A seascape approach for guiding effective habitat enhancement: Spatially explicit modeling of kelp–grazer interactions

In temperate to polar coastal environments, large brown macroalgae, commonly known as “kelp”, are highly valued as important primary producers and formers of structurally complex habitats for invertebrates and fish (Steneck et al., 2002; Krumhansl and Scheibling, 2012; Bertocci et al., 2015). However, kelp habitats are being lost in over one-third of their global distribution, threatened by a suit of interacting natural and anthropogenic drivers (Krumhansl et al., 2016; Kriegisch, Reeves, Johnson, and Ling, 2019). Globally, increasing water temperatures are exceeding the physiological thresholds of kelp, often allowing turf algae to replace them (e.g., Filbee-Dexter and Wernberg, 2018; Dijkstra et al., 2019). More locally, a major threat to kelp habitats is overgrazing by urchins (Ling et al., 2015), which can lock the system into a less diverse and productive “barren” state dominated by coralline algae (Steneck et al., 2013).

To counter kelp habitat loss, several restoration approaches have been proposed, including urchin removal (e.g., Sanderson et al., 2016), clearing turf from substrata (e.g., Perkol-Finkel and Airoldi, 2010), and transplanting kelp onto artificial structures (e.g., Terawaki et al., 2001; Perkol-Finkel et al., 2012), including artificial reefs, which are commonly used for habitat restoration in general (Fabi et al., 2011; Feary et al., 2011). More recently, the use of a modular design for artificial reefs allows separate units to be distributed across the seascape (Dyson and Yocom, 2015; Tessier et al., 2015).

In Eastern Canadian coastal ecosystems and in particular the Gulf of Saint Lawrence (GSL), modular reefs have been used to comply with regulations requiring offsets for damage to marine habitats caused by coastal development (Fisheries Act, 2019). This approach makes the GSL an ideal situation for the use of artificial reefs to enhance kelp habitats in a region where they are typically restricted in their distribution due to winter ice scouring in shallow areas and urchin overgrazing in deeper areas (Tamigneaux and Johnson, 2016). Indeed, due the lack of biological controls, exceptionally high densities of the green sea urchin, Strongylocentrotus droebachiensis (up to 500 ind. m^–2; Gagnon et al., 2004) occur in the GSL, making it an extreme case of a state of urchin barrens (Ling et al. 2015; Johnson et al., 2019).

In this context, if artificial reefs can be colonized by urchins, overgrazing poses a serious threat to kelp enhancement projects (Terawaki et al., 2003). However, since urchins are mainly associated with rocky bottoms (Christian et al., 2010), heterogeneous habitats with less favorable substrata (e.g., sand) interspersed among patches of preferred substrata (e.g., cobbles, boulders) could affect the activity of these consumers, thus offering refugia from herbivory for kelp growing on artificial reefs.

Despite significant research advances on the ecological functioning and quality of artificial structures as novel habitats for benthic assemblages (Dyson and Yocom, 2015; Bishop et al., 2017; Mayer-Pinto et al., 2017; Perkol-Finkel et al., 2018), experimental investigations have generally compared contrasting alternative categories (e.g., artificial–natural, vegetated–unvegetated, hard–soft bottoms) over the more heterogeneous situations typical of most bottoms. While such binary contrasts can be informative, they often represent the end points of environmental gradients or are mainly applied at larger spatial scales. Moreover, they may not adequately account for ecological interactions (e.g., herbivory, predation), especially those involving mobile consumers operating at smaller scales. For example, on coastal defense structures, the fate of algal transplants was largely determined by grazers (Ferrario et al., 2016).

