The effects of settlement depth on winter mortality in Pacific oysters
Hjalmar Stake
Degree project inbiology, Bachelor ofscience, 2021 Examensarbete ibiologi 15 hp tillkandidatexamen, 2021
Biology Education Centre, Uppsala University, and Centre for Coastal Research, University ofAgder Supervisor: Ane Timenes Laugen
1
The effects of settlement depth on winter mortality in Pacific oysters
Abstract
Biological invasions are a leading cause of biodiversity loss and cause extensive economic losses. Since its introduction to Europe, the invasive Pacific oyster (Magallana gigas) has successfully established far-reaching populations in Scandinavia. This achievement can in part be attributed to the species’ high tolerance to a wide range of environmental conditions.
However, the temperature range for survival and reproduction of this species has not been fully evaluated and is poorly understood. The relatively cold climate of northern Europe was initially deemed unsuitable for Pacific oyster populations as potential habitat. This would prove to be a vast underestimation of the species’ adaptability, however, since populations are now widely distributed in the region and their range still expanding. Nevertheless, exposure to severe winter conditions has been demonstrated to cause mortality events and change the contemporary distribution of oyster populations. Such conditions are common in Scandinavia and their effects on the oyster population are yet to be assessed. The purpose of this study was to estimate the extent to which mortality occurred among the Swedish Pacific oyster populations as a consequence of the relatively severe conditions during the 2020/2021 winter. Additionally, these surveys sought to investigate the relationship between winter mortality and the oyster settlement depth distribution. Our observations showed that survival probability generally increased with greater settlement depth. This was in accordance with previous findings. Furthermore, we found that there was considerable variation between geographical locations, with survival rates ranging between 63-98%. The average survival rate across all sampling sites was calculated to 87%. This suggested that the mortality of Pacific oysters in the winter 2020/2021 was influenced by other environmental factors in addition to depth. This project primarily aimed to provide an estimate of the relationship between the settlement depth of oysters and survival probability in a year with medium cold winter temperatures in Sweden. Additionally, a secondary purpose of this project was to develop standard procedures for analysing the survival probability of oysters in Scandinavia using generalised linear mixed models.
2
Table of Contents
Introduction ... 3
Biology of M. gigas ... 6
Methods ... 9
Sampling locations ... 9
Sampling methods ... 12
Depth correction ... 12
Climate data ... 14
Data analysis and visualisation ... 15
Results ... 16
Discussion ... 22
Acknowledgements ... 23
References ... 25
References to R packages used ... 27
Appendix 1 ... 28
3
Introduction
The geographical distribution of biological organisms is influenced by environment,
ecological niches, biotic interactions, history of colonization, and human activities (Guisan &
Thuiller 2005, Soberón & Peterson 2005, Sexton et al. 2009, Wiens 2011). Factors that limit a species’ range include physical barriers, mechanical stresses such as wind and ice, and tolerance to salinity and temperature. Biotic factors such as competition over resources, predation, and disease may further restrict a population’s ability to colonize new areas.
Species with high resilience towards a wide range of environmental conditions could in theory become broadly distributed, and the ability to inhabit different environments can be influenced by phenotypic plasticity and local genetic adaptations (Meyers & Bull 2002, West-Eberhard 2005, Kawecki 2008, Piersma & van Gils 2010). Thus, studying the
mechanisms underlying changes in the distribution of species is important for identifying the constraints imposed on a species capacity to expand its range and the potential for future biological invasions.
Marine species may naturally disperse over long distances via ocean currents or be transported by birds and other marine animals. Moreover, following the onset of human exploration and increased global trade in recent years, other vectors of dispersal such as ballast water and the shellfish trade have become increasingly important for the introduction of marine alien species to new areas (Drake & Lodge 2004, Ruesink et al. 2005,
Katsanevakis et al. 2013). Once established, some non-indigenous species have negative effects on native ecosystems and economies and are hence termed invasive (Kolar & Lodge 2001, Vilà et al. 2010). Biological invasions have severely affected Europe and is a leading cause of biodiversity losses (MEA 2005). Marine invasive species may negatively impact native species and habitats, ecosystem services, human health, and cause extensive economic damage (Grosholz 2002, Wallentinus & Nyberg 2007, Molnar et al. 2008, Vilà et al. 2010).
One of the best known non-indigenous marine organisms in Europe is the Pacific oyster; a highly successful and invasive species originating from the Japanese Sea (FAO 2021, Grizel
& Héral 1991). The Pacific oyster’s high tolerance to a wide variety of environmental conditions has made it an attractive species for cultivation and aquaculture, as well as a successful invader of many regions globally (Andrews 1980, Ruesink et al. 2005). Whether the Pacific oyster is a healthy addition to estuarine and coastal environments or an
unwelcome threat to native ecosystems is a matter of debate. While it has the potential of enriching biodiversity through the creation of reefs and a more heterogeneous environment, it may also outcompete indigenous bivalves. One such example is the European flat oyster (Ostrea edulis), of whose reefs an estimated 85 % have been lost globally and whose holdfast in Scandinavia is among the few remaining (Beck et al. 2011).
