Coral assemblages at higher latitudes favor short‐term potential over long‐term performance

Abstract The persistent exposure of coral assemblages to more variable abiotic regimes is assumed to augment their resilience to future climatic variability. Yet, while the determinants of coral population resilience across species remain unknown, we are unable to predict the winners and losers across reef ecosystems exposed to increasingly variable conditions. Using annual surveys of 3171 coral individuals across Australia and Japan (2016–2019), we explore spatial variation across the short‐ and long‐term dynamics of competitive, stress‐tolerant, and weedy assemblages to evaluate how abiotic variability mediates the structural composition of coral assemblages. We illustrate how, by promoting short‐term potential over long‐term performance, coral assemblages can reduce their vulnerability to stochastic environments. However, compared to stress‐tolerant, and weedy assemblages, competitive coral taxa display a reduced capacity for elevating their short‐term potential. Accordingly, future climatic shifts threaten the structural complexity of coral assemblages in variable environments, emulating the degradation expected across global tropical reefs.

Following plot set up, annual repeated surveys of all tagged colonies, up to and including 2019, then allowed us to estimate size-specific patterns in colony survival, size transitions (growth & shrinkage [Madin et al., 2020]), fragmentation, and recruitment.
Photographs, taken parallel to the benthic surface with a scale bar included for reference, were used to capture the visible horizontal extent of each tagged colony over successive surveys.Using the program ImageJ (Schneider et al. 2012), we traced along the colony edges in each photograph to obtain longitudinal records of horizontal surface area (cm 2 ) measurements for each colony.All colony size estimates were then log-transformed to ensure a normal distribution and enhance the resolution of smaller colonies.Next, pooling data across years and sites for each of the three life history categories (competitive, stress-tolerant, and weedy), we used generalised linear mixed models (GLMMs) to calculate size-specific patterns in colony survival, transitions in size, fragmentation probability, fecundity, and recruitment for each assemblage at our four focal geographical locations (AS, AT, JS, and JT; Table S1).
Table S1.Pooled number of colonies used to evaluate size-specific patterns in colony survival, transitions in size, fragmentation, and recruitment for each regional competitive, stress-tolerant, and weedy coral assemblage in Australia and Japan.(Australia vs. Japan), and ecoregion (tropical vs. subtropical) included as fixed effects (Fig. S1).We also included the random effects of colony identity and survey location to address any within-subject-variability and autocorrelation arising from our pooling of data across multiple years and sites.

Size transitions
Colony size transitions reflected the change in colony surface areas recorded across successive surveys, which we modelled as colony size at t+1 as a function of colony size at t using a polynomial GLMM (Fig. S2).As with survival we modelled colony size transitions with the variables of life-history classification, country, and ecoregion included as fixed effects, and the variables of colony identity and survey location included as random effects.
Separately we also modelled the relationship between the variance in colony size at time t+1 and colony size at time t.We determined this relationship by modelling the residuals from our initial colony size transition model as a function of colony size at time t, using a gamma GLMM to allow for a non-linear pattern whilst preventing negative variance.AIC scores confirmed the validity of this approach over an equivalent linear format (AIC: linear = 2413.5;gamma = 334.1).Again, we included life-history classification, country, and ecoregion as fixed effects, alongside the random effects of colony identify and site location.

Fragmentation
We recorded colony fragmentation in the event of observed colony breakage, recording the size (surface area, cm 2 ) of all remnants produced in each case.Using a polynomial binomial GLMM, we then modelled colony fragmentation probability as a function of colony size at time t (Fig. S3A).Initially, we performed this analysis using only a binomial GLMM (Fig. S3B).However, despite AIC scores indicating this binomial model was the most accurate (AIC: binomial = 1167.1;polynomial binomial = 1245.1),the polynomial binomial format offered an improved representation of visual patterns within our fragmentation data (Fig. S3).
As was the case across the other vital-rates, we included life-history classification, country, ecoregion, colony identity, and site location as fixed and random effects.
We also modelled the number and size of colony fragments produced during fragmentation events.With our observations of the number of fragments produced by fragmenting colonies representing count data, we modelled the number of fragments produced as a function of fragmenting colony size at time t using a Poisson GLMM (Fig. S4A).Meanwhile, we modelled fragment size as a function of fragmenting colony size at time t, using a polynomial GLMM (Fig. S4B), which provided a more representative fit than an equivalent linear format (AIC: linear = 2243.5;polynomial = 2239.1).Finally, using a gamma GLMM we also modelled the variance in fragment sizes as a function of fragmenting colony size at time t (AIC: linear = 1578.0;gamma = 1440.1).Across each of our models exploring size-specific patterns in the number and size of any fragments produced during fragmentation events we only included life-history classification and country as fixed effects variables.There was insufficient replication in our data for us to include the fixed effect of ecoregion and the random effects of either survey location or colony identity.Subsequently, it was necessary for our analyses to assume that size-specific patterns in the number and size of fragments produced during fragmentation events is consistent across tropical and subtropical conspecifics.