In contrast, landscape ecology—a well-established discipline in terrestrial studies—recognizes the role of the spatial configuration of the surrounding landscape in influencing how species use their habitats and interact locally. For instance, large herbivores adapt their movement and grazing activity to the types of land cover (e.g., refuge-offering forest stands) in their proximity and their exposure to predation risk (Fortin et al., 2005; Mason and Fortin, 2017), whereas forest heterogeneity determines paths of differential movement resistance affecting the outbreaks and spread of pests (Powell et al., 2018). However, in spite of difficulties in both mapping the bottom and tracking animals in marine systems, ecologists have increasingly considered the spatial configuration of seascapes (i.e., heterogeneous areas of the coastal sea floor; Boström et al., 2011) to better identify the mechanisms underlying the ecological patterns. A pioneering study showed that areas of seagrass acted as corridors for predatory crabs from salt marshes to distant oyster reefs (Micheli and Peterson, 1999). Similarly, predation risk for a Mediterranean urchin has been found to be a function of the arrangement of seagrass patches (Farina et al., 2016), and aerial imagery and seascape mapping demonstrated that grazing pressure on macroalgae decreases with the distance from the coral reefs that provide shelter to herbivorous fishes (Madin et al., 2011; Gil et al., 2017).

Seascape approaches and species distribution models that study the distribution of kelp beds and urchin barrens have been so far conducted at large scales and using only abiotic environmental parameters as predictors (e.g., Rinde et al., 2014; Parnell, 2015). While such large-scale studies highlight general trends, they may offer an oversimplified representation of the seascape, ignoring any effect of local habitat patchiness and heterogeneity on ecological interactions at smaller scales. Their utility may thus be limited for the design and deployment of artificial reefs. Indeed, understanding the effects of the seascape on local ecological interactions appears essential to assess the achievability of the ecological goals of artificial reefs, and this knowledge will assist managers and conservationists in minimizing the risk of failure of habitat enhancement projects.

In this work, we investigated whether seascape heterogeneity creates areas of lower grazing risk (i.e., refugia) for kelp transplanted onto modules of artificial substrata and characterized the properties that define this risk. We focused our attention on the effect of 3 seascape properties: bottom substratum (e.g., boulders, gravel, sand), algal cover, and spatial configuration (i.e., position of various elements relative to one another). Specifically, our research questions were (1) does kelp survival decrease in areas with higher rocky substrata due to a greater abundance by urchins there? (2) does the abundance of other algae impact urchin distributions thus potentially influencing kelp survival? and (3) is the distribution of urchins affected by the spatial arrangement of the modules, either promoting movement as “stepping stones” or reducing movement by acting as local “sinks”, that is, features that retain urchins at one location?

The study was conducted in the late summer 2015 at Baie du petit Métis, a small bay on the south shore of the St. Lawrence maritime estuary (48°40′37.30″ N, 68° 0′41.24″ W; Québec, Canada; Figure 1). The bay is separated on the northwest side from the St. Lawrence River by a narrow peninsula characterized by rocky shores, while sandy beaches become gradually predominant moving eastward. A wide intertidal mudflat with sparse rocky outcrops is followed subtidally by unconsolidated gravel, cobble and boulders heterogeneously arranged among large patches of sand. Fucoid seaweeds (i.e., Ascophyllum and Fucus spp.) are the most abundant vegetation in the rocky intertidal zone. Kelp in the bay are spatially confined to rocky intertidal pools and the shallow subtidal (approximately 1 m), primarily represented by perennial species Alaria esculenta, Saccharina latissima, and Agarum clathratum (Bégin et al., 2004; Krause-Jensen et al., 2012). No subtidal patches of kelp were present at the time of the study. Red foliose algae Ptilota serrata and Phycodrys rubens (Himmelman, 1991) are common in the GSL, and although we did not identify foliose algae taxonomically, these species were likely representing the majority of the seaweeds in the study area based on observations while diving.

The only sea urchin in the area is the green sea urchin S. droebachiensis, an omnivore that preferentially feeds on perennial kelp such as A. esculenta but includes other algae (including red algae) as well as animal tissues in its diet (Himmelman and Steele, 1971). Preliminary surveys indicated that urchins in the bay had a variable density of 8 ± 10 ind. m^–2 (average ± SD, n = 94).