The Pacific oyster was originally introduced in Scandinavia for cultivation purposes in the 1970s (Eklund 1977). The environmental conditions of Scandinavia, such as the relatively cold seawater temperatures, were initially assumed unsuitable for their reproduction, and the risk of the Pacific oyster establishing itself in the region was considered low. Additionally, water temperatures of >20°C and salinity of >20 PSU was thought to be necessary for
settlement in new locations (Muranaka & Lannan 1984, Fabioux et al. 2005). Gametogenesis and spawning for M. gigas has been observed to depend on water temperature and time of
4
exposure to a given temperature. For instance, research on gonadal development in M. gigas populations in Japan has suggested that temperatures above 23°C were necessary for gonadal tissue development to occur (Kobayashi et al. 1997). However, the specific water
temperature at which these events occur also depend on the genetic strain and geographical location (Mann 1979, Mann et al. 1991). Furthermore, research on the growth and evolution of the biochemical composition of M. gigas has found that the species can adapt to new, sub- optimal temperatures over time and recover lost respiration and filter-feeding capacity (Le Gall & Raillard 1988).
The Pacific oyster has undergone a northward range expansion in recent years, possibly facilitated by warmer summers, milder winters, and above-average seawater temperatures (Diederich et al. 2005, Wrange et al. 2010). Following its arrival in 2006 through natural dispersal from neighbouring areas, successful reproduction has created feral, self-sustaining populations, and the species has now established large populations along Sweden’s west coast (Wrange et al. 2010, Dolmer et al. 2014, Laugen et al. 2015, Faust et al. 2017). The range of Pacific oysters now covers most of the Swedish west coast, extending south to Malmö (García et al. 2018, Laugen et al. 2015) (Figure 1).Furthermore, a study on growth and reproduction of Pacific oysters in the Wadden sea suggests that the species has yet to reach its eco-physiological limits (Cardoso et al. 2007). With continued global warming the Pacific oyster is predicted to continue expanding its range for years to come (King et al.
2021).
5
Figure 1. Distribution of M. gigas in Scandinavia (filled circles) and monitored stations (open circles). Figure credit: Mark Wejlemann Holm (Roskilde University)
6
It has been suggested that reefs of Pacific oysters provide additional preferential habitats and microclimates for other marine species and thus have the capacity to increase overall biodiversity (Reise et al. 2017a, Reise et al. 2017b). However, research on Pacific oysters in the Wadden Sea suggests that the high growth rates and extensive settlements of Pacific oysters may impose restrictions on habitat use of native bivalves and compete with them over resources (Diederich 2006, Krassoi et al. 2008). Mussel beds constitute a highly suitable substrate for oyster settlement (Reise et al. 2017a). And whereas the growth of juvenile oysters appears to be unaffected by settlement substrate, growth of juvenile blue mussels (Mytilus edulis) has been observed to be significantly higher on sand flats than on oyster or mussel beds (Diederich 2006). Furthermore, reefs or aggregates of Pacific oysters can trap sediment and limit water passage in shallow regions (Ruesink et al. 2005). Consequently, this invasive species may compromise biodiversity and even transform native littoral ecosystems (Cognie et al. 2006).
Biology of M. gigas
M. gigas is a species of suspension-feeding bivalves, living attached to hard substrates in open coastal ecosystems of the intertidal zone where they form complex reef structures (Reise 1998, Dupuy et al. 1999). The Pacific oyster is an oviparous species and is known to release its gametes into the water column following a rise in water temperature (Quayle 1988). Following a pelagic phase of ~3 weeks, the planktotrophic veliger larva (a kind of larva common in mollusc species) settle onto a hard substrate such as a rock or the shell of another oyster (Figure 2). The oyster then becomes attached to the substrate via its lower valve by secreting a cement, and as the shell grows it will gradually assume the shape of the object to which it is attached (Gosling 2003, Reise 1998). The Pacific oyster is a filter- feeding species, mainly feeding off plankton and detritus, and can reach a length of 50 mm within its first year of settlement (Diederich 2006).
7
Figure 2. Life cycle of oviparous bivalve filter-feeders, of which M. gigas is one example. Figure credit: (Troost 2010).
Inhabiting the intertidal zone, the Pacific oyster is tolerant to changing abiotic environmental conditions, such as being periodically surface-exposed. Although not fully understood, the observed temperature span for survival ranges from sub-zero temperatures to roughly 30˚C.
