1. Macroalgal (seaweed) beds and forests fuel coastal ecosystems and are rapidly reorganising under global change, but quantifying their functional structure still relies on binning species into coarse groups on the assumption that they adequately capture relevant underlying traits.

2. To interrogate this ‘group gambit’, we measured 12 traits relating to competitive dominance and resource economics across 95 macroalgal species collected from the UK and widespread on North-East Atlantic rocky shores. We assessed the amount of trait variation explained by commonly-used traditional groups – (i) two schemes based on gross morphology and anatomy and (ii) two categorisations of vertical space use – and examined species reclassification into post hoc, so-called emergent groups arising from the functional trait dataset. We then offer an alternative, emergent grouping scheme of macroalgal functional diversity.

3. (i) Morphology and anatomy-based groups explained slightly more than a third of multivariate trait expression with considerable group overlap (i.e. low precision) and extensive mismatch with underlying trait expression (i.e. low accuracy). (ii) Categorisations of vertical space use accounted for about a quarter of multivariate trait expression with considerable group overlap. Nonetheless, turf species tended to display attributes of opportunistic forms. (iii) A nine-group emergent scheme provided a highly explanatory and parsimonious alternative to traditional functional groupings.

4. Synthesis: Our analysis using a comprehensive dataset of directly measured functional traits revealed a general mismatch between traditional groups and underlying traits, highlighting the deficiencies of the group gambit in macroalgae. While existing grouping schemes may allow first order approximations, they risk considerable loss of information at the trait and, potentially, ecosystem levels. Instead, we call for further development of a trait-based approach to macroalgal functional ecology to capture unfolding community and ecosystem changes with greater accuracy and generality.

2.1. Sampling

We measured eleven continuous and one categorical functional traits (Table 1) at the individual level across 95 erect intertidal macroalgal species, which spanned a great variety of form and function and hence, traditional functional groups (Table S1). Samples were collected from twelve rocky shores in the UK ranging from very sheltered to very exposed (Table S2): six sites in South Wales, four sites in Orkney (Scotland) and two sites in Cornwall (England). The collected species are commonly found on North-East Atlantic rocky shores (Ar Gall et al., 2016; Araújo, Sousa-Pinto, Bárbara, & Quintino, 2006; Martínez, Viejo, Carreño, & Aranda, 2012; Martins, Hawkins, Thompson, & Jenkins, 2007), and include species restricted to the region (~25% of species), more broadly distributed across multiple temperate regions (~35%), as well as cosmopolitan and non-native species (~40%). Sampling took place from May to September 2013 and 2015-2018 (Table S3). We collected an average of 6 replicates per species, ranging from 1 to 45 (mode and median = 6, S.D. = 7.55; Table S3). Such a large difference in replication was due to the rarity of some species and to our efforts in sampling abundant species across several sites to better capture natural variability. Replicates were sampled more than 2 m apart. Whenever possible, a replicate was made up of a single individual. However, when individuals were too small for the trait measurements, a sufficient quantity was collected for each replicate by pooling several individuals or tufts (Table S3). Whenever distinguishing between individuals was not possible (e.g., for turfs), the samples were collected by isolating tufts (Table S3). Replicates within species were sampled from a variety of microhabitats to better reflect natural variability, but belonged to the same life stage (e.g., the Trailliella intricata stage of Bonnemaisonia hamifera). Samples were kept in seawater in a cooler until brought back to the laboratory. They were then either screened fresh or frozen at -18°C until processed.