To evaluate how the seascape might influence survival of subtidal kelp, we selected a site of 256 m² featuring a heterogeneous bottom (i.e., a mix of sandy and rocky patches) at an 8-m depth within the bay (Figure 1b). The area of the site was chosen as a trade-off between the need of capturing bottom heterogeneity, the required workload in SCUBA, and the mobility of green sea urchins (up to 5 m d^–1; Dumont et al., 2006). The site depth was chosen to be representative of the subtidal bottom usually overgrazed by urchin in the GSL and well within the depth range typically occupied by kelp. Initial surveys mapped relevant seascape variables (see Bottom mapping) after which the survival of transplanted kelp was monitored (see Kelp survival and urchin distribution). We adopted a two-step modeling approach to describe first the survival of kelp as a function of different urchin metrics (e.g., presence/absence, density), and second the habitat use by urchins as a function of seascape properties. By coupling the 2 models, we mapped the grazing risk in the area.

We mapped the study site utilizing a photographic sampling method via SCUBA over a permanent 16 × 16-m grid divided into 64 2 × 2-m cells (Figure S1). Each cell was subdivided into 4 parcels and photographed using a GoPro Hero3⁺ camera mounted on a frame (Figure S2). We defined the areal extent a priori to optimize the workload.

Photographs were analyzed to quantify percent cover of substratum type and selected algal categories. Substratum type included 3 categories: “boulders” (>25 cm), “cobble and gravel” (0.2–25 cm, grouped to facilitate the analysis), and “sand” (<0.2 cm; sensu Wentworth, 1922). Algal categories included foliose algae (mainly red algae), turf-sediment, and crustose coralline algae. Algal cover was estimated at all sampling times to monitor changes, while bottom type classification was conducted only on photographs taken at the beginning of the experiments and then assumed to remain constant. Refer to Supplemental Material for further details.

We transplanted one reproductive individual of A. esculenta on each of 18 modules that each consisted of one concrete masonry unit (aka cinder block, but henceforth “block”) measuring 15 × 40 × 20 cm, W × L × H; Figure S3a) on August 5, 2015 (T_days = 0). Each block was placed on its side (thus with openings to hollow interior accessible directly from the bottom) and positioned in the center of cells randomly chosen from among the 36 cells of the inner grid (i.e., excluding cells located along the edge of the grid to allow calculation of neighboring statistics). The number of blocks used was chosen to occupy only half the available cells to have the variation in between-block distances. Random assignment of blocks resulted in 10 deployed in rocky cells, 8 in sandy cells (Figure 1). Kelp survival was assessed every 10–20 days until October 13 (T_days = 69) by confirming its primary meristem was still present (Mann, 1973).

We counted the number of visible urchins in each photograph using the Cell Counter plug-in for ImageJ 1.47, considering only individuals with a diameter greater than 20 mm (smaller urchins are generally cryptic). Urchins found inside the cavities of the artificial blocks (hereafter “hiding urchins”) were also counted.

Grazing was not the only possible cause for kelp loss, particularly with transplant experiments. Poor growing conditions, transplant stress, and dislodgment were other possible mortality vectors. To evaluate our transplant technique and test the suitability of the physical environment, we also transplanted kelp onto a frame suspended off the bottom (i.e., 3 ropes arranged in a pyramidal structure; Supplemental Material). This frame served as an effective herbivore deterrent while exposing them to similar hydrodynamic conditions as those transplanted onto blocks. We deployed the kelp frame in July 2015 with kelp plants spaced 30 cm apart along the lines (Figure S4). In October, we recovered all kelp transplants, recording for each (i) its position (depth) on the frame, (ii) whether it had “survived” (described above), (iii) whether transplants developed reproductive sporophylls, and, if so, (iv) sporophyll length.

