**1. Introduction**

Climate change and anthropogenic pressures have fragmented and restricted the distribution of many species worldwide, with significant shifts documented in a vast array of ecological communities [1–4]. In coastal ecosystems, numerous studies have reported a decline in biodiversity [5–12]. For example, Sorte et al. [9] noted a 60% decline in the abundance of the blue mussel *Mytilus edulis* over the past 40 years along the coastline of Eastern USA and linked this decline with that of several other species within the intertidal community. With at least a billion people expected to live within the lower-elevation coastal zone by 2060 [13] and up to 12.5 million km<sup>2</sup> of natural habitat potentially replaced by 2030 [14,15], there persists a need to identify how novel artificial habitats impact coastal ecology.

Introducing hard-engineering structures can negate many of the perceived negative geomorphological and economic impacts of coastal erosion, particularly in urban environs; however, these structures can have significant implications for the configuration of intertidal habitats and biodiversity [16–18]. Studies have predominantly focused on determining whether the communities found across analogous natural shores (largely situated in rural areas) versus artificial structures

(largely situated in urban or semi-urban areas) are comparable [18]. The consensus so far is that there is higher diversity on natural shores than analogous artificial rocky sea defences [19–26]. However, investigations have not been conclusive, identifying di fferences across tidal heights [27] and taxa [28,29], while others have found a strong similarity between structures [30,31].

Rock-rubble groynes are commonly implemented hard-engineering structures that run perpendicular to the shoreline, intercepting longshore transport of sediment. The intertidal habitat of these artificial structures o ffers a new rocky habitat that is typically located in predominantly sandy shorelines; however, there are very few studies investigating the ecology of rock-rubble groynes. Studies by Pinn and Rodgers [32] of a natural rocky shore and by Pinn et al. [21] of rock-rubble groynes in Dorset, UK, found higher species richness and abundances of most species on the natural rocky shore at Kimmeridge Bay than on the rock-rubble groynes at the nearby Sandbanks Peninsula. However, the ecological comparison between the structures was not the predominant focus of either study, and there was no statistical analysis beyond the simple comparisons of biodiversity. Similarly, in a study comparing the ecology of eight artificial structures (five of which were groynes) and eight natural rocky-shores in the UK, Firth et al. [23] found higher mean species richness on the natural rocky shores than the artificial structures, with no species unique to the artificial structures. Firth et al. [23] also compared habitats, finding that rock pools supported greater species richness than rock habitats, irrespective of structure. This contrasts with the results of Pinn et al. [21], who found more species on exposed rock than in the pools on the groynes at Sandbanks. The findings of these studies sugges<sup>t</sup> that, while groynes support a lower level of biodiversity than their natural counterparts, they could provide a refuge for intertidal communities found on rocky shores that are under pressure from increasing urbanisation.

Conversely, a major criticism of locating hard-bottom artificial structures in soft-bottom areas of coastline is that they can contribute to the decline of barriers (impassable areas of soft-bottom coastline) which isolate distinct regions of rocky shores: removing barriers may enable the dispersal of larvae and propagules of invasive species beyond their natural limits [33–35]. For example, Airoldi et al. [35] found that non-indigenous species were 2–3 times more abundant on artificial structures in part of the North Adriatic Sea, and several other studies have also found artificial structures to support non-indigenous species [36–39]. These studies conclude that there are more non-indigenous species on artificial structures than on nearby rocky shores because they act as points of invasion for many of the non-indigenous taxa. Because of these points of invasion and the resulting creation of stepping-stone dispersal corridors, hard-bottomed artificial structures may pose a serious concern for biodiversity [40].

The UK coastline is one of the most highly human-impacted ecosystems in the world [41]. With the projected coastal urbanisation and climatic changes, there is a pressing need to identify how intertidal communities di ffer between rock-rubble groynes in urban environs and analogous natural rocky shores in rural areas. Theoretically, rock-rubble groynes provide a conservation dilemma. They o ffer an opportunity for the presence of novel rock habitat that could have beneficial implications for populations of under-pressure intertidal species predominantly in highly-impacted urban environments, ye<sup>t</sup> they increase connectivity between isolated rocky shores, which may support populations of native species or increase the potential for non-native species invasion, or both. With only two studies [21,23] having compared the ecology of natural rocky shores with those of rock-rubble groynes (one of these only as an in-passing comment), and with contrasting results found in a single study area (Sandbanks peninsula, Dorset), the questions of whether rock-rubble groynes support ecological communities similar to natural rocky shores, and whether they represent a conservation opportunity or threat, remain open. Furthermore, we still lack basic knowledge of how diverse and abundant the rock-rubble groyne communities are.

Here we compare the ecological communities of both exposed rock and rock pools between rock-rubble groynes and nearby rocky shores, using a paired sampling design repeated in three locations around the coast of England. We focus on four main questions: do urban rock-rubble groynes and nearby rural rocky shores (hereafter we refer to the two collectively as rocky 'structure types') di ffer in terms of (1) their species richness and (2) their species' abundances? (3) Are there specific communities found on one of the structure types but not the other? (4) Do the rocky structure types di ffer in terms of presence and abundance of species not native to the area (considering native status at both a country and a within-country level)? In addressing these questions, we investigate the role of rock pools as well as rock surfaces and investigate all macro-organisms found, doing parallel analyses for algae, lichens, sessile animals and mobile animals.

