function in the base R package.
To assess differences in GUDs, we used linear mixed effects models (LMMs) with reef and station nested within reef as xed effects, and experimental day nested within reef as random effects using the lmer function in the lme4 package. We used a multivariate analysis of variance (MANOVA) to test if coral cover and benthic diversity differ signi cantly between reefs. Next, we used an analysis of covariance (ANCOVA) using the aov function in the stats package to assess how coral cover and benthic diversity may have impacted average station GUD. Since there was no interaction between the covariates, we removed the interaction term and used only the main effects. We used coral cover and benthic diversity as covariates and reef as the grouping variable.
Mean GUDs of each station were used to create reefscapes of fear heatmaps using the plot_ly function in the plotly package. Contours represent lines of equal mean GUD. All other graphs and maps were created using ggplot2, ggpubr, and ggmap

Benthic Community
Reefs varied in benthic community composition by location (PERMANOVA, F 4, 67 = 12.213; P = 0.001) and dispersion (F 4, 67 = 2.636, P = 0.04). Principal component analysis of benthic composition showed clear separation of the two algae-impacted reefs from the three algae-free reefs ( Fig. 2A). PC1 and PC2 explained 41.05% and 21.8% of the variation in benthic community composition, respectively. While Montipora spp. coral is more characteristic of algae-free reefs, Porites spp. coral is more characteristic of algae-impacted reefs ( Fig. 2A). Dead coral occurred at a greater density on algae-impacted than on algaefree reefs.
Reefs varied by percent coral cover (ANOVA, F 4,67 = 11.63, P < 0.001; Fig. 2B) in Kāne'ohe Bay. Algae-free reefs (12, 13N, 13S) had signi cantly more coral cover than algae-impacted reefs (14 and 16) (P < 0.001). Reef 13S had the highest coral cover with 2.6 times more coral cover than Reef 14. Within reefs, there was no signi cant relationship between coral cover and benthic diversity (Appendix S1: Table S7), though we did observe trends where benthic diversity declined with increased coral cover on algae-free reefs. This is because they were mostly dominated by one or two species (i.e. Montipora capitata). Algae-impacted reefs had a trend of increased diversity as coral cover increased.

Discussion
The ecology of fear provides insights into how a common Hawaiian coral reef sh perceives its environment. Degraded, or algae-impacted, reefs differ in benthic community composition (Fig. 2), yet were mostly similar to healthy, or algae-free, reefs in terms of sh communities (Fig. 3). Saddle wrasse were common on all experimental patch reefs. Despite this, they perceive differences in predation risk among reefs (Fig. 4). Speci cally, saddle wrasse perceive predation risk across larger spatial scales, rather than microhabitats. Patch reefs are scary if they are degraded. Habitat degradation, when viewed from the point of view of the sh, in uences the predator-prey dynamics of this coral reef system by altering the reefscape of fear.

Fear on coral reefs
We found habitat degradation alters the fear of predation of a common Hawaiian sh. Fish reduced foraging in experimental food patches on algae-impacted reefs, indicating perceived risk of predation increased on the degraded habitats (Fig. 4) Surprisingly, saddle wrasse did not perceive signi cant variation in predation risk at small scales within reefs associated with microhabitat features. Rather, the wrasses perceived differences in predation risk at macro-habitat scales. This may be due to the anti-predator strategies employed by the saddle wrasse.
First and foremost, saddle wrasse manage predation risk by diel activity patterns. Saddle wrasse are 'early to bed' and one of the last shes to resume activity in the morning, thus almost entirely avoiding the riskiest times of dawn or dusk (Hobson 1972). Once active, wrasse swim higher in the water column than other species (Fulton et al. 2001). This likely emphasizes their advantages. Like other wrasse (Labridae) species, saddle wrasse are highly visual (Barry and Hawryshyn 1999) and relatively fast swimmers (Walker and Westneat 1997). Their strategy to avoid predation is to detect predators and quickly swim away. Therefore, the availability of nearby refugia at a foraging site within a reef's interstices may not in uence their perceptions of risk. Instead, the availability of refuge on the patch reef as a whole may be a more important factor for perceived risk.
How a sh perceives risk may vary with species based on their particular anti-predator adaptations. While we focused on the behaviors of saddle wrasse in this study, we also found a difference in the sh community composition on risky reefs. Interestingly, there was a higher abundance of sh with morphological anti-predator defenses of spines or keels, such as Moorish idols (Zanclidae), surgeon shes (Acanthuridae) and butter y shes (Chaetodontidae), on the riskiest reef, reef 16. This suggests that predation risk may in uence community level responses.
We note similarities between anti-predator responses in marine and terrestrial systems (Wirsing and Ripple 2010). For instance, both terrestrial and marine prey exhibit morphological and behavioral responses to predation risk. Prey defend themselves from predators with colors, structures, chemicals, grouping behaviors, or altered activity patterns. In particular, regardless of marine or terrestrial environment, prey will reduce foraging activity in response to risk, there is spatial heterogeneity in perceived risk, and the landscape of fear may vary with anthropogenic in uence. Unlike terrestrial systems, marine prey must contend with risk varying in an aqueous environment. This means there will be differences in the visual and chemical cues a prey receives. It also means that risk may vary in a 3dimensions (Harvey and White 2017; Lester et al. 2020). Finally, prey on pristine coral reefs must cope with high biomass of erce predators, which is not typical of terrestrial predator-prey systems (Malone et al. 2020).
Similarly to this study, spatial scale is also an important factor in many terrestrial studies on predation risk (Morgan et al. 1997;Shrader et al. 2007). Different prey species may assess risk at different spatial scales (Berris 1997). Unlike many terrestrial studies on predation risk, the variability in risk in our study occurred at much larger spatial scales. This may be due to predation being diffuse on coral reefs, coming from a diverse assemblage of piscivorous shes that use a wide variety of predation strategies. According to published dietary analyses, the main predators of saddle wrasses are blue n trevally. Blue n trevally are roving predators that spend their day patrolling one patch reef and move between reefs at night (Holland et al. 1996), making the risk of predation somewhat unpredictable day to day. We observed acute behavioral responses to the presence of blue n trevally, but they did not result in signi cant differences in GUDs. GUDs are therefore a reliable method to quantify chronic predation risk on reefs.
We considered the possibility that GUDs were not, in fact, quantifying predation risk on reefs, but instead quantifying other factors such as resource availability. The average GUD between reefs correlated positively with wrasse densities, which may seem to indicate that the GUD is an indicator of productivity. According to optimal foraging theory, however, sh living on reefs with high resource availability should have high GUDs, while sh on reefs with low resource availability should have low GUDs. This is because the marginal value of the resource within the food patch is higher for sh on low productivity reefs. We did not observe these trends, which supports our conclusion that the GUD was a measure of predation risk.