We modeled kelp survival as a function of different sea urchin metrics using survival analysis. This approach models the “time until an event occurs”—that is, survival time—and accommodates censoring by retaining individuals with unknown survival time in the data set (Kleinbaum and Klein, 2005). We defined the survival time as the number of days from the start of the experiment (T_days = 0) to the midpoint between the sampling date when the loss of an individual kelp was first observed and preceding sampling date (e.g., for a kelp observed at T_days = 20 but not observed at T_days = 30, survival time = 25). We considered 6 predictor variables: “urchin frequency,” defined as the number of times that urchins were present in a cell with a block; “urchin density” as the median of the average urchin density recorded at each sampling time in a cell with a block; median urchin frequency in neighboring cells (i.e., the 8 cells surrounding the focal one); median of the urchin density in neighboring cells; “frequency of hiding urchins” as the number of times when urchins were found in cavities of the block; and finally the median of the number of urchins hiding in the block.

We fitted semiparametric Cox Proportional Hazards (CoxPH), Weibull, and exponential parametric survival models for each predictor and a null model (i.e., no effect of predictors) for a total of 21 candidate models and compared them via second-order Akaike information criterion for small sample sizes, AICc (Mazerolle, 2006). Because of the limited sample size (n = 18), we did not specified models with more than one predictor. We checked the assumption requirements (Kleinbaum and Klein, 2005) and hereafter report only the CoxPH models since they outcompeted their parametric alternatives.

We modeled sea urchin presence/absence as a function of 4 major seascape properties: bottom composition, algal cover, spatial configuration of patches and discrete natural features, and spatial arrangement of the artificial reef (i.e., reef setup). In addition to the variables described in the bottom mapping section, predictors included the median values of the cover of cobble and gravel, foliose algae, coralline algae, and turf-sediment in neighboring parcels. We assigned each parcel to either a sandy or a rocky patch and calculated spatial configuration predictors as the shortest distance of a parcel to a sandy patch, to a rocky patch, and to the closest parcel containing a boulder. Finally, we considered the reef setup calculating the distance from the closest parcel of a cell containing a block. All distances were in relative units (i.e., between parcels distance).

We checked all potential predictors for collinearity and retained those deemed more relevant. We modeled urchin presence in each parcel and sampling time using Binomial Generalized Linear Mixed Models with a logit link and “parcel” as a random effect following Zuur et al. (2009); percent cover predictors were arcsin transformed. We built a set of 17 candidate models that included selected combinations of seascape properties (see model set in Table 2) and compared them using AIC.

For both analyses of kelp survival and urchin habitat use, we considered models to have substantial support when they had a ΔAICc or ΔAIC < 2 (Mazerolle, 2006). Whereas only one model for kelp survival met this criterion and was thus selected as the “best” model, 3 models for urchin habitat use showed comparable support and were thus averaged to decrease the selection bias (Mazerolle, 2006).

All statistical analyses were performed using R 3.3.2 (R Core Team, 2016). We used packages “raster” for spatial analysis, “survival” and “lme4” for model fitting, and “MuMIn” for model averaging. All estimates of variance listed hereafter are standard error (SE). See Supplemental Material for more details on analyses.

Models describing the urchin distribution and kelp survival were coupled to create a map representing the risk for a kelp of being grazed in different areas throughout the site. Using the “urchin habitat use” model, we first mapped urchin frequency, assigning to each cell the maximum value between the maxima of predicted urchin presence probabilities in its parcels at different times. We then used this map as the input for the kelp survival model. The output of a CoxPH model is the log-hazard ratio, log(HR), here representing the risk for a kelp to be grazed and expressed relative to a cell with urchin frequency equal to 0. When the log(HR) equals 0 then two cells have the same risk.