2.2. Trait screening

The functional traits measured are hypothesised to capture two fundamental aspects of primary producer variability, (1) the economics spectrum and (2) competitive dominance. We consider multiple indicators (or ‘functional markers’ sensu Garnier et al., 2004) to provide a more integrated estimation of ecological strategy and function. Here, we briefly summarise the ecological significance of the traits with regard to photosynthesis, structural integrity, space use and complexity (Tables 1, S4). The suite of economics-related traits indicates the relative investment in resource acquisition versus resistance to (a)biotic stress and therefore resource conservation, tying in with the r- (‘fast return’) to K- (‘slow-return’) selection continuum (Pianka, 1970). Slow-return primary producers tend to display long lifespans, low maximum photosynthesis and productivity, reduced palatability, and slow decomposition (Littler & Littler, 1980; Smart et al., 2017; Wright et al., 2004). Traits a-g relate to photosynthesis and/or structural integrity, and hence, position on the economics spectrum: (a) Thallus Dry Matter Content (TDMC) is the ratio between dry and wet mass and represents the proportion of structural compounds and water-filled – and therefore mainly photosynthetically active – tissues (Elger & Willby, 2003; Littler & Littler, 1981; Schonbeck & Norton, 1979). (b) Thickness also increases with the amount of structural tissue, providing resistance to physical stress and herbivore grazing (Cappelatti, Mauffrey, & Griffin, 2019; Littler & Littler, 1980; Littler, Taylor, & Littler, 1983); (c) Carbon (C) content and its ratio with (d) Nitrogen (N) content, (e) C:N, more directly quantify recalcitrant structural compounds relative to N-rich photosynthetically-active tissues (Cornelissen et al., 2003; Weykam, Gómez, Wiencke, Iken, & Klöser, 1996). Analogously to Specific Leaf Area (Wilson, Thompson, & Hodgson, 1999), (f) Specific Thallus Area (STA), obtained by dividing surface area by dry mass, captures light- and nutrient- absorbing surfaces and increases with the extent of low density, water-filled, photosynthetically-active tissues relative to recalcitrant, structural compounds (Littler & Littler, 1980). Finally, because macroalgae absorb nutrients through the blades, the surface-area-to-volume ratio, or (g) SA:V, is associated with nutrient acquisition (Littler & Littler, 1980).

Traits h-l are hypothesised to relate to space use and complexity, and hence, competitive dominance. Plant height is a major determinant of competitive dominance (Díaz et al., 2016). Its macroalgal analogue, (h) maximum length, and by extension (i) aspect ratio, or the ratio between maximum length and width, relate to the ability of macroalgae to outcompete surrounding individuals by taking the position of canopy and emerges from a tradeoff between light capture and structural integrity (Carpenter, 1990; Littler & Littler, 1980). The presence of (j) pneumatocysts (i.e. air bladders) is another important predictor of macroalgal competitive dominance through canopy occupancy (Dromgoole, 1981). (k) Branching order, or the degree of branching of a thallus, and (l) the surface-area-to-perimeter ratio (SA:P) relate to three-dimensional complexity and resource acquisition, and hence, both competitive dominance and economics (Steneck & Dethier, 1994; Veiga, Rubal, & Sousa-Pinto, 2014). High complexity allows individuals to maximise light exposure (Stewart & Carpenter, 2003), provides greater nutrient and gas exchange, delays desiccation at low tide (Hay, 1981; Padilla, 1984; Taylor & Hay, 1984) and reduces the impact of herbivory (Padilla, 1984), but increases drag (Starko, Claman, & Martone, 2015). We measured both traits at the whole individual level to capture the complexity of the whole thallus, since all thallus parts, from holdfast to fronds, are important habitats for epibiota and nekton (Teagle et al., 2017).

Large or structurally complex individuals were subsampled, ensuring that all thallus parts were included at representative proportions (Table S3). We measured the surface area and perimeter of partly microscopic species on subsamples under the microscope and proportionally scaled them up to the whole sample (Table S3). The samples were cleaned of epibiota in seawater and rinsed in deionised water for elemental screening. To obtain TDMC, we recorded sample wet and oven-dried mass (g; Ohaus Scout Pro SP402, SPU602 and Pioneer Analytical PA114, Parsippany, NJ, USA). Thickness (mm) was averaged from ten measurements taken haphazardly along the fronds (Digital Micrometers Ltd, DTG03 0.005, DML3032 0.001 mm, Sheffield, UK), avoiding, when applicable, the midrib. Individuals were scanned (Epson Perfection V600, V39, Suwa, Japan) or displayed on a lightbox (MiniSun A1, Manchester, UK) and photographed directly (Pentax K3 digital camera, SMC DA L 18-55 mm, Tokyo, Japan) or via an imaging microscope (Leica S8AP0, Wetzlar, Germany, affixed with GT Vision GXCAM-H3, Sudbury, UK). We measured frond (when differentiated) or whole-individual surface area (mm²) and whole-individual perimeter (mm) using the software ImageJ (Schneider, Rasband, & Eliceiri, 2012), and calculated SA:V (mm² mL^-1), STA (mm² g^-1), and SA:P (mm). Volume (mL) was measured by water displacement. Maximum length (cm) was measured from the base of the holdfast to the tip of the longest blade. Aspect ratio was obtained by dividing maximum length by maximum width (i.e. largest width of a naturally spread out sample). Branching order was measured as the average number of divisions of the main axes of a thallus from its holdfast to the tip of the blades out of 5 measurements taken haphazardly within the sample. To obtain C and N content and C:N, ground samples were weighed with a microbalance (Sartorius CPA2P, 0.000001 g, Göttingen, Germany) and run through an elemental analyser (PDZ Europa 2020 isotope ratio mass spectrometer interfaced with an ANCA GSL elemental analyser and calibrated with acetanilide).

