Geographically Opposing Responses of Sea Urchin Recruitment to Changes in Ocean Climate

Abstract

Sea urchins are voracious herbivores that influence the ecological structure and function of nearshore ecosystems throughout the world. Urchin population growth rates may be particularly sensitive to climate change because adult reproduction and larval development can vary greatly with food availability and temperature, and the transport of their larvae, which spend months feeding in the plankton, are affected by changes in ocean currents. Yet how climate alters sea urchin populations in space and time by modifying larval recruitment and year-class strength remains untested. Using an unprecedented spatially replicated 27-year dataset we illustrate how ocean temperature and climate oscillations differentially affect larval recruitment of the purple sea urchin (Strongylocentrotus purpuratus) and give rise to geographic asynchrony between northern California (positive) vs southern California (negative). Importantly, we found a strong correlation between larval recruitment and regional year class strength suggesting that recruitment variation plays an important role in controlling population dynamics. These results are the first to show that climatic fluctuations shape broad-scale patterns of sea urchin larval recruitment and are likely to control dynamics of both populations and marine ecosystems that vary over the geographical range of their distribution.

Author contributions DKO, SS, and DR designed research. DKO designed, built and conducted analyses and wrote the initial manuscript. DKO, DCR and SS managed data. SS initiated and oversaw data collection and collaborated on all analyses. All authors contributed to revisions.

Introduction

Climate forecasts project large changes in ocean temperature, biogeochemistry and the nature and frequency of disruptive events that affect ocean circulation and primary productivity (e.g. Cai et al. 2014; Yeh et al. 2009). Such shifts are expected to impose wide-reaching ecological impacts, in part by altering animal recruitment in space and time (Sydeman et al. 2015). As a result, characterizing how climate shapes variability in recruitment can provide substantial benefits to the conservation and management of marine resources. For species with planktonic larvae, significant challenges remain in achieving such objectives due to the substantial effort needed to characterize spatial and temporal variation in larval abundance and the numerous sensitive vital rates that contribute to it.

For species like sea urchins, understanding causes and consequences of recruitment variability has both ecological and economic implications. Sea urchin grazing can alter the structure of some of the world's most diverse and productive marine ecosystems, including coral reefs (e.g. Edmunds & Carpenter 2001; Mumby et al. 2007), seagrass meadows (reviewed by Valentine & Heck Jr 1999) and kelp forests (reviewed by Filbee-Dexter & Scheibling 2014). In addition, sea urchins form the basis of important nearshore fisheries in many regions of the world (e.g. Andrew et al. 2003; Kato & Schroeter 1985). As a result, climate-driven changes in sea urchin populations have the potential to profoundly affect both marine ecosystems and associated fisheries. Much of the ecological research on sea urchins has focused on the roles of predation and disease in controlling adult abundance and their cascading influence on community structure (e.g. Burt et al. 2018; Estes & Duggins 1995; Filbee-Dexter & Scheibling 2014; Lafferty 2004; Mann & Breen 1972). However, despite the widespread recognition of the importance of recruitment variation in controlling population fluctuations in many marine species (Shelton & Mangel 2011), relatively few studies have examined the biotic and abiotic processes controlling the supply of sea urchin larvae, and the degree to which they affect the abundance and dynamics of older life stages.

Changes in climate can affect spatial and temporal patterns of larval supply by influencing the production of larvae by benthic adults and the survival of larvae in the plankton. Increases in ocean temperature and storm-related disturbances can impact larval production by (1) increasing adult mortality via physical disturbance (Ebeling et al. 1985) and the spread of water-borne pathogens (Behrens & Lafferty 2004; Lafferty 2004), and (2) reducing adult fecundity by altering food availability (Foster et al. 2015; Okamoto 2014) and inhibiting gametogenesis (Basch & Tegner 2007; Cochran & Engelmann 1975). Because sea urchins produce feeding larvae that spend weeks to months in the plankton, increases in ocean temperature can also affect larval development, growth and survival, either directly, or indirectly by altering the availability of their phytoplankton food source (Bertram & Strathmann 1998; Byrne et al. 2009; Hoegh-Guldberg & Pearse 1995; Meyer et al. 2007; Strathmann 1987). Finally, climate related changes in patterns of ocean circulation can affect the transport of larvae from source to destination (Siegel et al. 2008; but see Morgan 2014). Thus, the effects of climatic change on sea urchin recruitment represent cumulative impacts on adult abundance and reproduction, current patterns that transport larvae, and larval development and survival. Because patterns of ocean temperature, circulation and upwelling can vary dramatically in space, the effects of climate change on sea urchin recruitment potentially vary over large spatial scales. A dearth of long-term, high frequency, spatially extensive data has prevented characterizing temporal and spatial variability in larval recruitment in sea urchins, the degree to which it is explained by different sources of environmental variation, and the relative importance of these drivers in accounting for fluctuations in population size.

