Season‐specific impacts of climate change on canopy‐forming seaweed communities

Abstract Understory assemblages associated with canopy‐forming species such as trees, kelps, and rockweeds should respond strongly to climate stressors due to strong canopy‐understory interactions. Climate change can directly and indirectly modify these assemblages, particularly during more stressful seasons and climate scenarios. However, fully understanding the seasonal impacts of different climate conditions on canopy‐reliant assemblages is difficult due to a continued emphasis on studying single‐species responses to a single future climate scenario during a single season. To examine these emergent effects, we used mesocosm experiments to expose seaweed assemblages associated with the canopy‐forming golden rockweed, Silvetia compressa, to elevated temperature and pCO2 conditions reflecting two projected greenhouse emission scenarios (RCP 2.6 [low] & RCP 4.5 [moderate]). Assemblages were grown in the presence and absence of Silvetia, and in two seasons. Relative to ambient conditions, predicted climate scenarios generally suppressed Silvetia biomass and photosynthetic efficiency. However, these effects varied seasonally—both future scenarios reduced Silvetia biomass in summer, but only the moderate scenario did so in winter. These reductions shifted the assemblage, with more extreme shifts occurring in summer. Contrarily, future scenarios did not shift assemblages within Silvetia Absent treatments, suggesting that climate primarily affected assemblages indirectly through changes in Silvetia. Mesocosm experiments were coupled with a field Silvetia removal experiment to simulate the effects of climate‐mediated Silvetia loss on natural assemblages. Consistent with the mesocosm experiment, Silvetia loss resulted in season‐specific assemblage shifts, with weaker effects observed in winter. Together, our study supports the hypotheses that climate‐mediated changes to canopy‐forming species can indirectly affect the associated assemblage, and that these effects vary seasonally. Such seasonality is important to consider as it may provide periods of recovery when conditions are less stressful, especially if we can reduce the severity of future climate scenarios.

Simulating different future climate scenarios will better model climate change impacts by incorporating different levels of severity.
The Intergovernmental Panel on Climate Change (IPCC) provided several Representative Concentration Pathways (RCP) that predict changes in temperature and ocean pH by the year 2100 relative to present-day levels.For example, RCP 2.6 (+1°C/−0.1 pH units from ambient conditions) represents a low-impact scenario where emissions are stabilized by the 2020s while RCP 4.5 (+2°C/−0.2pH units) represents a moderate scenario where emissions are stabilized by the 2040s (IPCC, 2022).Observing the effects of climate change under multiple scenarios can reveal potential thresholds and offer greater predictability for management and conservation efforts (Thurman et al., 2020).Given (1) the uncertainty in the severity of future climate change, and (2) that small differences in temperature and/or pH can be biologically and ecologically meaningful (Araújo et al., 2018;Harrington et al., 2020;Wang et al., 2015), multiple scenarios need to be considered.
The severity of future climate change impacts on natural ecosystems may vary among seasons, but this variation also remains understudied (Russell et al., 2012).As future climate change continues to shift mean temperature and pH, anticipating how these shifts will vary between seasons is critical for timing-dependent endeavors such as restorative transplanting (Li et al., 2014;Pearce-Higgins et al., 2015;Richardson-Calfee et al., 2004).Well-known climate change alterations to seasonal events such as droughts, marine heatwaves, coastal upwelling, growing periods, and weather whiplash (i.e., rapid fluctuations between extreme conditions; Lee, 2022) have already disrupted phenological cycles and restructured communities across a wide range of ecosystems (Beas-Luna et al., 2020;Bell et al., 2021;Donham et al., 2022;Ernakovich et al., 2014;Ooi et al., 2014).However, much of our understanding of species' responses to climate change derives from studies conducted under static laboratory conditions lacking seasonality (Kroeker et al., 2020).Including seasonal factors in climate change studies will provide clearer insights into performance impacts and mitigating these impacts.
Similar to studies performed under static conditions, population-level studies have provided important insights (e.g., taxa-specific effects of elevated pCO 2 ; Cattano et al., 2018, Fernández et al., 2015, Kim et al., 2020, Lefevre, 2016).However, they may not accurately predict climate change impacts on whole communities because they do not allow for species interactions (Ockendon et al., 2014).Despite the staggering increase in climate change-related research during the past two decades, the ratio of single-species studies to community-level studies remained nearly the same (i.e., single-species studies continue to comprise ~60% of studies in this field, Bass et al., 2021).Additionally, when papers published between 2010 and 2019 were subdivided into those focusing on single species versus species assemblages, single species studies were three times more common (Bass et al., 2021;Wernberg et al., 2012).Reducing this gap and successfully predicting the impacts of climate change on natural populations will require increased efforts toward studying these impacts on natural assemblages.
Because canopy-forming species interact strongly with their understory assemblages, climate change impacts may be particularly strong in communities containing such interactions (Edwards & Connell, 2012).Canopy-forming species form densely branched overhead structures that modify the local environment (Edwards, 1998;Gonzales et al., 2017;Hondolero & Edwards, 2017;Joly et al., 2017;Ørberg et al., 2018).These effects can provide a more favorable habitat for understory species recruiting beneath or to the canopy (Clark et al., 2004;Flukes et al., 2014;Kitao et al., 2018;Roberts & Bracken, 2021).In turn, understory species can affect canopy-forming species by augmenting the recruitment and survival of juvenile of canopy-forming species (Barner et al., 2016;Beckley & Edwards, 2021).Under future climate conditions, these reciprocal interactions will likely alter the response of individual species to climate-related stressors.For example,  (Hirsh et al., 2020).Consequently, the performance of understory species can be directly and/or indirectly affected by climatemediated changes (Edwards & Connell, 2012;Kim et al., 2020;Koch et al., 2013;Ragazzola et al., 2012).Given the complexity of these interactions, community-level approaches should therefore be especially pertinent for these canopy-dependent assemblages.
For this study, we identified a community that might be sensitive to the combined effects of climate change, seasonality, and species interactions.This community consisted of the canopyforming, intertidal, temperate rockweed, Silvetia compressa (henceforth Silvetia), and its understory assemblage.Rockweed canopies transform inhospitable intertidal areas such as exposed boulders into refuges by trapping moisture and stabilizing substrate and water temperature (Bertness et al., 1999).The assemblage associated with these refuges includes fleshy, turfing, and calcifying seaweeds, mobile and sessile intertidal invertebrates, and juvenile subtidal species such as fishes and lobsters (Sapper & Murray, 2003;Vercaemer et al., 2018).These understory species enhance primary productivity (Tait & Schiel, 2018), provide settlement cues and substrate for commercially important invertebrate larvae (Morse & Morse, 1984), and feed higher trophic levels (Ellis et al., 2007).Such interactions and corresponding services could be heavily altered should Silvetia populations decline.Recently, Silvetia declines have co-occurred with ocean warming associated with the 2015-16 El Niño (Graham et al., 2018;MARINe, 2023).
Future climate conditions resulting in similar levels of warming but across a prolonged period would likely exacerbate the decline of Silvetia communities.
Intertidal communities like this one might be particularly prone to season-specific impacts of climate change for several reasons.
First, future summers could become more stressful than other seasons in these habitats because intertidal organisms often live near their thermal maxima (Madeira et al., 2012).Second, tidal ranges are often season-specific such that intertidal organisms will encounter more extreme conditions during these seasons (Erftemeijer & Herman, 1994;Flick, 2016).Third, seasonality may interact with climate in intertidal systems because the reproduction, dispersal, and recruitment of marine species are often seasonal (Ådahl et al., 2006;Edwards, 2022).
To understand the impacts of multiple climate change scenarios on Silvetia communities, we used mesocosms to expose Silvetia and its understory assemblages to three levels of ocean change conditions (Ambient,RCP 2.6,and RCP 4.5).These experiments also manipulated Silvetia presence to distinguish between the direct and indirect effects of climate change on the dominant understory species.We repeated this experiment in the summer and winter to assess seasonal variation in these effects.Because future climate scenarios were expected to suppress Silvetia growth, we also conducted field manipulations of Silvetia to understand the consequences of canopy loss on natural understory assemblages at two levels of understory biomass.