The spatial configuration of our study site was characterized by 2 sandy patches divided by 1 continuous patch of rocky bottom (Figure 1b and c); however, the 2 bottom types were almost equally represented (27:37, sandy: rocky cells). Rocky patches were mainly composed of cobbles and gravel with boulders covering only 5.73 ± 0.1% (mean ± SE), while sand accounted for 19.1 ± 3.0%. Sandy patch composition was more homogeneous with sand covering 91.3 ± 1.7% of the bottom with a sparse presence of cobble and gravel, and boulders (7.0 ± 1.6% and 1.5 ± 0.5%, respectively). Throughout the experiment the density of visible sea urchins in the rocky patch was 0.92 ± 0.08 ind. m^–2 (mean ± SE, n = 740) and of 0.13 ± 0.02 ind. m^–2 (n = 540) in the sandy patches.

In the absence of herbivory from sea urchins (i.e., on the kelp frames), 83% of kelp transplants survived and all developed reproductive sporophylls (sporophyll length = 26.1 ± 0.1 cm, Figure S5), indicating that very little mortality was due to the transplant technique and that the physical environment was suitable for kelp growth. There was no statistical effect of depth on the number of lost transplants (p = 0.17, Generalized Linear Model with a Poisson distribution) or on sporophyll length (p = 0.4, Figure S5).

40% of initially transplanted kelp survived until the end of the study (Figure 2). We observed a sharp decrease in survival (40% decrease) in the first 25 days followed by a constant decrease. Kelp survival was significantly related to the local urchin frequency, log(HR) ± 0.95 confidence interval (CI) = 5.8 ± 2.9; Likelihood Ratio: χ² = 21.1, 1 df, P < .001; Table 1. The risk exposure decreased (i.e., hazard ratio < 1) as urchin frequency decreased from the average value toward zero, whereas it greatly increased (hazard ratio > 1) as urchin were more frequently present in a cell (Figure 3).

Three models resulted as good candidates between the ones tested and were therefore averaged (ΔAIC < 2; Table 2). The averaged model retained the variables boulders, cobble and gravel, foliose algae, coralline algae, neighboring foliose algae, and distance to block. All variables, except distance to block, were positively correlated with urchin frequency, although only boulders and cobble and gravel were significant, while coralline algae were almost significant (Table 3). The percentage cover of boulders in a parcel had the greater effect size, meaning that an increase in boulder cover contributes the most in the frequency of urchin presence (Figure 4). The averaged model showed a good calibration (i.e., a good agreement between observed and fitted values) for fitted values below 0.4, otherwise the model seemed to slightly underestimate the observed values (Figure S6). Model discrimination (i.e., the ability of model-predicted values to discriminate between those with and without the outcome), assessed by the area under the receiver curve, was satisfactory (Area Under the receiver operating characteristic Curve [AUC] ± .95 CI = 0.95 ± 0.01; perfect discrimination when AUC is equal to 1).

The grazing risk was concentrated on but not limited to the rocky patch where it was generally higher than on the sandy part of the bottom (Figure 5). The distribution of risk intensity also differed in the 2 patch types: risk intensity in rocky cells ranged between 0.6 and 5.2 and presented an overall uniform distribution, while log hazard ratio on sand ranged between 0 and 4.0, but 55% of the cells were associated with the lowest risk (Figure S7).

A visual analysis of the map of grazing risk revealed the existence of hotspots of risk in both rocky and sandy regions (i.e., darker cells in Figure 5): Cells associated with the highest risk were distributed at different positions in the grid and separated from each other by lower risk cells. The position of the risk hotspots seemed to be driven primarily by the distribution of boulders in the grid. Indeed, Figure 4 showed that an increase in the percent cover of boulders in a given cell determined a more direct increase of urchin frequency compared to the amount of cobbles and gravel, the other significant variable in the model (Table 3). Interestingly, kelp surviving to the end of the experiment in 2015 were still alive at the beginning of the next summer (Figure 5).