(Le Gall & Raillard 1988, Bougrier et al. 1995). Additionally, ecological niche modelling predicts that the Pacific oyster can subsist in areas where surface seawater temperatures range from -2 to 30˚C and air temperatures from -23 to 31˚C, respectively (Carrasco & Barón 2010). Daily measurements from Tjärnö Marine Laboratory reveal that sea surface temperature at 1 m depth range between -2 and 24˚C, and measurements from the Koster islands show that air temperature ranges between -12 and 28˚C (SMHI). Severe winter conditions have been observed to have a negative effect on Pacific oyster survival in Sweden (Strand et al. 2012), but the long-term population-level effects of these conditions in
Scandinavia remain largely unknown.
8
A study compiling Pacific oyster data from independent surveys carried out in Sweden, Denmark and Norway suggests that winter mortality increased with latitude, which is likely explained by colder climates (Strand et al. 2012). The study also found site-specific
conditions such as depth to correlate with mortality. Following the winter of 2009/2010, which imposed especially severe conditions, mortality was found to be high among populations of Pacific oysters. Large numbers of the species still exist in Scandinavia, however, suggesting that while harsh winter conditions may result in a temporary decline of oyster numbers, the species can persist and continue its range expansion in Scandinavia (Strand et al. 2012, Durkin, Laugen, Strand et al. unpublished data). Furthermore, a similar study of Pacific oyster mortality in the Wadden Sea following the ice winter 2009/2010 found mortality rates reaching up to about 90% (Büttger et al. 2011). These losses were
comparatively much higher than in the milder winters of previous years and suggest that while Pacific oysters are considered highly frost-tolerant, enduring cold-water conditions accompanied by stress induced by extensive ice cover can have detrimental effects of oyster survival (Büttger et al. 2011). Additionally, Pacific oysters exposed to simulated winter conditions in the lab have been found to be very tolerant to low temperatures, with 50% of individuals surviving -22°C after 24 hours (Strand et al. 2011).
Air and sea surface temperatures in Scandinavia generally range from -12 to 30°C and -1 to 24°C, respectively, and extended periods of sub-zero temperatures are commonplace (DMI, SMHI, MET). Understanding how these climatic conditions affect populations of the invasive Pacific oyster is therefore essential for evaluating the species’ potential for continued survival and proliferation in Northern Europe. An ongoing, multi-year study by Laugen et al.
(unpublished) has in three separate field studies collected data on winter mortality of Pacific oysters from the Swedish west coast from years of both extreme winter conditions
(2009/2010 and 2010/2011) and normal winter conditions (2012/2013). In addition to
determining annual winter survival, the study aims to estimate the water depth at which most oysters settle and relate this to the probability of winter survival, as well as potentially assessing recolonization success following a winter-induced mortality event. The results of this study will ultimately serve to evaluate the extent to which harsh winter conditions may impose constraints on population-level survival of Pacific oysters in Sweden. Owing to exposure of low seawater temperatures and mechanical stress from ice formation during winter, the hypothesis for this study that the depth distribution of oysters will influence survival. Individuals living at shallower water levels are hypothesized to experience more extreme winter-induced stresses and would therefore be subject to higher mortality rates. In contrast, individuals living at greater depths are expected to exhibit lower mortality rates due to milder exposure to the winter conditions. It was found in previous years of this multi-year study that the thick ice cover in 2011 compared to 2013 increased the depth at which
mortality occurred. Findings from previous years of this multi-year study showed that survival probability increased with settlement at greater depths. Following the winter of 2009/2010, the depth at which 50% and 90% of oysters survived was estimated to -69.1 and - 80.2 cm, respectively. In contrast, the equivalent depths for 2012/2013 were -19.7 and -25.7 cm (Laugen et al., unpublished data).
Since the conditions during the winter of 2020/2021 could be considered more extreme compared to several preceding years, winter mortality of the Swedish Pacific oyster populations was again investigated. With an average air temperature of 0.4 °C during the
9
months December to February, the winter of 2020/2021 was on average colder than several preceding winters (Table 2.). Compared to the winters of 2009 through 2013, after which the previous surveys of this multi-year study were performed, the winter of 2020/2021 was not as severe as those of 2009/2010 and 2010/2011 but neither as mild as that of 2011/2012 nor 2012/2013. Thus, the main aim of the current project was to provide an estimate of the relationship between settlement depth of oysters and survival probability in a year with medium cold winter temperatures. The secondary aim of the project was to develop standard procedures for analysing survival probability of oysters in Scandinavia using generalised linear mixed models.
The 2021 field study and subsequent data analyses form the basis for this bachelor’s degree project.
Methods
Sampling locations
We carried out surveys of Pacific oysters between the 28th of March and the 4th of April 2021 off the coast of Bohuslän, Sweden (Figure 5). A total of 14 study sites were surveyed, eight of which had been used as study sites in all previous years of this multi-year study (Table 3).