Here we used a unique 27-year weekly to biweekly time series of newly metamorphosed larvae (which we refer to as “larval recruits”), collected on artificial substrates placed at seven locations distributed across >1000 km of coast in California to: (1) quantify variation in larval recruitment across different temporal and spatial scales; (2) evaluate whether larval recruitment on artificial substrates predicts year-class strength in natural populations; and (3) determine the relative importance of factors contributing to the observed variation in larval recruitment, namely those that affect reproductive output (such as benthic adult abundance and per-capita fecundity) and oceanographic conditions that affect planktonic larval survival and transport. Due to the correlative nature of our analyses our primary aim was to reveal potential causal relationships that merit further evaluation as drivers of sea urchin recruitment dynamics at spatial and temporal scales relevant to natural populations.

Study system

Populations of the purple sea urchin (Strongylocentrotus purpuratus) occupy shallow subtidal and intertidal rocky substrata from at least 27°N on the western coast of the Baja Peninsula (Olivares-Bañuelos et al. 2008) to at least 59°N on the Kenai Peninsula in Alaska (Field & Walker 2003). Purple urchins are broadcast spawners and the seasonality of their spawning in California is generally thought to be driven by photoperiod and temperature (Cochran & Engelmann 1975; Gonor 1973; Pearse et al. 1986). Fertilized zygotes develop into planktonic echinoplutei and are obligate planktivores that consume phytoplankton (Strathmann 1987). After spending several weeks to months in the plankton individuals begin final metamorphosis and settle to the benthos (Strathmann 1978) at a size of ~500 μm in diameter (Okamoto unpublished data). Larval recruitment varies dramatically among locations at both small and large spatial scales (Ebert 2010).

Collection of newly settled urchins along the coast of California, USA (1990–2016)

Recruitment of newly metamorphosed purple sea urchins was sampled in three major regions along the California coast from 1990 through 2016, with a total of 54,588 observations. Sampling regions (from south to north) included two sites in San Diego (Scripps Pier and Ocean Beach Pier), four sites in the Santa Barbara Channel (Anacapa Island, Stearns Wharf, Ellwood Pier and Gaviota Pier) and one site at Fort Bragg (Figure 1). San Diego and the Santa Barbara Channel lie within the Southern California Bight and Fort Bragg is in northern California. At each site, urchins were collected using nylon-bristled scrub brushes (2.5 cm long bristles and a 6 × 9 cm wooden base) suspended 1 to 2 m from the benthos (Ebert et al. 1994). The majority of deployments included 4–8 replicate brushes collected weekly at each site from 1990 to 2003, and biweekly thereafter through 2016. Upon collection brushes were transported to the laboratory in plastic bags and rinsed through a 350 μm mesh sieve. Purple urchins were then identified to species, sorted from other organisms, counted and preserved see (Ebert et al. 1994) for further details.

Figure 1:

Map of larval recruitment collection locations in California. Colors represent the average spatial gradient of sea surface. The hatched white lines show the shoreline buffer used to constrain sea surface temperature and sea surface irradiance (the chlorophyll index) to local larval recruitment observations.

We conducted three primary analyses using these data (see Methods for full details): (1) we examined biweekly, seasonal and interannual trends and spatial correlations in S. purpuratus larval recruitment using a Bayesian multivariate time series model; (2) we tested whether trends in larval recruitment corresponded to year-class strength using a generalized linear mixed effects model; and (3) we nested a linear regression into the Bayesian time series model from (1) to evaluate the degree to which various biotic and abiotic factors explained patterns of sea urchin larval recruitment in northern and southern California. In this latter analysis we linked larval recruitment data to long-term data on sea surface temperature, chlorophyll a (larval food), kelp coverage (adult food), regional upwelling indices, regional adult abundance indices and global climate indices.