2.3. Categorisation of species into functional groups

We allocated species to the groups defined by Littler and Littler as well as Steneck and Dethier based on a review of the literature (Table S1). The species we screened belonged to five traditional groups: ‘filamentous’, ‘sheet’, ‘coarsely branched’, ‘thick leathery’ and ‘articulate calcareous’ for Littler and Littler’s functional-form model, and five groups defined by Steneck and Dethier (1994), ‘filamentous (S)’, ‘foliose’, ‘corticated’, ‘leathery’ and ‘articulated calcareous’, in increasing order of cortication. Although both schemes contain groups with similar or identical names, they were originally defined using different approaches and are not assumed a priori to be analogous. We also tested Steneck and Dethier’s detailed scheme, which includes two subgroups (Supplementary information, section 1). We used two common categorisations of vertical space use: the binary canopy vs. turf scheme and a three-level canopy/subcanopy/turf scheme adapted from Arenas et al. (2006). Turfs were considered macroalgae with little to no three-dimensional structure (compared with kelp and other canopy-forming macroalgae) that form a dense layer of fine filaments, branches, or plumes on the substratum (Filbee-Dexter & Wernberg, 2018). This broad definition of turf macroalgae allowed categorisation of all species within our study. Vertical structure in the water column is somewhat community-dependent, so we categorised species into the three-level scheme based on what we judged was the most common scenario on the rocky shores screened.

2.4. Creation of the emergent grouping schemes

We performed all analyses in R 3.5.3 (R Core Team, 2019). Species trait averages were transformed to bring their distribution as close to normality as possible and to reduce differences in scale across traits (Table S4). We imputed the four percent of average trait values that were missing from the dataset (function ‘mice’ in eponymous R package; van Buuren & Groothuis-Oudshoorn, 2011) to then reduce the dimensionality of the data using a Principal Coordinate Analysis (PCoA; function ‘cmdscale’). We favoured a PCoA over a Principal Component Analysis because it allowed us to include pneumatocyst presence. The PCoA was run on a weighted Gower matrix (function ‘daisy’ in ‘cluster’; Maechler, Rousseeuw, Struyf, Hubert, & Hornik, 2019), with equal weighting to traits associated with photosynthesis, structural integrity, space use, and complexity (Table S4). We created emergent groups from the weighted Gower matrix using k-medoids clustering (k-medoids), a top-down clustering approach whereby species are assigned to a chosen number of groups based on multivariate distance from group medoids, making it rather robust to noise and outliers (Reynolds, Richards, de la Iglesia, & Rayward-Smith, 2006; using ‘pam’ in package ‘cluster’; Maechler et al., 2019).

Clustering to generate emergent groups provided a tool for generating a more functionally informative alternative to traditional functional groupings. To allow the data to inform not only the assignment of species to groups but also the number of groups, we ran k-medoids while allowing an increasing number of groups (from five upwards), searching for an emergent grouping scheme that maximised overall explanatory power and parsimony while maintaining statistically significant differences (P < 0.05) between all pairs of groups. Finally, to evaluate the robustness of the resulting grouping scheme to intraspecific trait variability, we computed the explanatory power (PERMANOVA; ‘adonis’ in ‘vegan’; Oksanen et al., 2019) of the new grouping across 999 iterations of the species x traits matrix, randomly assigning 1/3 of rows to each of species-level mean, mean + 1 S.E., and mean - 1 S.E. in each iteration.

While we observed a monotonic increase in the extent of trait variation explained as well as parsimony, a nine-group emergent scheme maximised explanatory power and parsimony while maintaining significant differences between all groups. This scheme explained about two thirds of multivariate trait expression (PERMANOVA, R² = 0.69, P < 0.001), and all emergent groups were significantly different from each other (pairwise PERMANOVA, P < 0.05; Table S6). The scheme’s explanatory power remained substantial even when we conducted resampling to allow for intraspecific variability in species’ traits (PERMANOVA, R² = 0.57 ± 0.02 [mean ± S.D.]).