| Study site
We selected sites adjacent to two southern California (Point Loma, San Diego, Figure 1) long-term monitoring sites; Navy South (32.68306°N, −117.24963°W;hereafter NASO) and Navy North (32.692312°N, −117.25297°W;hereafter NANO).These Multi-Agency Rocky Intertidal Network sites (MARINe) contain dense patches of Silvetia, perhaps because of the rarity of some stressors such as trampling (Denis, 2003;Tydlaska & Edwards, 2022) and runoff (Whitaker et al., 2010).We surveyed the Silvetia assemblages, collected the algae and grazers used in the mesocosm experiment, and conducted the field experiment at NASO.We added NANO as a secondary collection site to minimize the impact on algae and grazer populations at either site.Collecting procedures were identical between both sites.Silvetia at both sites grows on emergent substrata at intertidal elevations between 0 and 1 m above Mean Lower Low Water (hereafter MLLW).Average water temperatures at these sites are ~18°C and maximum summer water temperatures reach ~24°C (SeaTemperatures, 2023).

| Mesocosm experiment
To examine the impacts of projected changes in ocean temperature and pH on Silvetia assemblages, we conducted a mesocosm experiment at San Diego State University's Coastal and Marine Institute (CMI, Figure 2a) that exposed the assemblages to three ocean climate conditions (Ambient, RCP 2.6, RCP 4.5).Ambient conditions represent current levels of temperature and pH.RCP 2.6 is a global emissions pathway representing low levels of climate change that will be experienced in the year 2100 (in line with the theoretical stabilization of global emissions by ~2020 leading to an average change of +1°C/−0.1 pH units on global oceans; IPCC, 2022).RCP 4.5 represents moderate levels of climate change (+2°C/−0.2pH units).
Importantly, our experiments used flow-through seawater, which allowed for natural variation in ambient conditions.Thus, our future scenarios that manipulated pH and temperature relative to ambient conditions also experienced such variation.
Each mesocosm consisted of a clear polypropylene plastic box (15 × 15 × 7.6 cm; l × w × h) that had three 5-cm diameter holes in each of two opposite sides.Window screen mesh covered these holes and the box tops to retain box contents and allow water exchange.We crossed climate scenario (Ambient, RCP 2.6, RCP 4.5) with Silvetia canopy (Present, Absent) treatments.Replicate mesocosms (n = 10) were randomly assigned to three outdoor water tables (1.8 × 0.9 × 0.3 m; l × w × h) that received flow-through seawater from San Diego Bay.Each water table was then randomly assigned to one of the three climate scenarios.Because water temperatures in San Diego Bay are warmer than the average water temperatures at rocky shores where Silvetia occurs, we chilled the incoming seawater using a flow-through seawater chiller (Aqualogic, Inc.) but still allowed the temperature to vary with natural ambient fluctuations (Figure S1).Seawater delivered to the future scenario mesocosms (i.e., RCP 2.6, RCP 4.5) was then altered in a header tank using aquarium heaters and CO 2 injections to match the projected temperature and pH values of the respective scenarios before entering experimental mesocosms.Seawater was delivered to each header tank at 2270 L/h, which then flowed via gravity to the experimental mesocosms.To create realistic tidal conditions, ball valves connected to drains were opened and closed using a digital watering timer (DIG Model C002, DIG Corporation), which resulted in the mesocosms being submerged at tide heights 0.5 m above MLLW, and emerged at tide heights below this.This tide height is representative of intertidal elevations where Silvetia occurs in southern California (Littler, 1980).
We added realistic assemblages of understory algae to each mesocosm.To determine the species that comprised these representative assemblages, we surveyed natural understory algal communities in the field at NASO.We collected, identified, and weighed all the understory algae found within six 0.15 × 0.15 m quadrats that were placed beneath haphazardly selected Silvetia individuals.This identified five genera that made up 83% of the total understory algal biomass; namely Chondracanthus, Centroceras, Corallina, Gelidium, and Laurencia (Figure 2b).Because we were unable to find enough Gelidium during future collections for our experiment, we removed it from the study.The remaining four genera made up 76% of total understory biomass.To create realistic understory assemblages, we calculated the biomass density of each genera in the field (grams per m 2 ) and scaled these calculations to match the surface area of the mesocosm floors.