Results

(1) Estimated Patio-temporal Trends in Purple Urchin Larval Recruitment

Substantial interannual variability in larval recruitment of S. purpuratus was observed with years of poor recruitment (e.g. 1995, 1998, 2005, 2016) that were, on average 1–2 orders of magnitude lower than neighboring years (Figure 2, Figure S1). Larval recruitment was highly synchronous among sites within each of the two regions in the Southern California Bight indicated by mean interannual correlations of r = 0.73 and 0.85 for sites within the Santa Barbara Channel and San Diego, respectively (Figure 2b, c). Within the Santa Barbara Channel, pairwise correlations in interannual trends involving Gaviota, Ellwood and Stearns Wharf were higher (r = 0.86 to 0.90) those involving Anacapa (r = 0.46 to 0.71). This resulted in part because different rates of interannual declines from high larval recruitment in 2011 to a common low value in 2016. Anacapa began to decline in 2012 while the other sites, while the declines at the other sites did not begin until 2014 (Supplementary Figure 1 b-e). While Santa Barbara Channel sites all exhibited a period of strong increases following the low in 2005, San Diego sites remained below average for the remainder of the time series (Figure 2 c, Supplementary Figure 1 f-g). In contrast, interannual trends in recruitment at Fort Bragg were asynchronous with sites in the Santa Barbara Channel (Figure 2a vs. 2b, mean r = −0.23) and uncorrelated to sites in San Diego (Figure 2a vs. 2c; mean r = 0.12).

Figure 2:

Standardized annual scale trends (log-scale, standardized to one standard deviation) of larval recruitment of purple urchins (S. purpuratus) from 1990 through 2016 for: (a) Fort Bragg, (b) sites in the Santa Barbara Channel, and (c) sites in San Diego. Colors within each panel represent individual sites associated with the legend. Interannual trends are estimated using a 3/2 Matern Gaussian process covariance within a Bayesian multivariate state-space model accounting for spatiotemporal correlations.

Larval recruitment in southern California was highly seasonal, with similar patterns among sites (Figure 3, Figure S2). On average 90% of recruitment occurred from March to July with a single peak in late April/early May (Figure 3a). By contrast recruitment at Fort Bragg in northern California extended over a longer period of time (90% occurred, between January and September) and often included two peaks per year (a large consistent peak around July and a smaller peak in February and March of some years; Figure 3a, S3). The seasonal peaks in recruitment in southern California coincided with the peaks in sea surface chlorophyll (Figure 3a vs. 3b) and troughs in sea surface temperature (Figure 3a vs. 3c). In northern California the primary peak occurred slightly after the peak in chlorophyll a.

Figure 3:

Seasonal trends at each site for (a) standardized larval recruitment of S. purpuratus, (b) biweekly mean sea surface chlorophyll a (chla) and (c) biweekly mean sea surface temperature (SST). The seasonal trend in larval recruitment at each site was standardized by the mean maximum seasonal value. Chla and SST were measured using satellite imagery within 25 km of the coastline and 150 alongshore km of the location where recruitment was sampled. See Methods for details.

(2) Relationship Between larval recruitment and subsequent juvenile year-class-strength

Recruitment of juvenile purple urchins at shallow subtidal reefs in the Santa Barbara Channel exhibited a significant, positive correlation with larval recruitment to brushes two years prior (Figure 4, , p = 0.002). Years with high larval recruitment corresponded with a nearly three-fold increase in the mean density of juvenile urchins two summers later.

Figure 4:

The relationship between the summer density of juvenile purple sea urchins on reefs in the Santa Barbara Channel and the density of recruits on larval collectors during March-July of the previous year (i.e., 12–20 months prior). Points and error bars represent annual means (+/- SE) averaged over all sites. The line with the band represents the estimated relationship and 95% uncertainty interval.

(3) Relationships with Environmental Conditions

Larval recruitment of purple urchins showed strong correlations with local sea surface temperature (SST, Figure 5 a-g, Figure 6 a), and climate indices, specifically ENSO - measured as the multivariate El Niño Southern Oscillation Index (MEI - Figure 5 h-n, Figure 6 f) and the North Pacific Gyre Oscillation (NPGO –Figure 5 o-u, Figure 6 g). The sign of the correlation was opposite between Northern California (Fort Bragg) and Southern California (Santa Barbara Channel and San Diego). The Pacific Decadal Oscillation showed a correlation with larval recruitment only at Fort Bragg (Figure 6 h), but the estimate was highly collinear with that of the MEI leading to inflated variance of both coefficients (Supplemental Figure 3). No correlation was observed between larval recruitment and adult urchin density or canopy kelp (data that were only available for southern California sites; Figure 6 b, e), or the Bakun Upwelling Index (Figure 6c). Sea surface chlorophyll showed no correlation with larval recruitment at Fort Bragg and in the Santa Barbara Channel and a negative correlation in San Diego – the opposite of the hypothesized relationship that more food in the plankton would lead to higher abundances of larvae and higher numbers of larval recruits (Figure 6 d). The network of correlations among these variables are depicted in (Supplemental Figure 4).