Most sites were centred around the area of the island of Tjärnö (N 58° 56.2', E 11° 10.4'), with the remaining sites scattered along the west coast. The study sites were selected based on the criteria that 1) the number oysters exceeded 100 individuals, and 2) the maximum density of oysters reached at least 10 individuals × m-2 (Table 3). The sampling locations varied in levels of wind and wave exposure, average depth, and various other abiotic
conditions. Sampling sites ranged from shallow island shores or bays to narrow sounds with sandy or soft bottoms. Oysters were found in high densities at all locations, usually alongside lower densities of other bivalve species such as blue mussels (Mytilus edulis) and European flat oysters (Ostrea edulis).
10
Figure 3. A typical day in the field conducting surveys at an oyster bank. Sampling location: Öddö.
Figure 4. A typical Pacific oyster (Magallana gigas) bank partially surface-exposed at low tide. Sampling location:
11
Figure 5. Map of the Swedish west coast showing oyster sampling locations.
12 Sampling methods
At each site, we laid out two to three parallel transects depending on its size. Every transect started close to shore, or sometimes on the shore if any oysters were found there. The transect then continued in a straight line in a given direction, such as towards a landmark on the opposite shore in the case of narrow sounds, until no more oysters could be found, or it reached the opposite shore. For every transect meter a metal frame was placed, within which the water depth and total number of live and newly dead oysters were recorded. A 0.25 m2 square was used for areas with higher oyster densities and a 0.5 m2 square for areas with lower densities. Oysters were classified as having died during the previous winter if the shells were hinged and the insides were clear without any sediment, algal growth, or fouling. In many cases dead oyster tissues remained attached to the valve’s interior. In the case of the sites Svallhagen, Skredsvik, Kockholmen, Getevik and Smalsundet oyster sizes were also measured. Using callipers, we measured the length from umbo to the edge of the shell, width from one side of the shell to the other perpendicular to the first measurement, and depth as the broadest part along the length of the shell.
Note that what is referred to as 'dead' oysters throughout this data analysis does not refer to the total number of dead individuals found at a particular sampling location, but only those determined to be newly deceased (i.e., did not survive the last winter). This was evaluated based on the overall condition of the animals, such as whether there were any dead tissues remaining, the amount of epibiont growth on the inside of oyster shells and the amount of sediment accumulated inside, as estimates of how long a given oyster had had its valves open.
Depth correction
To compare the distribution of winter survival of oysters across different sites and time periods, the square-specific depth data recorded had to be adjusted according to the average water level at the time of sampling within a given area. To account for tides and differences in atmospheric pressure across time, data on average water level from the Swedish
Meteorological and Hydrological Institute (SMHI) was used. For correction of the southern sites Getevik and Smalsundet, average water level data from the Smögen station was used, and for the remaining northern sites data from the Kungsvik station was used. Data on average water level was then plotted against the time-period over which a survey at a given site was performed, followed by the generation of a well-fitting model for calculating standardized water level (Figure 6). The model was produced using the trendline function in Excel and R2 values were used to determine the goodness-of-fit for each model. In some cases, the sampling period from a given site spanned across local maxima or minima and resulted in poor model fits. This was resolved by splitting the function at the extremum and generating two separate well-fitting functions for the periods before and after the extremum, respectively. The best-fitting function was then used to produce square-specific values for corrected depth using the time recorded for sampling of each square as the predictor variable.
The corrected depth values then replaced the recorded depths throughout the data analysis.
13
Figure 6. Upper plot: Example of the local fluctuations in average water level across a 24-hour period at one of the sampling sites; Smalsundet, May 20, 2021. Lower plot: Example of how a model of good fit was generated for calculating standardized water level; Smalsundet, May 20, 2021. The regression line equation was
subsequently used for calculating water level.
14
Table 1. Example of how corrected depth was calculated using the cubic equation derived from the trendline of best fit from Figure 6.
Time (UTC)
Recorded depth (cm)
Calculated water level (cm) Corrected depth (cm) 09:19
09:23
-12
-14
= -4646,4*Time3 + 4470,5*Time2 - 1190,3*Time + 66,334 = 6,14
= -4646,4*Time3 + 4470,5*Time2 - 1190,3*Time + 66,334 = 6,63
= Recorded depth +
Calculated water level = -5,86
= -7,37
Climate data
A species ability to survive and reproduce within a given environment is influenced by a wide array of abiotic conditions. To understand the impact of environmental conditions on winter mortality of Pacific oysters in Sweden, we compiled data on air temperature, sea surface water temperature, and water level and analysed them alongside the results obtained in this study (Table 2).