Using this approach, each mesocosm received 4 g of Centroceras, 10 g of Chondracanthus, 9 g of Corallina, and 2.5 g of Laurencia.
Additionally, because invertebrate grazers can alter algal-algal interactions such as by controlling epiphyte growth (Hoffmann et al., 2020;Rogers & Breen, 1983), they were included in all mesocosms.To add ecologically realistic densities of these grazers relative to Silvetia biomass, we scaled field densities (# of grazer individuals per gram of Silvetia) reported in a previous study (Jones, 2016) to our mesocosms.As a result, we added six Tegula funebralis, six Lottia strigatella, six Lottia scabra, 10 Littorina scutulata, and one Cyanoplax hartwegii to each mesocosm.
Grazers and understory algae were collected during the establishment of the field experiment (see below) and held at the CMI for a 10-day acclimation period.Each of the four understory seaweeds was weighed to the predetermined biomass (±5%), attached to a rock with superglue, and placed into one of the four corners of the mesocosms.Additional rocks covered the bottom of the mesocosms to provide a refuge for grazers.We alternated the position of each algal type between replicates in a Latin Square design.For the Silvetia Present treatments, pre-weighed Silvetia (72.0 ± 5% g) were laid across the assemblage inside mesocosm containers.Actual average Silvetia weight was 72.5 ± 1.3 g (mean ± SE).Once assembled, the mesocosms were placed into the water tables containing ambient seawater.Header tanks containing treated seawater would then alter the pH and temperature of the experimental tables to match treatment values over the course of approximately 1 day.During our 42-day experiment (August 5th-September 17th, 2021), we measured the pH and temperature of the seawater as it flowed from each header tank into the experimental mesocosms every morning using a probe (Oakton 300 Series pH/DO meter), except on days 32, 38, and 39, which were not measured due to logistical constraints.
After 42 days, we ended this experiment as most of the understory algae in the Silvetia Absent treatments had bleached or disintegrated (Figure 2c).We categorized the algae tissue as being either bleached (dead) or unbleached (living) and measured the biomass of each group in each replicate after blotting them dry.After separating the bleached tissue, the remaining biomass of each understory genus was then calculated as the percentage of final unbleached tissue weight relative to its initial weight.To assess Silvetia health, we measured quantum yield (a ratio of variable fluorescence [Fv] to maximal fluorescence [Fm]), which estimates the light-harvesting efficiency of photosystem II (PS II), using a pulse amplitude modulated (PAM) fluorometer (sensu Bews et al., 2021;Edwards & Kim, 2010).
Because we observed within-individual variation in tissue health, we measured the quantum yield of each individual at five randomly selected sections of each thallus and averaged these measurements for each Silvetia replicate.
To understand seasonal differences in how the Silvetia assemblage responded to climate change, we repeated this experiment in the winter (November 9-December 20, 2021).We followed the same protocols described above but made three changes: (1) We shortened the acclimation period from 10 to 5 days, (2) replaced all algae and grazers with newly collected individuals from a nearby site (NANO instead of NASO), and (3) the water tables were randomly reassigned different climate treatments.Pre-weighed Silvetia for this experiment averaged 71.0 ± 1.6 g.Although we did not see as much understory degradation in the Silvetia Absent treatments during this experiment, we maintained the 42-day experimental duration to facilitate comparisons between the two trials (hereafter summer and winter).single "Silvetia Present" treatment (n = 20) and compared this pooled treatment to the Silvetia Canopy None (henceforth "Silvetia Absent") treatment; n = 10.Because we were unable to relocate some plots during subsequent surveys, our final samples varied from 14 to 16 and 6 to 8 for Silvetia Present and Silvetia Absent treatments, respectively.To manipulate the understory assemblages, the existing assemblages in half of the plots of each Silvetia treatment were removed using scrapers and chisels (Understory Cleared treatments) while the assemblages in the other half were left unmanipulated (Understory Full treatments).We measured the percent cover of the understory assemblages in October (hereafter fall) and December (hereafter winter) 2021.