Figure 5:

Correlation between purple urchin recruitment at each site and (a-g) sea surface temperature, (h-n) the multivariate ENSO index, and (m-u) the North Pacific Gyre Oscillation (NPGO). Small grey points represent biweekly means and black points and lines represent annual means.

Figure 6:

Standardized Bayesian multiple regression coefficients for relationships between S. purpuratus and local covariates (a-e) or global climatic indices (f-h). Models with global and local variables were fit separately. Coefficients were estimated using a multiple regression nested within a Bayesian multivariate time series model. Coefficients are a-priori biased towards zero using a horseshoe prior. For details see Methods.

In southern California recruitment was orders of magnitude lower during warm, El Niño conditions and during the negative phase of the NPGO with the more southern sites in San Diego responding more strongly to temperature than those further north in the Santa Barbara Channel (Figure 6 a). In contrast, Fort Bragg, the northern California site, responded positively to warmer El Niño conditions and negatively to the NPGO. Importantly, sites in southern California showed opposite relationships from those at Fort Bragg in northern California (Figure 6 f, g). [Note that we fit models with local variables (SST, adults, kelp coverage, chlorophyll, and Bakun upwelling index) and ocean-scale indices (ENSO/NPGO/PDO) separately because of strong functional dependence of the latter with SST]. The correlations between recruitment and both SST and ENSO occurred on roughly 3–5 year cycles, while the correlation with NGPO occurred on decadal scales.

Discussion

Climate change poses significant threats to marine ecosystems. Yet determining how many species and communities will respond to shifting temperature and ocean storm regimes remains a difficult task. Using a multi-decadal, high frequency, spatially extensive time series we show how large-scale climatic variation can give rise to geographically different responses in larval sea urchin recruitment that were synchronous within regions and asynchronous between them. Larval recruitment of purple sea urchins varied by orders of magnitude among years and responded oppositely to climate in northern versus southern California. Our finding that larval recruitment was a significant factor in determining year class strength in natural populations indicates that large fluctuations in larval recruitment of this prominent herbivore could have far reaching ecological impacts that resonate throughout the regional marine ecosystems (Pearse 2006).

Climate related changes in the supply of sea urchin larvae are thought to be a primary factor contributing to increases in sea urchin populations that have caused the dramatic transformation of benthic communities (Fabricius et al. 2010; Hernández et al. 2010; Ling et al. 2008). Yet until now, no long-term data have corroborated whether urchin larval supply, regulated by climate, can alter year class strength. Demonstrating this link has historically proved challenging because “larval recruitment” can be attenuated by early post-settlement mortality (Connell 1985; Rowley 1989). Our analysis indicates that the year class strength of purple sea urchins in a region is significantly influenced by larval supply. This result points to the possibility that future oceanographic changes may reduce purple urchin recruitment in southern California, but increase it in northern California with potentially cascading effects on benthic ecosystems. The fact that our study shows similar correlations between oceanographic factors and recruitment over a 27-year period, which included multiple ENSO events, suggests that such climatic associations will continue.

While many recruitment-environment correlations break down over time (Myers 1998), our study shows a strong association between urchin larval recruitment and climate over a nearly 3 decade time scale. Although sea urchin larval recruitment data prior to 1990 were not available in the fine temporal scale characterized by our study, there is evidence of ENSO-related recruitment failures in populations in southern California since 1960s. Between 1969 and the early 1980’s, recruitment of juvenile sea urchins was anomalously low in El Niño years (Ebert 1983; Tegner & Dayton 1991). Between 1969 and 1977, juvenile recruitment was lowest during the three El Niño years of 1970, 1973, and 1977 at False Point, California near San Diego (Ebert 1983; Tegner & Dayton 1991) and at nearby Point Loma, low recruitment followed the 1982–83 El Niño event (Ebert 1983; Tegner & Dayton 1991). How patterns of larval recruitment may change in the future depends largely on the process underlying their associations with El Niño.