#### **2. Materials and Methods**

#### *2.1. Study Area and Data Collection*

We selected three stretches of the English coastline that each contained a rural natural rocky shore and urban (or semi-urban) rock-rubble groynes in reasonably close proximity to each other (Figure 1, Table SI1). This gave us three pairs of study sites (Supplementary Information 1), which we sampled in summer 2008. In each site with artificial structures, we randomly selected three rock-rubble groynes of the same age to survey (Table SI1). To control for variations in tidal shore height, we separated each structure into low-, mid- and high-shore sections (we call this variable 'level') based on the mean low- and high-water spring tides. We did not identify tidal height boundaries by the limits of organisms, as often recommended [42] because both physical and biological components influence the zonation of organisms [43], meaning there is the possibility to introduce considerable cross-site error.

**Figure 1.** Location of study sites around the coast of England. Devon, Dorset, and Norfolk are counties. Ipswich and Scarborough are locations mentioned in the text.

We used a stratified random sampling technique. With 2 structure types (natural, groyne), 3 counties (Devon, Dorset, Norfolk), 3 replicates of each structure type in each county and 3 levels (high, mid, low) per structure, we had 54 sampling sections in total. Due to the uncertain influence of aspect and exposure on intertidal biological assemblages [21,44], we only surveyed the side of the groynes that did not face the dominant longshore current (therefore sheltered from wave action). Each section of the sampling area was divided up into 0.5 m<sup>2</sup> squares to reduce the chance of recording the

same individual twice and selected using a random number generator [42]. On open rock surfaces, we randomly located 12 quadrats in each sampling section, each of 0.25 m × 0.25 m, making 648 rock-surface quadrats in total, and where necessary moved quadrats to avoid sampling very deep crevices or between boulders to allow for consistent surface areas [29]. In each sampling section, we also selected three rock pools (of similar size to the quadrats across structures) in the same way, or all rock pools in the section if fewer than four were present. Ten of the sampling sections (nine of them on groynes) contained no rock pools at all. Overall, we placed quadrats in 123 rock pools: 74 on natural rocky shores and 49 on groynes. In addition, we counted the total number of rock pools present on each entire groyne or equivalent area of natural shore. We did not measure the depth and perimeter of rock pools as is sometimes recommended [42] due to the time constraints associated with the survey, and we acknowledge the limitations of this approach on the analysis in the discussion.

We used a non-destructive sampling method, recording species as present if observed within the quadrat. We recorded species richness as the total number of species in each quadrat. For mobile organisms, we recorded abundance as the count of individuals in each quadrat or pool, and for lichens, algae and sessile animals, we recorded abundance as percentage cover using a grid. We grouped the twelve rock quadrats in each sampling section into four sets of three neighbouring quadrats and grouped the three rock-pool quadrats into one set of three rock-pool quadrats (where we had su fficient numbers), with each data value being the arithmetic mean of these quadrats (e.g. the mean number of animal species A across three quadrats). Thus, each unit in the analysis represented a small section of habitat sampled using three quadrats, rather than a single very small quadrat. This was to allow a better representation of the local community in each unit of analysis and to reduce any variation and uncertainty associated with slight di fferences in pool volume and surface area due to the uneven surface of the study areas. Overall, the data analysed included 216 rock surface samples and 46 rock pool samples.

#### *2.2. Data Analysis*

We used a three-way analysis of variance (ANOVA) to explore the di fferences between rock-rubble groynes and natural rocky shores (the 'structure type' explanatory variable). We also included the following factors to account for their expected influence: 'county' (Devon, Dorset, Norfolk—reflecting the pairing of the sites) and [tidal] 'level' (low, mid, high). We used Levene's test to examine the assumption of homogeneity of variance, and we visually examined the model residuals for patterning and tested them for normality using Kolmogorov–Smirnov tests. These diagnostics caused us to square-root transform the response variable in the analyses of species' abundances. We ran parallel analyses for rock and pool habitats. We then used detrended correspondence analysis (DCA) to identify ecological communities and significant environmental centroids within the full species dataset among sites which had species recorded within them. DCA reveals the dispersion of points in ordination space, which reflects species abundances within sampling sites [45]. Due to the high frequency of rare species in pools, analysis was undertaken for both rock and pool habitats together, but rare species were not down-weighted as sometimes suggested [46] due to minimal di fferences in the results when both habitats (rock and pool) were considered together and the importance of the rarer species to ecological communities on natural shores. We chose DCA over other ordination analyses due to the long gradient lengths [47] and the fact that DCA is based on the underlying unimodal model of species distributions [48]—a key foundation of our research questions. We explored the implementation of another method (nonmetric multidimensional scaling: nMDS), but found that the long computation time, coupled with the lack of model convergence and the impact of rare species, made manually exploring model options and subsequent interpretation of the results overly complex. Moreover, results from the nMDS were largely congruen<sup>t</sup> with those of DCA, with the exception of the extreme impact of rare *Idotea* and *Gammarus* spp (results not shown). Environmental factors were passively projected on the ordination plot. All analysis was implemented in the open-source software R 3.6.2 [49], with the DCA implemented using the *vegan* package [46]; see Supplementary Information 2 for data and Supplementary Information 3 for R code.

We compared recorded presences in our study with species distribution maps from Gibson et al. [50] and the Marine Life Information Network (MarLIN; [51]) to determine whether we recorded species beyond their distributions, as judged by the two sources. Gibson et al. [50] provide a generalised range of each species, while MarLIN [51] maps are generated from published species records and verified sightings. We chose these sources to correspond with the period of data collection, rather than the most up-to-date distributional data.