| Statistical analyses
All data were analyzed using R-Studio and Primer + PERMANOVA 7.
Prior to analyses, data were checked for normality and heteroscedasticity using Shapiro-Wilk's and Levene's tests, respectively.
For the mesocosm experiment, measurements of quantum yield required square-root transformation to meet assumptions of normality.Silvetia biomass and measurements of quantum yield within the mesocosms were compared among the three climate treatments using separate one-way ANOVAs (for each season).This was done as separate analyses rather than a two-way ANOVA that included season as a factor because the experimental mesocosms were broken down, cleaned, randomized, and reassigned with new assemblages prior to the winter trial.Tukey's HSD post hoc tests between pairs of climate treatments were then used when the ANOVAs returned significant differences.To visualize shifts in the understory algal assemblages between the Climate and Silvetia canopy treatments within each trial, Principal Coordinates Analysis (PCoA) based on Bray-Curtis dissimilarity matrices was used to map similarities in the algae comprising each assemblage.Two-way PERMANOVAs were then used to determine if the assemblage shifts differed between the Climate and Silvetia canopy treatments.Due to a high number of zeroes for certain taxa in the Silvetia Absent treatments, the data were square-root transformed and the PERMANOVAs were run with a zero-inflated Bray-Curtis similarity indices using a dummy variable of 1 (Clarke et al., 2006;Smith, 2017).A priori post-hoc permutation tests were then used to examine pairwise differences in the assemblages between Climate and Silvetia canopy treatments.SIMPER analyses were used to identify the relative contribution of each understory taxon to assemblage dissimilarity between treatments.As discussed above, these analyses were run separately for the summer and winter trials.
For the field experiment, a three-way PERMANOVA was used to assess differences in the understory communities (based on percent cover) between Silvetia canopy treatments, Understory treatments, and Seasons.Unlike the mesocosm experiments, the season was included as a factor because the field experiment was run continuously.Due to consolidating Silvetia High and Low plots into a single treatment, our experimental design was unbalanced, and this was compounded by the loss of plots due to storms.PERMANOVA is the most robust test under this design, but because it still loses reliability with increasing heterogeneity, we acknowledge the potential statistical error for this analysis (Anderson & Walsh, 2013).Following the PERMANOVA, a priori permutation post-hoc tests were used to determine differences in understory assemblages between the Silvetia canopy treatments within each Understory treatment and season.SIMPER analyses were used to determine the percent contribution of each general to the observed differences.All analyses were evaluated at an α-level of .05.

| Mesocosm conditions
Seawater conditions within the climate treatments representing future climate scenarios approximated the desired target values for temperature and pH of +1°C and -0.1 pH units (RCP 2.6), and +2°C and -0.2 pH units (RCP 4.5) relative to Ambient conditions.In summer, average RCP 2.6 parameters measured +1.2°C and -0.16 pH units while average RCP 4.5 parameters measured +1.8°C and -0.25 pH units.In winter, average RCP 2.6 parameters measured +0.6°C and -0.14 pH units while average RCP 4.5 parameters measured +1.8°C and -0.22 pH units (Figure S1, Table S1).On average, all three treatments varied with natural ambient fluctuations and were warmer and more acidic during the summer trial than during the winter trial.
In contrast, Silvetia biomass increased significantly under Ambient (p = .004)and RCP 2.6 (p < .001)conditions relative to its starting biomass during the winter trial but did not change under RCP 4.5 conditions (p = .467).Consequently, biomass under Ambient and RCP 2.6 conditions remained similar to one another (p = .828)in the Ambient and RCP 2.6 mesocosms but were both higher than in the RCP 4.5 mesocosms (p < .001for both).Overall, final Silvetia biomass was higher in the winter trial (74.2 ± 5.6 g, mean ± SE) relative to the summer trial (37.8 ± 11.8 g, mean ± SE).When comparing the same climate treatments across seasons (e.g., Ambient in summer to Ambient in winter), the biomass of all three winter treatments was higher than those of the summer treatments (Figure 3).