Although climate related changes to both the benthic and planktonic stages of purple sea urchins can affect larval recruitment via their effects on the production of larvae, our analysis shows no significant correlations between larval recruitment and adult abundance, adult food (kelp) or larval food (phytoplankton) – (Figure 6 and Supplemental Figure 3). This narrows down the range to two plausible hypotheses to the regionally varying effects of ENSO and temperature on larval supply and recruitment: (1) temperature-related effects on larval production and survival, and 2) ENSO-related changes in currents that vary between southern and northern California, which affect larval delivery. While adult purple urchins can tolerate temperatures up to 25°C (Farmanfarmaian & Giese 1963), temperatures above 17°C during ENSO events in southern California may adversely affect larval supply via effects on gamete production (Basch & Tegner 2007; Cochran & Engelmann 1975), fertilization (Schroeder & Battaglia 1985) larval development and gene expression (Padilla-Gamiño et al. 2013; Runcie et al. 2012; Wong et al. 2018), and larval survival (Azad et al. 2012). Such responses to high temperatures may, in part, explain why recruitment at Fort Bragg, where sea surface temperatures are rarely above 16°C, was not negatively correlated with temperature.

Like variation in temperature, regional circulation is hypothesized to be an important driver of larval recruitment for many species, including those in the Southern California Bight (Blanchette et al. 2006; Broitman et al. 2005; McManus & Woodson 2012; Woodson et al. 2012). ENSO can cause major changes in patterns of ocean circulation along the US west coast, with warm El Niño years often characterized by anomalously strong poleward flow and decreased local retention (Lynn & Bograd 2002; Mitarai et al. 2009). Importantly, these effects differ between northern and southern California and may partly explain the divergences in climatic responses in recruitment between these two regions. In northern California, El Niño events are associated with relaxed upwelling and downwelling Kelvin waves (Chavez et al. 2002) which hypothetically improves larval retention, but is thought to have a more nuanced impact in the Southern California Bight (Lynn & Bograd 2002; Mitarai et al. 2009). Although we cannot definitively partition out the effects of climate associated changes in current patterns and temperature on larval recruitment, our observations on larval recruitment and subsequent year-class strength point strongly to broader-scale ocean climate effects.

Our study demonstrates a consistent and geographically variable relationship between climate and larval recruitment and year-class strength of purple sea urchins over a three-decade time period. These unique time series provide an unprecedented illustration of climate related impacts on larval recruitment dynamics that are notoriously difficult to investigate. We show that fluctuations in larval recruitment can influence year-class strength in an herbivore that is known to severely alter the composition of an entire benthic ecosystem. Although these results are based on correlations, they were observed consistently over three decades (with further anecdotal evidence spanning six decades) that have included multiple ENSO events. Moreover, our findings provide important insights into the environmental drivers of larval recruitment, which is essential for understanding and predicting the influence of climate on regional population and community dynamics of marine species. Oceans are experiencing simultaneous shifts in temperature, water chemistry, productivity and circulation; thus future investigations aimed at determining the specific biotic and abiotic processes that regulate larval recruitment should provide much needed knowledge of how climate change is influencing regional population dynamics to alter the dynamics and structure marine ecosystems.

Methods

(1) Estimated patio-temporal Trends in Purple Urchin Larval Recruitment

We estimated annual, seasonal, and spatial trends in larval recruitment using an integrated spatio-temporal model that accounted for both the intercorrelated and heterogeneous nature of the multivariate time series. Specifically, we used a Bayesian hierarchical model with spatiotemporal correlations in the underlying annual and seasonal scale trends, and spatial correlations in the process error terms. We used this model to simultaneously estimate biweekly, seasonal and interannual trends in recruitment, and spatial synchrony while accounting for observation noise.

Using this model, we calculated the log-scale biweekly mean trend in larval recruitment as the sum of the estimated annual trend, seasonal trend and process error in the statistical model. The annual trend was estimated using a separable spatio-temporal covariance prior where the spatial covariance is unstructured (i.e. each element estimated) and the temporal covariance determined by a Matérn 3/2 prior. Functionally, we constructed the covariance matrix for the annual trend using Kronecker matrix-vector operations with the Cholesky decompositions of the spatial and temporal correlation functions independently for efficiency (Steeb & Hardy 2011). The seasonal trend within each site was estimated using a periodic temporal covariance (MacKay 1998). The process error (deviations from the sum of smoothed annual and seasonal trends) was modeled with an unstructured spatial covariance, assuming temporal independence. We used a Poisson likelihood to link the expectation (biweekly mean trend) to the observed counts of S. purpuratus larval recruits.