For kelp-urchin dynamics (e.g., Steneck et al., 2002; Scheibling et al., 2011; Ling et al., 2015) and the design and functioning of artificial reefs (Bulleri and Chapman, 2010; Fabi et al., 2011; Feary et al., 2011; Lima et al., 2019, and references therein), the binary contrast of alternative habitats (i.e., kelp bed vs. barren ground or artificial vs. natural reefs) examines the end points of what is actually a continuum. By specifically examining small-scale seascape heterogeneity, we provide a novel spatially explicit framework that incorporates ecological processes and is more relevant for the scale for the design of artificial reefs relative to more commonly used species distribution models. Using this framework, we investigated for the first time the role of the local seascape heterogeneity in modulating the interaction between urchins and kelp. We showed that grazing risk intensity is modulated by the seascape properties, including biological characteristics and discrete seascape features, to locally alter microhabitat use by this herbivore, even within the same patch. Importantly, we showed that we can quantify grazing risk, that is, kelp refugia, at small spatial scales based on local seascape heterogeneity.

Seascape heterogeneity often results from multiple concurrent organizational levels (Didham, 2010). Here, we examined heterogeneity over 3 levels: bottom patchiness, within-patch variability, and spatial configuration. The patchiness of bottom composition played a major role in determining kelp refugia. Sandy patches were overall safer for kelp compared to rocky patches, an expected outcome as urchins prefer hard substrata. This result confirms the utility of coarse bottom classification into binary categories for preliminary identification of kelp refugia in the seascape when planning the deployment of artificial reefs modules.

Next, we resolved patch heterogeneity at a finer scale using our two-step experimental approach to couple kelp survival with urchin habitat use. By identifying the factors affecting kelp survival, we could translate within-patch heterogeneity into variability of grazing risk, which did not follow an all-or-nothing pattern corresponding to the binary bottom classification. Boulders emerged as a key seascape feature, creating hotspots of grazing risk within both rocky and sandy patches. This source of heterogeneity drove the urchin distribution at our study site by locally enhancing the structural complexity of the bottom. Elsewhere, urchins greatly favor boulders over cobble beds, apparently for reducing predation risk (Scheibling and Hamm, 1991). Similarly, urchins prefer microhabitats that minimize dislodgment risk (Frey and Gagnon, 2016). The cover of foliose algae, both in focal and neighboring cells, also contributed to this within-patch heterogeneity and had a predictive value for urchin presence and therefore for grazing risk in the area. As a potential food source, their presence could either attract and/or retain urchins or simply interfere with movements. Urchins often move randomly (Lauzon-Guay et al., 2006) and thus in an area dominated by foliose algae would more frequently encounter algal patches increasing the chances of remaining in the area. Similarly, the percent cover of coralline algae was a predictor of urchin presence, but more likely as the result of past grazing activity (e.g., either exposing coralline algae after the removal of erect algae or freeing the substratum for coralline expansion) rather than acting as a resource.

Finally, the heterogeneity resulting from the spatial configuration of the landscape can affect species interactions (Didham, 2010). For example, meadow proximity influences the trajectory of free-ranging bison moving in a patchy forest landscape (Dancose et al., 2011), corridors of vegetation ease the spread of mountain pine beetles (Powell et al., 2018), and the spatial arrangement of seagrass affects predation on urchins (Farina et al., 2016). Our models included variables describing the spatial configuration of both the seascape and arrangement of the modules (Table 2). However, the limited spatial extent of our study or the abundance of cells containing boulders (Figure 4) possibly hindered our ability to capture an effect of seascape configuration. Nonetheless, the slightly negative effect of the distance from a block on urchin frequency suggests that blocks could have possibly acted as urchin attractors. Indeed, we commonly observed urchins occupying the block cavities (Figure S8), a behavior consistent with their preference for sheltered microhabitats (Scheibling and Hamm, 1991).