| Silvetia quantum yield
Similar to biomass, Silvetia quantum yield varied among the climate treatments in both summer (ANOVA: F 2,27 = 6.5, p = .005)and winter (F 2,27 = 0.5, p = .635,Figure 4, Table S3), but this appeared to differ between the two seasons.Specifically, quantum yield was generally higher in the winter (0.63 ± 0.05 Φ PSII , mean ± SE) than in the summer (0.46 ± 0.12 Φ PSII , mean ± SE).In summer, Silvetia quantum yield varied among climate change treatments and was significantly lower under RCP 4.5 conditions relative to Ambient conditions (Tukey's: p = .004),but otherwise, it did not differ between Ambient and RCP 2.6 conditions (p = .302)or between RCP 2.6 versus RCP 4.5 conditions (p = .115).In contrast, the quantum yield did not vary among the climate treatments in the winter trial (ANOVA: F 2,27 = 0.5, p = .635),though the quantum yield of every winter climate treatment was higher than the summer counterpart (Figure 4).
Specifically, during the summer trial when the biomass of the Silvetia  S5).However, the understory assemblages were not different between RCP 2.6 and RCP 4.5 (p = .583).In contrast, no shifts in the understory assemblages occurred under either climate change scenario relative to Ambient in the absence of the Silvetia canopies (Pairwise tests: p = .165& .420for comparisons of RCP 2.6 & RCP 4.5 to Ambient, respectively, Table S5), and the understory assemblages were again not different between RCP 2.6 and RCP 4.5 (p = .460,Figure 5b).Similarly, understory shifts during the winter trial occurred in the presence of a Silvetia canopy, but only under RCP 4.5 conditions (i.e., when Silvetia biomass was lower than it was in Ambient & RCP 2.6 treatments, Pairwise tests: p = .001for both, Figure 5c).In the absence of a canopy, like the summer trial, there were no differences between understory assemblages of either climate change scenario relative to Ambient (Pairwise tests: p = .831& .065for comparisons of RCP 2.6 & RCP 4.5 to Ambient, respectively, Table S5) or between RCP 2.6 and RCP 4.5 (p = .655).
When future climate scenarios shifted the understory assemblage beneath a Silvetia canopy compared to the Ambient treatment (i.e., RCP 2.6 & RCP 4.5 in summer, and RCP 4.5 in winter), we observed the same ranking in taxa with respect to their contribution to this dissimilarity.From most important to least important, this ranking was the same in these three comparisons-Centroceras, Corallina, Chondracanthus, Laurencia (Table 1, taxa rankings for Silvetia Absent treatments can be found in Table S6).
Overall, the top two genera (Centroceras and Corallina) consistently declined under future climate scenarios relative to Ambient climates in these three treatments (RCP 2.6 & RCP 4.5 in summer, and RCP 4.5 in winter, Figure S2).The other two genera showed more variable responses to these treatments.For example, Chondracanthus decreased in two of these treatments (summer, RCP 2.6 and winter, RCP 4.5) but increased in another (summer, RCP 4.5).
Similarly, Laurencia decreased in one of these treatments (winter, RCP 4.5) but increased in two other treatments (summer, RCP 2.6 and summer, RCP 4.5).

| Field assemblage
The effect of Silvetia removal on the understory assemblages in our field plots depended upon season and the initial state of the understory assemblages (PERMANOVA: pseudo-F 1,84 = 3.6, p = .002,F I G U R E 5 Principal Coordinates Analysis (PCoA) plots depicting the change in the understory assemblage across Climate and Canopy treatments for the mesocosm experiment.The assemblage composition of (a) Silvetia Canopy Present treatments during the summer, (b) Silvetia Canopy Absent treatments during the summer, (c) Silvetia Canopy Present treatments during the winter, and (d) Silvetia Canopy Absent treatments during the winter.Assemblage composition of each plot was generated using the biomass of each algal genera recovered at the end of each trial.
Importantly, the top two species that responded to Silvetia loss were Centroceras and Corallina were the same top two species that were impacted by the climate manipulations in our mesocosm experiment.
As a caveat to these results, prior to initiating the field manipulations in summer, a priori testing revealed that assemblages differed between Understory Full treatments (p = .009,Tables S8 and   S9, Figure S3).Thus, the effect of Silvetia removal on fall, Understory Full assemblages could be confounded with the starting state of the assemblages.However, because the starting patterns of some genera were not consistently maintained across every season (e.g., Corallina had similar starting abundances in summer but decreased in the absence of a canopy in fall, Figure S4), it is likely that the differences found during subsequent sampling resulted from manipulating the canopy and understory rather than a holdover from the starting state of the assemblage.