Table 2. Climate data (water temperature (°C) at 1 m depth at Tjärnö Marine Laboratory, air temperature (°C) at the Koster islands outside Strömstad, and deviation of sea surface water level (cm) from the yearly mean water level measured at Kungsvik) for summer (June-August) and winter (December-February) for the years 2009- 2021. Data on water temperature and water level from the years 2013-2020 missing. Sea surface temperature data from Tjärnö Marine Laboratory, air temperature and water level data from SMHI.
Water temperature (oC)
Air temperature (oC) Water level (cm)
Year Season Average Max Min Average Max Min Average Max Min
2009/2010 Winter 2 8 -1 -3 7 -15 -8 47 -73
2010/2011 Winter 1 6 -1 -3 4 -14 -8 86 -83
2011/2012 Winter 3 8 -1 2 10 -14 8 117 -73
2012/2013 Winter 2 9 -1 -2 6 -12 -14 69 -89
2013/2014 Winter 3 9 -10
2014/2015 Winter 3 8 -8
2015/2016 Winter 1 11 -12
2016/2017 Winter 2 10 -9
2017/2018 Winter 0,8 9 -13
2018/2019 Winter 5 9 -5
2019/2020 Winter 2 10 -9
2020/2021 Winter 0.4 8 -10 -5 80 -81
15 Data analysis and visualisation
I plotted the mean survival probability at each site against the mean, maximum, minimum depths recorded at each site. To illustrate the effect of depth on survival, I also plotted mean survival against settlement depth for each sampling site. The variation indicated that there were significant differences between sites in how survival related to settlement depth.
Moreover, I was not interested in the parameter estimates for specific sites per se but
recognized that there was variation between sites that had nothing to do with depth, but which still influence mortality/survival rates. Thus, I subsequently analysed the effect of depth on winter survival while controlling for the variation due to site-specific characterizations by fitting site as a random effect.
To evaluate the effect of settlement depth on oyster survival probability, I applied a
generalized linear mixed-effects model as implemented in the lme4 package in R (Bates et al.
2015, version 4.0.5; R Core Team 2021). Survival probability being a binomially distributed variable with possible values from 0-1 (i.e., proportion data), I applied binomial family structure and a logit link function when developing the model. A generalized linear model (GLM) compares the transformed values specified by a link function, in this case a logit link function, from a linear predictor to the observations. A linear mixed model (LMM) is an extension of the simple linear model where random effects are fitted alongside fixed effects.
This acknowledges that the observations are not necessarily independent and means that there are fewer degrees of freedom than what the data suggests. For this analysis, we used a
combination of a generalized linear model and a linear mixed model to create a generalized linear mixed-effects model. This approach acknowledges both the binomially distributed proportion response variable (survival probability) and the possibility that mortality differs between sampling sites. Oysters sampled from within the same site reside under similar conditions and are thus not entirely independent.
Using the glmer command in the lmer4 package in R (version 4.0.5; R Core Team 2021) I fitted winter survival as a two-vector combination of the numbers of live (successes) and dead (failures) Pacific oysters per square (Crawley 2013) as the response variable. Since survival was a binomially distributed variable with possible values of 0 (dead) or 1 (alive), I applied a binomial family structure and a logit link function. As a fixed-effect explanatory variable I first fitted settlement depth (cm) as a continuous explanatory variable. I
subsequently fitted both depth and site as random effect variables allowing both the intercepts and slopes in the logistic regression of survival to vary between sites. This was done because estimating site-specific parameters by fitting site as a fixed effect was not interesting in this context. However — as stated above — the exploratory plots indicated that accounting for variation in intercepts and slopes across sites were necessary. Moreover, fitting site as a random effect is appropriate as oysters sampled from the same site reside under similar conditions and observations are thus not entirely independent.
To illustrate how the model fit our observations, I visually compared the predictions from three sets of models; 1) a simple logistic regression model containing no random effects and pooled information from all sampling sites to fit a single regression line for a general effect of settlement depth on survival, 2) a random-effects model fitting an individual regression line for each site (random intercept and slope) but without fitting a general effect of depth and thus not pooling any of the data, and 3) a mixed-effects model with so-called partial pooling
16
where depth was fitted as both a fixed and random effect and site as a random effect with varying intercept and slope (for R code, see Appendix 1).
Results
The mean survival for each site is given in Table 3 and Figure 7. This shows a relatively large variation in survival among sites; from 98 % at the site Kockholmen to 62 % at Trälsundet.
The calculated mean survival rate for the entire survey was 87%.
Table 3. Coordinates, number of squares used, mean recorded depth and mean survival rates for each of the 14 sampling sites surveyed in this study.