Recruitment
During the repeated surveys of our permanent coral plots, we recorded the number and size of all new colonies <5cm diameter appearing within each plot.However, since our approach to document our permanent coral plots over time involved taking photographs of each overall plot, it was possible to document the occasional appearance of new colonies larger than our predefined size threshold.Consequently, our documented sizes of newly establishing colonies ranged between 0.03cm 2 to 265.9cm 2 across our surveys, encompassing both new recruits and colonies that settled but remained undetectable in previous years.Using these counts of newly establishing corals we quantified annual and regional variation in the recruitment densities of competitive, stress-tolerant, and weedy coral populations (Table S2).We also used the size of newly established corals to estimate assemblage-specific recruit size distributions (Fig. S5).With the parental lineage of newly appearing colonies unknown, we modelled new colony size at time t+1 independent to existing colony sizes at time t using a linear regression, extracting the mean colony size and standard deviation for each assemblage.Initially, we included life-history classification, country, and ecoregion as fixed effects within this recruitment model allowing us to quantify both inter-assemblage and regional variation in the size of newly establishing colonies (Fig. S5).However, the majority of the variation between each population's new colony size distribution was solely generated by their life-history group classification (ANOVA.F2,1108 = 48.8,p < 0.001), with the country and ecoregion variables providing only a small contribution (ANOVA.Country: F1,1108 = 6.4,p = 0.01; Ecoregion: F1,1108 = 0.007, p = 0.932; Fig. S5).Subsequently, to maximise our sample size for estimating recruitment parameters we subsequently dropped both the ecoregion and country terms from the model.

Fecundity
Due to the logistical challenges associated with insitu measurements of colony fecundity (Gilmour et al. 2016), we did not empirically record the fecundity of our tagged colonies.
Instead, we modelled size-specific patterns in colony fecundity using a relationship linking colony size and larval output (larval volume, mm 3 ) recorded in the coral communities at Lizard Island, on the Great Barrier Reef (Hall and Hughes 1996).Firstly, we categorised the coral species surveyed by Hall & Hughes (1996) as competitive, stress-tolerant, or weedy according to their shared life-history characteristics (sensu Darling et al., 2012).Using a polynomial GLMM we subsequently quantified a relationship between colony size and larval output for competitive, stress-tolerant, and weedy coral taxa (Fig. S6).
We acknowledge here that our approach to modelling fecundity does imply an assumption that all larvae produced by an assemblage will reseed back into that same assemblage; an assumption that is inappropriate for coral populations which typically exist as open populations with larvae capable of dispersing away from their source populations (Graham et al. 2008, Yau et al. 2014).We corrected this assumption by parameterising a recruit survival function (ϕ) into our IPMs.This recruit survival function serves to convert estimates of larval output from a measure of volume into the proportional contribution of colonies towards observed recruit densities, as a function of their size.Thus, although we have modelled fecundity using data from a distinctly different community, our use of the recruit survival function ensures that recruitment patterns within our IPMs were determined by empirical counts made within our focal communities and made no assumptions regarding the initial source of new recruits.Our IPMs were therefore not sensitive to changes in colony fecundity, and our inclusion of this vital rate merely allowed us to close the loop between the dynamics of existing colonies and the dynamics of recruitment in order to quantify measures of long-term population performance and transient potential.We estimated the recruit survival function as the ratio between the total expected larval output of a population in any given year and the corresponding annual recruitment count for that population (sensu Bramanti et al., 2015;Cant et al., 2021).