| DISCUSS ION
Realistic assemblages of the intertidal canopy-forming rockweed, Assemblage composition of each plot was generated using the percent cover of each algal genera.
Silvetia similarly shifted the understory community, but only in the fall when the understory was intact.
Season-specific impacts of climate change (e.g., in the mesocosm trials, RCP 2.6 suppressed Silvetia and shifted the understory during the summer trial but not in the winter trial) suggest that seasonal factors may determine how climate change affects intertidal algal communities.This phenomenon has been observed for various taxa such as insects (Johansson et al., 2020), plants (Gordo & Sanz, 2010), and migratory animals (Robinson et al., 2009) when climate changeinduced warming intersects with critical, season-dependent phenological periods such as mating, flowering, or migration.With Silvetia, climate change may exacerbate mortality during the summer when it encounters temperatures near its thermal maximum, which may then reduce reproduction and recruitment in the winter (Moeller, 2002).In support of this hypothesis, Silvetia only grew in our mesocosms during the winter trial when photosynthetic quantum yields were higher and abiotic conditions were more benign.In summer, relative to winter, seawater pH was lower, seawater temperatures and irradiances were higher, and peak irradiance coincided more frequently with periods of low tide (an effect likely to be more pronounced in areas with larger seasonal and tidal variation; McLachlan et al., 1996), all of which may have suppressed Silvetia Taxa resistant to direct effects of climate change may be susceptible to indirect effects via changes to canopy-forming species (Edwards & Connell, 2012).For example, ocean acidification and warming can negatively affect canopy-forming species (Brown et al., 2014;Shukla & Edwards, 2017) but often do not directly impact turfing algae, such as Centroceras (Christie et al., 2019;Ober et al., 2016).Consistent with this finding, Centroceras increased under future climate scenarios relative to Ambient in the absence of Silvetia during the winter mesocosm trial.In summer, however, TA B L E 2 Cumulative percent contribution for each algal genera, generated by SIMPER, to significant assemblage dissimilarity in the field experiment between Silvetia Present and Silvetia Absent treatments.Centroceras required a Silvetia canopy for survival regardless of climate treatment, potentially because ambient abiotic conditions (e.g., temperature) were too stressful during this season (Figure S1).This demonstrates how the climate-mediated loss of canopy-forming species may impair members of the understory assemblage which are otherwise resistant to the direct effects of climate change and that this interaction may only occur seasonally.
Understory seaweeds that are sensitive to direct impacts of ocean acidification, such as calcifying taxa like Corallina, may be particularly prone to climate change because of both direct (Kim et al., 2020) and indirect effects.Ocean acidification can directly reduce the growth and performance of calcifying seaweeds, in part because of reductions in calcification rates (Cornwall et al., 2022).
Ocean acidification can also indirectly affect these understory species by reducing the cover provided by canopy-forming species, thereby increasing desiccation, photoinhibition, and pH stress (Fales & Smith, 2022;Irving et al., 2004;Schmidt et al., 2011).Although we are unable to parse out all these effects here, the trend for Corallina loss under future climate scenarios in the presence of Silvetia (that occurred in both seasons) and a weak or lack of a trend in the absence of Silvetia suggest that some of the Corallina declines were indirect effects of canopy loss unrelated to an increase in photic or desiccation stress.
The effects of climate on fleshy algae such as Laurencia and Chondracanthus followed different patterns relative to turf and calcifying algae.For example, during the winter trial, Chondracanthus and Laurencia both exhibited declines under future climate scenarios relative to Ambient when without a canopy, while Centroceras increased under these conditions.This decline in Laurencia and Chondracanthus could have resulted from a lower thermal tolerance threshold, the lack of a biomechanism to utilize high concentrations of CO 2 such as carbonic anhydrase, or a heavier reliance on canopies for physical and chemical amelioration (Hirsh et al., 2020;Jueterbock et al., 2013;Kim et al., 2016).These patterns, potentially driven by physiological differences and species interactions, indicate a differing response between seaweed functional groups to canopies, seasonality, and the interaction of these factors with climate change.
Under natural field conditions, assemblages also shifted in response to Silvetia loss depending on the season and successional stage.In the fall, the assemblages in the Understory Cleared plots did not differ between Silvetia Present versus Absent treatments, indicating a lack of reliance on Silvetia canopies by early successional species, which are generally robust to abiotic stressors (Table S12 , Farrell, 1991;Sousa, 1979).The mature assemblage of Understory Full plots, however, had diverged between Silvetia treatments and the effect of Silvetia canopies on these assemblages had similarities to the mesocosm experiment (Table 2).For example, in this survey and the winter mesocosm trial, Corallina declined in the absence of a canopy while Centroceras increased.
When resurveyed 2 months later in winter, the mature assemblages had homogenized, perhaps due to the recovery of species sensitive to Silvetia loss following cooler conditions (Cheung-Wong et al., 2022).The assemblages within Understory Cleared plots, however, had now shifted between Silvetia treatments, possibly because late-successional stage species, such as Gelidium and Gigartina, which are better competitors for space but are also reliant on canopies at higher elevations, had developed (Sousa, 1979).
However, because bare rock was the primary contributor of dissimilarity, this shift may have also resulted from unrelated factors (e.g., stochastic scouring during winter storms).Regardless, if Silvetia cover declines under future climate conditions as seen in our mesocosm experiment, shifts in natural assemblages, such as those observed in our field experiment, will likely occur.
Climate change-mediated shifts in the Silvetia assemblage will ultimately reduce or restructure intertidal communities, altering individual fitness, species interactions, and ecosystem services (Kroeker et al., 2020).Declines of Silvetia alone will lead to loss of nursery habitats for subtidal species during periods of submergence (Schmidt et al., 2011;Vercaemer et al., 2018) and a lack of refuge for mobile and sessile intertidal species during periods of emergence (Sapper & Murray, 2003).Loss of canopy-forming seaweeds can also result in reduced primary production (Edwards et al., 2020;Spector & Edwards, 2020;Sullaway & Edwards, 2020), especially in the upper-mid intertidal zone (Vadas Sr et al., 2004).
Indirect effects of canopy loss will include the reduction of available habitat for understory species facilitated by canopies as well as the ecosystem services they provide (Fales & Smith, 2022).For example, future climate scenarios in our mesocosms led to decreases in Corallina.Because Corallina provides settlement cues and substrate for invertebrate larvae (Morse & Morse, 1984;Seabra et al., 2019), climate change may reduce invertebrate recruitment via changes to Silvetia and Corallina.Additionally, if the understory also facilitates a canopy-forming species (e.g., by providing a hospitable surface for the settlement of canopy-forming recruits), then climate-mediated canopy loss may lead to feedback loops, causing further canopy declines and exacerbating disruption at the community level.