Grazing risk for kelp is thus an emergent property resulting from the interaction between urchins and local habitat heterogeneity. The survival of kelp was negatively correlated to the frequency of urchin presence in a given cell. Since urchins are mobile and capable of moving among cells within a day (Dumont et al., 2006), their movement behavior ultimately underlies their local distribution including their density, which was the second-best explanatory variable for our predictions of risk. Unfortunately, we were not able to determine whether higher urchin frequency or densities were due to an accumulation of urchins in particular cells or to a more dynamic reshuffling of individuals between samplings (tracking the movement of individuals would have been necessary to distinguish these two possibilities but is logistically complex). Our results confirmed that urchins are associated with particular characteristics or hard bottom features of the seascape but not exclusively. Indeed, recent studies (MacGregor, 2020) have shown that sand is not a barrier to urchin movement, but it affects urchins in that they will cross patches of sand but not remain on them. This behavior would help explain the higher risk associated with certain landscape features within sandy patches (i.e., isolated boulders) as well as the colonization by urchins of our blocks placed there. The propensity of urchins to move onto soft substrata is still poorly known, but it likely underlies the low grazing risk for kelp in such habitats. If individuals are less likely to venture onto soft substrata, then the probability of encountering small rocks or isolated boulders would be reduced. Moreover, once on a sandy bottom, S. droebachiensis appears to move further across the bottom (possibly either more quickly or more directly) than on rocky bottoms (Dumont et al., 2006), which would also reduce the encounter probability. Finally, once on an isolated hard substratum, an urchin might be more likely to leave due to the limited suitable habitat.

Seascape ecology offers a conceptual framework to identify areas with differential intensity of species interaction, and while of fundamental interest, it also informs ecosystem management. Our modeling approach offers a method to inform stakeholders with both predictions of risk at the site scale and an indication of which predictor variables need to be measured. In our specific case, we assessed grazing risk at the small scale at which artificial reefs are used in restoration and habitat enhancement projects. This predictive capability could allow the successful development of artificial reefs in locations that are more relevant, namely the heterogeneous bottoms that are more typical of shallow coastal waters. In particular, the ability to deploy reef modules closer to natural rocky bottoms reduces some drawbacks of past approaches while promoting connectivity between artificial and natural habitats (Dethier et al., 2003; Bishop et al., 2017). For example, to minimize grazing impacts, reefs are usually deployed in areas of extensive sandy bottoms or built to greater heights (Terawaki et al., 2001; Terawaki et al., 2003; Reed et al., 2006) where greater exposure to wave action reduces urchin foraging (Blain and Gagnon, 2014). These measures likely increase costs (e.g., more construction material, further distances) and result in the isolation of kelp beds. Moreover, the introduction of hard substrata into soft-sediment habitats alters their ecology (e.g., hydrodynamics, sedimentation, organic matter inputs; Bishop et al., 2017; Heery et al., 2017), yet grazers can still colonize through recruitment, removing macroalgae and possibly creating opportunities for invasive species (Dafforn et al., 2012). In contrast, deploying reefs modules in proximity of existing rocky bottoms but within areas of reduced grazing risk could promote positive feedback on natural habitats. In particular, kelp refugia within such seascapes would enhance recruitment (kelp are generally poor dispersers; Gaylord et al., 2004) and productivity. Indeed, the addition of locally derived kelp detritus could enhance food resources for urchins, enhancing their commercial value (Claisse et al., 2013; Cresson et al., 2014) and reducing destructive grazing perhaps to the point that the ecosystem switches to the alternative stable state of a kelp forest (Kriegisch, Reeves, Flukes, et al., 2019).