Occasionally we encountered samples with sufficiently large numbers of larval recruits (hundreds to thousands of urchins per sample - 1,070 out of 53,478 samples or ~2%), which prevented efficient identification of all urchins to species. In these cases, S. purpuratus were identified from a homogenized subsample of 30–100 individuals. We accounted for the uncertainty in S. purpuratus abundance introduced by this subsampling routine by using a beta prior with parameters determined by the subsampled S. purpuratus counts and the non-S. purpuratus counts. This prior was derived as the analytical posterior distribution of the proportion of the true, latent fraction that was S. purpuratus given a Jeffrey’s prior on a binomial model of the observed counts.

All equations for the model are given in Supplementary Table 1 and parameters and variables defined in Supplementary Table 2. We estimated the models using Stan (Carpenter et al. 2016; Hoffman & Gelman 2014; Stan Development Team 2016b, a) via R (R Core Team 2017) with three 1000 iteration chains after 1000 iteration burn-in. Stan code for the model is provided in the appendix.

(2) Relationship Between larval recruitment and subsequent juvenile year-class-strength

We tested whether the mean of our estimates of larval recruitment in the Santa Barbara Channel (mean of Anacapa, Stearns Wharf, Ellwood Pier and Gaviota Pier) was predictive of subsequent juvenile recruitment on natural reefs in the region. We calculated juvenile densities from the Channel Islands Kelp Forest Monitoring (KFM) Program (Kushner et al. 2013), including only sites whose time series extend from 1990 through 2016. To test for correlations, we used a Bayesian generalized linear mixed effects model (GLMM) with a negative binomial likelihood and survey site as a random effect.

The KFM program surveys purple sea urchins at the Santa Barbara Channel Islands by counting the total number of individuals in defined areas and measuring approximately 100 individuals at each site. Thus, to model density we used the number of juveniles measured divided by the total number of individuals measured, multiplied by the total number of individuals counted, divided by the area surveyed.

We modeled density in the GLM using the number of measured juvenile urchins (2.5 cm in test diameter the approximate cutoff size for reproduction; ref Kenner & Lares 1991). Let μ_i,t represent expected juvenile density at site i in year t. We constructed the regression as:

We allowed the intercept (α_i) to vary by site nested within each island because of overall differences in mean juvenile density among sites and islands. We used a negative binomial (NB) with the using direct mean and variance parameterization form:

We constructed the model in this format (i.e. with the density denominator in the left hand side of the equation) to maintain the sample size and integer nature of the data while modeling the mean density. We tested the hypothesis of no correlation using a likelihood ratio test. Models were estimated using glmmTMB (Magnusson et al. 2017) in R.

(3) Relationships with Environmental Conditions

We tested how larval recruitment trends correlated with physical and biological variables using two core analyses. Covariates included the following:

(i) Oceanographic climate indices (monthly, 1990–2016, all sites)

We used three major global indices of oceanographic climate. The multivariate El Niño Southern Oscillation Index (MEI) provides a metric of the intensity of El Niño/La Niña fluctuations and is derived from several metrics of sea surface temperature, surface winds, sea level pressure, and cloudiness of the sky (Wolter & Timlin 1993; Wolter & Timlin 1998). The Pacific Decadal Oscillation (PDO) is the 1st empirical orthogonal function analysis mode of sea surface temperature variability and height in the Northeast Pacific (Mantua & Hare 2002) and the North Pacific Gyre Oscillation (NPGO) is the 2nd mode (Di Lorenzo et al. 2008).

(ii) Coastal upwelling index (monthly, 1997–2016, all sites)

The Bakun index (Bakun 1973) provides an index of coastal upwelling and specifically describes the volume of water that is transported offshore from Ekman transport (http://www.pfel.noaa.gov/). Negative values indicate downwelling and surface waters that moved onshore, while positive values indicate upwelling and surface waters moving offshore. This index has been used in relation to delivery of larvae onshore and advection from shore in addition to its value as a metric of coastal productivity. For the Southern California Bight sites, we used data from 33N-119W, while in Northern California we used data from 39N-125W.