| CON CLUS ION
Our experiments considered (1) realistic assemblages that allowed

CO N FLI C T O F I NTE R E S T S TATE M E NT
The authors declare that there are no conflicts of interest.

O PE N R E S E A RCH BA D G E S
This article has earned an Open Data badge for making publicly available the digitally-shareable data necessary to reproduce the reported results.The data is available at http:// doi.org/ 10. 5061/ dryad.05qft tf8b.
Applied ecology, Biodiversity ecology, Community ecology, Conservation ecology, Ecosystem ecology, Global change ecology, Population ecology the canopy of giant kelp, Macrocystis pyrifera, may reduce the effects of climate change on understory species by absorbing excess CO 2

F
Map of Point Loma, San Diego (Google Maps, 2023).Algae and grazers were collected from Navy North and Navy South.Mesocosm experiments were conducted at the Coastal and Marine Institute.Field experiment plots were established at Navy South.F I G U R E 2 (a) Header tank and water table setup for the mesocosm experiment prior to removal of Silvetia canopies from the Silvetia Absent treatments and attachment of mesh lids.(b) An example of understory algae arrangement (i.Centroceras, ii.Chondracanthus, iii.Laurencia, iv.Corallina, v. Gelidium) within individual mesocosms.Note that due to availability, Gelidium was removed from the mesocosm experiment.(c) An example of understory algae bleaching over the course of the mesocosm experiment.
Experimental field plots were established at NASO to simulate the effect of climate change-mediated loss of Silvetia on its understory assemblage.Because the effect of canopy loss on the assemblage could depend upon the successional stage of the assemblage, we also manipulated the assemblage biomass of the understory by clearing half of the plots at the start of the experiment.We crossed Silvetia Canopy (High, Partial, None) with the initial state of the Understory (Full, Cleared); n = 10.We established these plots in the summer (July 2021) because we hypothesized that the effects of Silvetia loss should be most pronounced during the less favorable summer conditions.Plots containing Silvetia (0.15 × 0.15 m) were marked at their corners with Z-spar Splash Zone epoxy and were randomly assigned to the different treatments.Plots were positioned just below the existing Silvetia holdfasts to study the understory species beneath where the Silvetia canopy drapes over the substrate during low tide.Prior to manipulations, we recorded the percent cover of each genus within the plots using 25-point intercepts within 0.15 × 0.15 m quadrats.Plots assigned to the Silvetia Canopy None treatments simulated the effects of climate change-mediated loss of Silvetia by trimming Silvetia to its holdfast using shears.This allowed the thallus to eventually regrow, while still subjecting the assemblage to any effects associated with an absent canopy for the duration of the experiment.In previous mesocosm experiments, future climate conditions caused Silvetia to discolor, shrivel, and lose biomass across its entire thalli (J.D. Long, 2015, unpublished data).To examine the consequences of partial Silvetia loss, we trimmed Silvetia in Partial Canopy treatments from multiple layers originating from a single holdfast to a single thallus layer.The remaining plots containing Silvetia were left unmanipulated and represented our High Canopy treatments.However, because (1) we observed large within-treatment variation and (2) the Silvetia Canopy High and Partial treatments provided similar canopies, we pooled High and Partial Silvetia treatments into a