Ultimately the performance of kelp-enhancing artificial reefs must be sustainable through time, ideally without active maintenance. Although our work was not designed to look at seasonality and long-term performance, some speculation can be advanced. Grazing risk may decrease overwinter primarily because urchins tend to migrate in deeper areas or to seek refuge in crevices to escape increased wave disturbance during winter storms (Konar and Estes, 2003; Adey and Hayek, 2011). Additionally, food intake in S. droebachiensis reaches minima during winter (Siikavuopio et al., 2007) indicating that seasonality seems to affect feeding rates more than temperature (Suskiewicz and Johnson, 2017). From a kelp perspective, winter represents a relatively inactive season because of reduced light availability. However, winter irradiance levels might become less limiting if the current trend in climate change induced sea-ice cover reduction persists (Krause-Jensen et al., 2012; Galbraith et al., 2019). We witnessed the overwintering of kelp when we opportunistically revisited our study site at the beginning of the next summer season (Figure 5). Despite having just an anecdotal connotation, this observation is encouraging, and we cautiously suggest that seasonality might not negatively affect the kelp-urchin dynamic described here. Finally, having observed that artificial reefs on isolated sandy areas in the GSL can successfully maintain kelp population for several years (Johnson and Ferrario, personal observations, August 2018), we are confident that long-term performance of artificial reefs will not be an issue provided that the number of sea urchins reaching the reef can be limited.

Since restoration projects are necessarily implemented at the local scale, our seascape approach to assess the intensity of species interactions can act as a guide. Several factors, however, need to be considered when applied more widely. First, the mobility (e.g., range, behavior) of relevant species needs to be clearly assessed as it determines the spatial scale and resolution at which seascape data need to be acquired. Unfortunately, this information is often largely absent. Second, accurate mapping of the bottom of coastal habitats, with a particular attention to biological data (e.g., vegetation, benthic invertebrates) is needed. Finally, artificial reef projects should be carefully monitored to identify factors leading to failure, but relevant information is often not collected or is not publicly available (Tessier et al., 2015).

Our work demonstrates that with such information, we can understand how the intensity of species interactions is modulated by the surrounding seascape and can be readily incorporated in the planning phase of restoration projects to increase their rate of success and ensuring lasting results.

Data sets accompanying this article are archived on Scholar Portal Dataverse and accessible at https://doi.org/10.5683/SP2/7P4TMB.

The supplemental files for this article can be found as follows:

Text S1. Figures S1–S8.Docx.

We thank Chris McKindsey (Maurice Lamontagne Institute of Fisheries and Oceans Canada), the Association des résidents de la pointe du phare, Ted Savage, and Diana Lafleur for logistic support as well as Benoit Dumas, Heather Hawk, Kevin Ma, Katie MacGregor, Leonardo Miranda, Carla Narvaez, and Anne-Sara Sean for their assistance in the field. FF dedicates this work to the late Dr Shimrit Perkol-Finkel, bright scientist, successful entrepreneur, amazing woman, and mother, for her friendship and mentorship and for being an inspiring example of what we can achieve when passion motivate us: “if you want, you can.”

This study was supported by Fonds de recherche du Québec—Nature et technologies (FRQNT) Merit scholarship program for foreign students (Filippo Ferrario) and the Canadian Natural Sciences and Engineering Research Council (NSERC) Discovery Grant and Strategic Grant programs (Ladd Johnson).

The authors declare no competing interests.

Contributed to conception and design of the study: FF, LEJ.

Contributed to performing the experiment and acquisition of data: FF, TS.

Contributed to analysis and interpretation of data: FF.

Contributed to leading the writing of the article: FF.

Contributed to critical revision and approval of the article: FF, LEJ, TS.

How to cite this article: Ferrario, F, Suskiewicz, T, Johnson, LE. 2021. A seascape approach for guiding effective habitat enhancement: Spatially explicit modeling of kelp–grazer interactions. Elementa: Science of the Anthropocene 9(1). DOI: https://doi.org/10.1525/elementa.2021.00013

Domain Editor-in-Chief: Steven Allison, University of California, Irvine, CA, USA

Associate Editor: Martha A. Sutula, Southern California Coastal Water Research Project, Costa Mesa, CA, USA

Knowledge Domain: Ecology and Earth Systems