Site (abbreviation) Lat Long Number of
squares
Mean depth (cm)
Mean survival
Långörännan N (LanN) 58.9512° 11.1143° 37 -69.3 0.95
Långörännan S (LanS) 58.9491° 11.1159° 49 -31.8 0.93
Trälsundet (Tral) 58.9129° 11.1952° 30 -20.9 0.63
Öddö (Oddo) 58.9115° 11.1564° 41 -34.0 0.86
Saltösundet (Salt) 58.8743° 11.1458° 60 -18.3 0.92
Svallhagen (Sval) 58.8685° 11.1552° 34 -51.5 0.86
Krokesundet N (KroN) 58.8637° 11.1779° 35 -33.4 0.69
Krokesundet S (KroS) 58.8617° 11.1747° 56 -39.5 0.88
Kollholmen (Koll) 58.8620° 11.1659° 71 -39.6 0.97
Rossö (Rosso) 58.8573° 11.1856° 44 -44.7 0.83
Kockholmen (Kock) 58.8325° 11.1415° 53 -43.8 0.98
Skredsvik (Skre) 58.4873° 11.3100° 29 -26.7 0.79
Getevik (Gete) 58.2754° 11.5061° 44 -31.6 0.71
Smalsundet (Smal) 58.2490° 11.4401° 59 -27.9 0.83
17
Figure 7. Mean survival (and 95 % binomial confidence interval) of Pacific oysters across 14 sampling sites on the Swedish west coast after the 2020-2021 winter. The sites are sorted by order of decreasing survival probability. See Table 3 for site abbreviations.
18
Figure 8. Mean (A), minimum (B), and maximum (C) depths (cm) recorded at the 14 sites, plotted against survival probability. Trendlines were applied using the stat_smooth command in R.
19
Our observations demonstrate that survival probability of Pacific oysters generally increased with greater settlement depth (Figure 8). However, noticing that the site-specific relationship between the depth and survival differed depending on which depth variable were used (see site name labels in Figure 8), more detailed plots and analyses were necessary. By plotting survival against settlement depth separately for each site (Figure 9), a more complex pattern was detected. Firstly, while most of the sites displayed the expected negative relationship between the two parameters, in one of the populations (LanN) the relationship was positive.
Secondly, even among the sites with the expected negative slope, there was substantial variation in trendline intercepts and slopes, which suggested that winter mortality of Pacific oysters may be influenced by other environmental factors in addition to depth and warranted a mixed-model approach (Figure 9).
Figure 9. Subplots for each sampling site with data on survival probability (logit-transformed) as a function of depth. A regression line (orange) has been fitted to each subplot using the stat_smooth() command in the ggplot2 package to illustrate the overall trend. See Table 3 for site abbreviations.
20
Thus, to account for the variation observed between sites, I fitted a generalised linear mixed model to the data. The GLMM revealed a significant effect of depth on survival whereby the probability of oyster survival increased with settlement at greater depths (Table 4). The random intercept and slope coefficients generated by the model varied substantially between sites (Figure 10).
Table 4. Results from the generalized linear mixed model generated of the effects of settlement depth on survival probability of Pacific oysters (Magallana gigas) in Sweden in 2021.
Fixed effects Estimate ± stderr z-value p-value
Intercept 0.936056 ± 0.402274 2.327 0.02
Depth -0.037171 ± 0.009075 -4.096 4.21e-05
Random effects
Groups Name Var ± stderr
Site Intercept 1.8926097 ± 1.3757
Depth 0.0008759 ± 0.0296
21
Figure 10. Subplots for each sampling site with data on survival probability (logit-transformed) as a function of depth. The “No pooling” model fits a separate regression line for each site, i.e., random intercept and slope. The
“Complete pooling” model has combined the information from all sites to fit a single static line, i.e., no random effects. The “Partial pooling” model has pooled information from all lines of the previous two models together to produce better estimates of each individual line. See Table 3 for site abbreviations. R code adapted from Tristan Mahr’s post “Plotting partial pooling in mixed-effects models”, URL: https://www.tjmahr.com/plotting- partial-pooling-in-mixed-effects-models/
Due to significant variation in both intercept and slope between sampling sites, the model containing both intercepts and slopes as random effects (called “Partial pooling” in Figure 10) was chosen for producing parameter estimates (Table 3) as it appeared to best fit our observations (Figure 10). The extensive individual variation found between these estimates suggests that the individual depth-survival profiles are dependent on a variety of biotic and/or abiotic factors.
22
Discussion
The 87 % survival rate found in the 2021 survey was relatively high compared to data collected in previous years that has been analysed so far (Laugen et al. unpublished data).