Figure S6.
Size-specific patterns in larval output (cm 3 ) estimated for competitive, stresstolerant, and weedy coral populations using data obtained from coral communities on Lizard Island, on the northern Great Barrier Reef (Hall and Hughes 1996).

Section S2: Classifying tagged corals according to shared morphological and ecological trait characteristics
Table S3.Huang et al. (2014), 8. Benzoni et al. (2010), 9. Schmidt-Roach et al. (2014).autocorrelation describe the correlation between successive elements within a series, such that positive autocorrelation reflects the condition whereby the properties of any element are closely related to those preceding it (Sokal and Oden 1978).Next, we estimated the frequency spectrum of each time series.The frequency spectrum of a timeseries reflects the periodicity of any recurrent variability across the series, with higher frequencies associated with shorter-term fluctuations (Greenman and Benton 2005).The frequency spectrum of a time series is equal to its spectral exponent (β) and calculated as the slope between the log spectral density and log frequency of the time series (Gilljam et al. 2019).We calculated the frequency spectra of each of our SST time-series using the spectrum function from the stats R package (R Core Team 2019).
Table S4.The sea surface temperature (SST) regimes experienced by coral assemblages in tropical and subtropical regions of Australia and Japan, quantified using measures of mean monthly SST (x̄sst), monthly SST variance (cvsst), monthly SST autocorrelation (asst), and monthly SST frequency spectrum (βsst).Measures estimated from 69-year SST timeseries obtained from the Met Office Hadley Centre climate dataset (Rayner et al. 2003).Finally, prior to conducting partial least squares analyses into the association between the long-term performance and transient potential of coral assemblages with patterns in thermal conditions it was necessary for us to evaluate for collinearity across our abiotic variables.We tested for collinearity using the measure of tolerance which describes an inverse measure of the correlation between multivariate predictor variables with estimates of <0.1 evidence of collinearity (Fox 1991).We calculated measures of tolerance for our abiotic variables using the function multicol from the fuzzySim package (Barbosa 2015).Our test for multicollinearity, when we included all four SST variables, returned tolerance estimates of ~0 highlighting a strong correlation between one or more of the variables.Subsequently, we explored collinearity across each triple-wise combination of our four abiotic variables and determined that the triple-wise combination of the variables of mean monthly SST, monthly SST variance, and monthly SST frequency spectrum exhibited the least collinearity (Table S6).Accordingly, we omitted the variable of monthly SST autocorrelation (asst) from further analyses.

Figure S1 .
Figure S1.Colony survival probability as a function of colony size, showing the regional and

Figure S2 .
Figure S2.Colony size at time t+1 as a function of colony size at time t, showing the

Figure S3 .
Figure S3.Comparison between size-specific patterns in fragmentation probability modelled

Figure S4 .
Figure S4.(A) Number and (B) size of fragments produced as a function of fragmenting

Figure S5 .
Figure S5.Regional and interspecific variation in the size densities of newly establishing

Table S2 .
Densities of new colonies of competitive, stress-tolerant, and weedy coral taxa observed during the repeated surveys of our permanent coral plots in2017, 2018, and 2019.
Table of coral genus and species to which our tagged colonies were identified, alongside the proportion of each taxonomic rank assigned to each of the four assemblage classifications: Competitive, Generalist, Stress-tolerant, and Weedy.Greyscale used to differentiate between colonies tagged in subtropical Australia, tropical Australia, subtropical Japan, and tropical Japan.

Table S5 .
(Rayner et al. 2003)ce temperature (SST) regimes experienced by coral assemblages in tropical and subtropical regions of Australia and Japan between 2017 & 2019, quantified using measures of mean monthly SST (x̄sst), monthly SST variance (cvsst), and monthly SST frequency spectrum (βsst).Measures estimated from 12-month SST timeseries obtained from the Met Office Hadley Centre climate dataset(Rayner et al. 2003).Note that we have omitted the measure of monthly SST autocorrelation from this table as it was omitted from our main analyses due to collinearity.

Table S6 .
Tolerance estimates obtained for each of the sea surface temperature (SST) measures of mean monthly SST (x̄sst), monthly SST variance (cvsst), monthly SST autocorrelation (asst), and monthly SST frequency spectrum (βsst) across each triple-wise combination possible with the four abiotic variables.