(iii) Sea surface chlorophyll (monthly, 1997–2016, all sites)

Satellite imagery of sea surface chlorophyll provides a spatially and temporally resolved estimate of phytoplankton biomass that is not available from in situ sampling. Thus, as a metric of planktonic food availability we used version 3.1 of the OC-CCI merged ocean color time series (Sathyendranath et al. 2018) from SeaWIFs (O'Reilly et al. 1998), MERIS (Rast et al. 1999), MODIS (Esaias et al. 1998) and VIIRS (Schueler et al. 2002) that allows for the temporal and spatial coverage required for this study. Because sea surface chlorophyll concentration data are spatial grids of time series, we included averaged cells meeting particular geographic conditions for each site. For mainland sites, all cells had to lay within 10 km offshore of the mainland coastline that stretched 150 km in coastline length from the larval recruitment collection site or within 5 km of any island coastline of any island within a 150 km radius. For the site at Anacapa Island, we included any point within a 150 km radius and within 10 km of any coastline. We used 150 km because that is the average Lagrangian estimate for dispersal distances for species with a planktonic larval duration (PLD) of 30-days (Siegel et al. 2003). However, this is potentially an overestimate for a 30-day PLD as behavior and complex current patterns may reduce this distance substantially (Shanks 2009). Figure 1 depicts the aerial extent of chlorophyll data used in the analysis.

(iv) Sea surface temperature (monthly, 1997–2016, all sites)

We used sea surface temperature data derived from Pathfinder AVHRR (advanced very high resolution radiometer) that were optimally interpolated on daily and 0.25 degree latitude/longitude resolution (Reynolds et al. 2007). To produce monthly scale time series, we ran data through a 30-day moving average filter with a rectangular, backwards looking window in order to capture general temporal trends in the data. We spatially aggregated data using the same buffers as for chlorophyll indices. Figure 1 illustrates mean spatial trend in winter sea surface temperature for the entire study region.

(v) Fall kelp canopy biomass (annual, 1996–2015, Santa Barbara Channel & San Diego only)

The regional biomass of giant kelp Macrocystis pyrifera can fluctuate dramatically from year to year (Bell et al. 2015). Giant kelp is a preferred food and a major constituent of S. purpuratus diets in southern California (Foster et al. 2015). Given that fecundity in S. purpuratus can vary by orders of magnitude as a result of food supply (Okamoto 2014), these fluctuations in M. pyrifera may in part influence urchin reproduction. Thus, biomass of M. pyrifera was derived from Landsat Thematic Mapper satellite imagery (Bell 2017). Data were aggregated for the Santa Barbara Channel (including islands and mainland from Point Conception to Santa Monica Bay through Ventura County) or the San Diego region (mainland coast from the US-Mexico through Orange County including Santa Catalina and San Clemente Islands). We used the 3-month running mean of kelp biomass during the period leading into the spawning season (Sept-Nov) because marked declines in reproductive capacity require several months of consistently low food supply (Okamoto 2014).

(vi) Adult density in the Channel Islands (annual, 1997–2016, Santa Barbara Channel only)

As an index of regional adult reproductive density of sea urchins in the Santa Barbara Channel, we used the geometric mean of adult biomass density using survey data from The Channel Islands KFM Program (Kushner et al. 2013). We used the same sites and data as described above (see relationship between urchin larval recruitment and benthic juvenile recruitment) with the exception that instead of juvenile density we used adult biomass density. This was calculated as the biomass of adults based on: (1) the mean relationship between size and individual biomass, (2) the size frequency of adults at a given site in a given year, (3) the total number of urchins, and 4) the total area surveyed.

To assess which variables exhibited correlations with larval recruitment of purple sea urchins, we nested multiple regression analyses within a time-series modeling framework, where coefficients varied among major regions (Fort Bragg, Santa Barbara, and San Diego sites). The regression analyses included either all global covariates (ENSO, PDO, NPGO) or all local covariates (temperature, chlorophyll, adult densities, kelp coverage, and the Bakun upwelling index). We inserted a centered and standardized regression covariate matrix into the time series model of sea urchin recruitment, replacing the explicit estimates of the interannual trends (Supplementary Table 1). The regression coefficients were given a regularized horseshoe prior (Carvalho et al. 2010; Piironen & Vehtari 2017) that biases the model coefficients towards zero without affecting coefficients of relevant variables.

Variable Correlations

To illustrate the intercorrelated nature of the covariates and responses, we estimated a network model for Gaviota Pier that includes the independent model estimates of S. purpuratus settlement and its seasonality. To do so, we first constrained network structure by excluding all nonsensical interactions (i.e. chlorophyll does not cause ENSO events and is thus eliminated a priori) and including all known directional interactions (i.e. estimated seasonality affecting settlement for is forced into the network). Learning of the network skeleton is achieved via the Hill Climbing algorithm using the bnlearn¹ package in R. Note that strong collinearity can result in the weaker correlation being ignored (e.g. ENSO over SST with Larval Recruitment).