Results from the survey conducted in 2010 showed that only 13% of Pacific oyster survived the winter of 2009/2010, but 99% in 2010/2011 despite similar climatic conditions to the preceding winter. A likely explanation for the extremely high survival in the winter 2010/2011 was that the high mortality event in 2009/2010 resulted in low recruitment at shallow areas in the following spawning season, and/or that many oysters in shallower waters were wiped out the year before. In contrast, the winter of 2019/2020 was relatively mild (Table 1), and the low mortality rate observed for the winter 2020/2021 was likely due to other factors. These results suggest that the winter conditions of 2020/2021, with average and minimum air temperatures of 0.4 and -10 oC respectively (Table 2), were severe enough to inflict some damage to the Swedish oyster population. However, it appears that this winter was not enough to cause a massive mortality event such as the one observed in the winter of 2009/2010, which sported average and minimum temperatures of -3 and -15 oC respectively (Laugen et al. unpublished data, SMHI). Unfortunately, data on sea surface temperatures for the winter 2020/2021 measured at the Koster islands is missing from SMHI and thus air temperature was used for comparing the severity of winters.
Analyses from previous years showed that winter mortality rates of oysters did not differ notably between the two main study areas: Tjärnö in the North and Kristineberg in the South (85% and 89%, respectively). Additionally, although wave exposure differed between the surveyed sites, it was found not to significantly affect winter mortality (Laugen et al.
unpublished data). Study areas and site-specific wave exposure were therefore omitted from the final model used for generating parameter estimates in previous years of this study.
The model applied in the data analysis for estimating the effect of settlement depth of oyster survival probability was a generalised linear mixed-effects model with correlated random intercept and random slope (Bates et al. 2015). This model corresponds to, and is illustrated by, the “Partial pooling” model seen in Figure 10. In a random intercepts and random slopes model, both coefficients are allowed to vary across groups, which in this case would be sampling sites. Albeit the most complex kind of mixed effects model, it is arguably the most realistic. It produces better estimates for different contexts and each individual case in exchange for less predictive power in the more general case. The fixed effects model, as illustrated by the “Complete pooling” model in Figure 10, pools information from all groups (sampling sites) together to make predictions of the general case, whereas the “No pooling”
model with both random intercepts and slopes generates estimates for each individual group independently of one another. In contrast, the “Partial pooling” model can be thought of as a combination of the two and gathers information from both preceding models to improve the overall estimates. The site-specific regression line generated for Långörännan N (abbreviated LanN) by the “No pooling” model (Figure 10) has a positive slope, meaning that the model estimated that survival probability would decrease with greater depth. This is opposite to what was hypothesized and does not make much sense intuitively. One explanation for this is that more oysters were sampled at shallower depths combined with the fact that the mean survival at this site was very high (>95%). This is supported by the “Partial pooling” model which, after taking the results from other sites into account, estimates that the slope should be negative. This is in accordance with the initial expectations. In summary, this model produces
23
partially group-specific parameter estimates while retaining some predictive power of the fixed effects model. As illustrated in Figure 10, the “Partial pooling” model fits the observations relatively well, hence the choice of this model for the data analysis.
In general, there was a negative correlation between survival probability and the mean settlement depth within sampling sites. This negative correlation supports the hypothesis and is also in accordance with our observations and findings from previous years of this multi- year study (Laugen et al. unpublished), as well as other independent studies (e. g., Strand et al. 2012). However, there was considerable variation involved as winter mortality differed significantly between sampling sites. For instance, despite the sites Saltösundet and
Trälsundet having similar mean depths of 18 and 21 cm respectively, the survival probability varied greatly with rates of 92% and 63% respectively (Table 2). These findings suggest that other contributing factors are involved, such as differences in wind and wave exposure, ice cover during the last winter, and general oyster density. Another possible explanation for these findings stems from the fact that the general mortality rate was found to be markedly low at certain sampling sites, including Saltösundet (Table 2, Figure 7). A low general mortality rate coupled with a shallower mean depth or a high-density oyster bank might help explain some of this variation.
As an intertidal species, Pacific oysters frequently become exposed to the surface during low water levels. During the survey, square-specific depth was recorded using a yardstick.
However, measuring the height above sea-level during low tide for surface-exposed
individuals was difficult and, in these cases, the recorded depths were estimates rather than actual measurements. This is a potential source of error in our data.
There is a correlation between the random intercepts and random slopes from the model used in this data analysis. These estimates also relate to survival probability, where lower survival rates entail steeper negative slopes and smaller intercept values. Three sampling sites stand out in this respect and exhibit the highest mortality rates, namely Trälsundet, Krokesundet N, and Getevik (Figure 7, Table 3). While these three sites do have relatively shallow mean depths, our analysis suggest that there is no significant correlation between mean depth and survival probability. Moreover, these sites are relatively well protected from wind and wave exposure which may have facilitated more extensive ice cover leading to greater winter mortality. However, this is inconsistent with our observations from other sites such as Saltösundet which despite being well protected still exhibited a survival rate of 92%. Our results suggest that mortality rates differ significantly between geographical locations, but whether it is largely affected by these conditions, the settlement depth distribution, or an interaction between the two, is yet to be resolved.