Impacts of habitat degradation
We demonstrated the long-lasting impact of an altered reefscape of fear due to habitat degradation caused by the chronic presence of invasive algae. Invasive algae, Euchema spp. and Kappaphycus spp., were uncommon on study reefs during the 2016 season because of herbivore biocontrol (Neilson et al. 2018). Even though the invasive algae were not a major component of the benthic community at the time of this study, their impact was still visible through the higher amount of dead coral (Fig. 2) observed on Reef 16. The unseen ecological impact of habitat degradation is the reduction in foraging behaviors across patch reefs. Changes to this non-consumptive effect of predation will impact the predator-prey dynamics of the system.
Habitat degradation may alter the directionality of non-consumptive effects. Here, we show how the nonconsumptive effects of predation can also be impacted from the "bottom-up". Altered habitat due to degradation changes the refuge availability and thereby the predation risk perceived across the reef. In contrast, non-consumptive effects can be altered from the "top-down". Changes in predator populations drive anti-predator responses in prey and result in trophic cascades (Madin et al. 2019). It is unlikely the differences in predation risk in this study were due to differences in predator populations, as reefs were in close proximity (within 1.5 km of each other) and none of the reefs are within protected areas.
Tradeoffs between resource availability and safety -conservation implications: Habitat degradation may in uence the balance between resource availability and safety. When approached through the lens of tradeoffs, we can classify coral reefs into ecologically relevant categories based on resource availability and predation risk (Appendix S1: Table S21) (Kotler et al. 2016). This classi cation approach provides a complementary metric to classical reef assessments. Saddle wrasse are good candidate species for this approach as they are a common benthic invertivore and the primary species lling this niche on Hawaiian reefs. In this study, wrasse density provides a metric of invertebrate productivity (Fig. 3C) and GUDs a metric of predation risk (Fig. 4). Through the concepts of ideal free distribution theory and density dependent habitat selection, we expect resource availability to be re ected through a positive correlation with wrasse abundance. According to patch use theory, GUDs in risky habitats will be higher than those in safe habitats. Based on wrasse abundance and GUDs we can classify the reefs in our study. Algaeimpacted reef 16 is an inferior reef for saddle wrasse as it has the lowest density of saddle wrasse and the highest risk. Algae-free reef 13S is core wrasse habitat; it has the highest wrasse density and lowest risk. Interestingly, saddle wrasse did not view algae-impacted reef 14 as being inferior habitat (risky and resource poor), but instead a core predator habitat (risky and productive). This may be due to the high amount of coral rubble and sand on the reef which saddle wrasse preferentially use for foraging (M.A.M unpublished data). Reefs 12 and 13N can both be categorized as refuge habitat, as they are relatively safe but resource poor.
This habitat classi cation approach according to predation risk and resource productivity has conservation implications and can serve as a direct behavioral indicator of populations and communities (Kotler et al. 2016). This paradigm has been successfully applied to several terrestrial systems (Reid 2004; Vickery 2010; Morris and Dupuch 2012; Abu Baker and Brown 2013). In common with these terrestrial applications, the wrasse's distribution and behavior revealed habitat quality. In contrast, we investigated habitat degradation impacts on food and safety tradeoffs. By considering tradeoffs, managers can assess habitats and the e cacy of conservation efforts from the point-of-view of inhabitants themselves. Coral reefs are threatened worldwide and management strategies should focus on the ecological implications of habitat degradation. We advocate GUDs as a standardized technique, which complements other metrics.

Conclusion
Habitat degradation alters the reefscape of fear on coral reefs. Increased predation risk on degraded reefs alters predator-prey interactions and may fundamentally alter how predation structures populations and communities. We advocate taking an integrative approach to study the ecology of fear on coral reefs. In particular, studies of habitat degradation, as measured by sh abundances, coral properties, and resource abundances, would bene t from a measure of giving-up densities. Allowing sh to reveal their perceptions of habitat through their patch use behavior provides a critical piece of information for both assessing and monitoring reefs.
Declarations Figure 1 Map of Kāne'ohe Bay, O'ahu, Hawai'i with study area indicated by black box (See Fig. 4). Top right: T.