\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\ \\ Stan Code Required to Estimate Trends \\ \\ Author: D.K. Okamoto \\ \\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\ data { // input information int NO; // number of observations int NS; // number of sites int NM; // total number of periods per site int NYM; // number of periods within a seasonal trend int NSUB; // number of observations with subsamples // input data vector [NSUB] P1; // SP counts +0.5 vector [NSUB] P2; // Other urchin counts + 0.5 int SUB [NSUB]; // indicator variable for which observations are subsampled int N[NO]; // total count (on the original scale) int OBS[NO]; // continuous month-site observation index int PRED_MONTH[NM]; // month-site observation index associated with MA vector[NO] D; // number of days cholesky_factor_cov[NYM] D_seas; // seasonal cholesky covariance factor cholesky_factor_cov[NM] D_ann; // annual cholesky covariance factor } parameters { // mean volatility vector<lower=1e-7>[NS] sigma_S_mu; // spatial volatility correlation matrix cholesky_factor_corr[NS] L_Omega_S; // cholesky_factor_corr[NS] L_Omega_Spatial; // seasonal and annual variance vector<lower = 0>[NS] sigma; vector<lower = 0>[NS] sigma2; // mean site settlement row_vector<lower= -10, upper= 10>[NS] mu_S; // iid errors for seasonality matrix<lower= -10, upper= 10>[NYM,NS] z_s; // iid errors for annual trends matrix<lower= -10, upper= 10>[NS,NM] w_z; // iid errors for residuals matrix<lower= -4, upper= 4>[NS,NM] e_z; // proportion of samples that are purps vector<lower= 0,upper= 1>[NSUB] theta; } transformed parameters { matrix[NYM,NS] LS; // site specific seasonal trends matrix[NM,NS] S; // estimated abundance vector [NO] S_exp; // normal scale estimated sp for each observation vector [NO] theta2; // fraction that is SP matrix[NM,NS] w; // spatially and temporally correlated, standardized annual trends matrix[NM,NS] e; // spatially correlated process errors LS = D_seas*z_s; w = D_ann*transpose(L_Omega_Spatial*w_z); e = transpose(diag_pre_multiply(sigma_S_mu,L_Omega_S)*e_z); theta2 = rep_vector(1.0,NO); theta2[SUB] = theta; // caculate expected values S= rep_matrix(mu_S,NM)+diag_post_multiply(LS[PRED_MONTH,],sigma)+ diag_post_multiply(w,sigma)+e; S_exp = exp(to_vector(S)[OBS]) .*D ./theta2; } model { // site SDs for annual , seasonal trends and process errors sigma ~ cauchy(0, 0.5); sigma2 ~ cauchy(0, 0.5); sigma_S_mu ~ cauchy(0,2.5); // spatial correlations for annual and seasonal trends L_Omega_S ~ lkj_corr_cholesky(2); L_Omega_Spatial ~ lkj_corr_cholesky(2); for(i in 1:NS){ z_s[i]~ normal(0,1); mu_S[i] ~ normal(0,5); // site specific log-scale means } // uncertainty on proportions of purples in the sample // jeffrey's prior given subsampling theta~beta(P1, P2); // observation likelihood N ~ poisson( S_exp ); } generated quantities { corr_matrix[NS] Omega_S; corr_matrix[NS] Omega_Spatial; Omega_S = multiply_lower_tri_self_transpose(L_Omega_S); Omega_Spatial = multiply_lower_tri_self_transpose(L_Omega_Spatial); }

Acknowledgements

Funding was provided by the California Urchin Commission, the South Bay Cable Committee, the California Department of Fish and Game Director’s Sea Urchin Advisory Committee, and the National Science Foundation’s support of the Santa Barbara Coastal Long Term Ecological Research program. We are indebted to Tom Ebert and John Dixon, who initiated larval recruitment time series with SS. John Richards, Clint Nelson, Shannon Harrer, Jenny Wolf and many students assisted with field collections, laboratory sampling and data compilation. Sally Holbrook, Russ Schmitt and Cherie Briggs contributed discussions during initial development of analyses. Kyle Cavanaugh, Dave Siegel and Tom Bell provided valuable discussions about giant kelp canopy data. Kyle Cavanaugh and Rachel Simons gave important comments on previous manuscript drafts.

Footnotes

• ¹ Marco Scutari (2010). Learning Bayesian Networks with the bnlearn R Package. Journal of Statistical Software, 35(3), 1–22. URL http://www.jstatsoft.org/v35/i03/