F
Box and whisker plots representing (a) initial Silvetia biomass and final Silvetia biomass between Climate treatments for the (b) summer trial and (c) winter trial of the mesocosm experiment.Horizontal lines and asterisks between Climate treatments denote levels of significant differences within each trial (NS, not significant).F I G U R E 4 Box and whisker plots representing the final Silvetia quantum yield between Climate treatments for the (a) summer trial and (b) winter trial of the mesocosm experiment.Horizontal lines and asterisks between Climate treatments denote levels of significant differences within each trial (NS, not significant).canopy was reduced under future climate scenarios, the understory communities shifted relative to Ambient (Pairwise tests: p = .012& .013for comparisons of RCP 2.6 & RCP 4.5 to Ambient, respectively, Figure 5a, Table Silvetia, and its understory, exhibited season-specific responses to ocean climate change.Future climate scenarios similar to those projected by the IPCC suppressed Silvetia growth, reduced Silvetia photosynthetic efficiency (measured by quantum yield), and shifted the understory seaweed communities.These effects, however, were season-specific; both future climate scenarios (RCP 2.6 & 4.5) indirectly influenced the understory by reducing Silvetia cover in summer, but only the more severe scenario (RCP 4.5) produced the same effect in the winter.Similarly, future climate reduced Silvetia photosynthetic efficiency in the summer but not the winter.The summertime reductions in Silvetia cover under future climate scenarios were then associated with shifts in the understory communities.Specifically, future climate scenarios reduced Centroceras and Corallina cover but had season-specific impacts on Chondracanthus and Laurencia (e.g., Chondracanthus increased in summer but decreased in winter).Field removals of F I G U R E 6 Principal Coordinates Analysis (PCoA) plots depicting shifts in the understory assemblage between Silvetia Canopy and Understory treatments for the field experiment based on surveys conducted in fall and winter.The assemblage composition for (a) fall, Understory Full treatments, (b) fall, Understory Cleared treatments, (c) winter, Understory Full treatments, and (d) winter, Understory Cleared treatments.
biomass during the summer trial.The season-specific impacts of RCP 2.6 versus the consistent impacts of RCP 4.5 on Silvetia suggests the potential for recovery from climate change effects if less intense climate change scenarios are realized.For example, Silvetia encountering biomass loss under RCP 2.6 conditions in the summer may be able to recover in the winter, though whether other processes such as reproduction will also recover remains untested.In support of this hypothesis, we only observed Silvetia growth in the winter trial when Silvetia was exposed to Ambient and RCP 2.6 climates.The realization of RCP 2.6, which hinges on extensive and immediate mitigation of greenhouse emissions, is unlikely given current trends while RCP 4.5, which calls for substantial mitigation efforts by the year 2040, appears more realistic.Consequently, the potential for recovery from climate change during at least parts of the year may be rapidly waning.However, because Silvetia individuals were replaced between trials, it is unclear if Silvetia is capable of net growth, or perhaps longer-term acclimation when it experiences future climate conditions through consecutive seasons.More comprehensive conclusions would be drawn from experiments assessing year-round climate change impacts on the same individuals of Silvetia.
for species interactions and indirect climate effects, (2) multiple future climate scenarios, and (3) seasonality.Using realistic assemblages revealed that climate change affected understory assemblages largely via indirect interactions with a canopy-forming species.Including multiple future climate scenarios highlighted gradients in the response of Silvetia assemblages to increasing climate severity.Lastly, repeating our mesocosm experiment and conducting field surveys during two time periods allowed us to assess the interaction between climate change and season.Canopy-understory interactions shape multiple communities outside of rocky intertidal habitats and it is likely for all three of the factors we tested in this experiment to be relevant for those communities.Incorporating realistic assemblages, climate scenarios, and seasonality will ultimately help better inform how important species and communities respond to climate change.AUTH O R CO NTR I B UTI O N S Anthony T. Truong: Conceptualization (lead); data curation (lead); formal analysis (lead); investigation (lead); methodology (lead); project administration (lead); resources (lead); supervision (lead); writing -original draft (lead); writing -review and editing (lead).Matthew S. Edwards: Formal analysis (supporting); methodology (supporting); resources (supporting); validation (supporting); writing -review and editing (supporting).Jeremy D. Long: Conceptualization (supporting); funding acquisition (lead); investigation (supporting); methodology (supporting); project administration (supporting); resources (supporting); supervision (supporting); validation (supporting); writing -review and editing (supporting).ACK N OWLED G M ENTS This work was supported by grants received from the National Parks Service (P20AC00867) and San Diego State University.Thank you to members of the Navy Marine Ecology Consortium for coordinating access to the field sites used in this study (J.Curran and S. Graham) and for providing invaluable field assistance (B.Hong, K. Osterkamp, E. Pollard, C. Salles, and B. Saunders).We are grateful for the additional assistance provided by K. Arzuyan, L. Booher, R. DeSantiago, G. Kalbach, J. Patzlaff, A. Puelicher, M. Sato, and V. Van Deusen.Aid with mesocosm design and equipment was provided by R. Angwin, R. DeSantiago, M. Edwards, and T. Polizzi.Assistance with statistical analysis and data visualization was provided by M. Edwards, L. Pandori, and L. Strope.Comments from M. Edwards and A. Levine greatly improved the quality of the manuscript.This project was conceived, designed, and performed by A. Truong and J. Long.This manuscript was written by A. Truong, M. Edwards, and J. Long.This is Contribution No. 86 of the Coastal and Marine Institute, San Diego State University.

summer, canopy present) Ambient versus RCP 4.5 (winter, canopy present) Average dissimilarity: 0.26 Average dissimilarity: 0.33 Genus Ambient RCP 4.5 Av.Diss SD Ratio Contrib. % Cum. % p
Left column: Summer trial.Right column: Winter trial.The average ratio of final recovered biomass to initial biomass of each algal genera was used for the analysis of dissimilarity.

Canopy present versus canopy absent (fall, understory full) Canopy present versus canopy absent (winter, understory cleared) Average dissimilarity: 0.50 Average dissimilarity: 0.44
Left column: Fall, Understory Full treatments.Right column: Winter, Understory Cleared treatments.Percent cover was used for the analysis of dissimilarity.The table only includes genera that cumulatively contribute >70% dissimilarity.