As an invasive and highly adaptable species, the consequences of the recent northward range expansion and successful establishment of Pacific oysters in Scandinavian waters are yet to be fully evaluated. The cold seawater temperatures having previously been thought unsuitable for the species’ proliferation at northern latitudes, the current settlements of Pacific oysters in the region shows that its spreading potential and tolerance to abiotic conditions has been underestimated. We might predict for instance that the warm waters of the gulf stream may allow the species to spread further north, while the lower salinity of the Baltic Sea may prove intolerable. More research is needed to fully evaluate the fundamental niche of Pacific oysters make more accurate predictions of its future distribution. However, extreme winter
24
conditions such as the those experienced in 2009/2010 have shown to push the Pacific oysters towards their lower thermal distribution limits. Furthermore, our findings suggest that normal climatic conditions do not cause significant problems for the species’ survival and
reproduction in the region. Thus, while temporary reductions in population numbers are possible, the species seems unlikely to be eradicated as a result of the Scandinavian climate.
Climate change is predicted to reduce the time of the species to reach maturity and accelerate its expansion considerably in the coming decades (King et al. 2021). The recent range
expansion of the Pacific oyster is one example of how global warming and biological
invasions have the potential of changing ecosystems worldwide. While the Pacific oyster has the potential of compromising local biodiversity, the species may also provide useful
ecosystem services through oyster fisheries and cleaning water by filter-feeding. Whether the effects of its invasion to a non-native environment are deemed net positive or negative is highly contextual, however. Monitoring the distribution changes in Pacific oyster populations is therefore essential for evaluating its effects on both ecosystems and human society, and studies like the one described here may play an important part in that process.
25
Acknowledgements
Many thanks to my supervisor Ane Timenes Laugen for arranging this project and for allowing me to come along for an exciting week of fieldwork to Tjärnö Marine Research Station on the beautiful west coast. Thanks for your guidance, helpful meetings, and teaching me much about coding in R and good research practices.
I want to give a big thank you to my fieldwork assistants Mariela Johansson, Kari Løe and Magnus Jansson for your hard work and great company. This project would not have been possible without you. Thanks also to Luc Bussiére for excellent cooking.
26
References
Andrews JD. 1980. A critique of the introduction of Crassostrea gigas to Europe. Marine Resource Report, doi 10.25773/V5-BDQ7-VZ69.
Bates D, Mächler M, Bolker B, Walker S. 2015. Fitting linear mixed-effects models using lme4.
Journal of Statistical Software, doi 10.18637/jss.v067.i01.
Bougrier S, Geairon P, Deslous-Paoli JM, Bacher C, Jonquières G. 1995. Allometric relationships and effects of temperature on clearance and oxygen consumption rates of Crassostrea gigas (Thunberg). Aquaculture 134: 143–154.
Büttger H, Nehls G, Witte S. 2011. High mortality of Pacific oysters in a cold winter in the North- Frisian Wadden Sea. Helgoland marine research 65: 525–532.
Cardoso JFMF, Langlet D, Loff JF, Martins AR, Witte JI, Santos PT, Veer HW van der. 2007. Spatial variability in growth and reproduction of the Pacific oyster Crassostrea gigas (Thunberg, 1793) along the west European coast. Journal of Sea Research 57: 303–315.
Carrasco MF, Barón PJ. 2010. Analysis of the potential geographic range of the Pacific oyster Crassostrea gigas (Thunberg, 1793) based on surface seawater temperature satellite data and climate charts: the coast of South America as a study case. Biological Invasions 12: 2597–
2607.
Cognie B, Haure J, Barillé L. 2006. Spatial distribution in a temperate coastal ecosystem of the wild stock of the farmed oyster Crassostrea gigas (Thunberg). Aquaculture 259: 249–259.
Crawley MJ. 2013. The R book. Wiley, Chichester.
Diederich S. 2006. High survival and growth rates of introduced Pacific oysters may cause restrictions on habitat use by native mussels in the Wadden Sea. Journal of Experimental Marine Biology and Ecology 328: 211–227.
Diederich S, Nehls G, Beusekom JEE, Reise K. 2005. Introduced Pacific oysters in the northern Wadden Sea: invasion accelerated by warm summers? Helgoland marine research 59: 97.
Dolmer P, Holm M, Strand Å, Lindegarth S, Bodvin T, Norling P, Mortensen S. 2014. The invasive Pacific oyster, Crassostrea gigas, in Scandinavian coastal waters: A risk assessment on the impact in different habitats and climate conditions. Fisken og Havet 2: 1–67.
Drake JM, Lodge DM. 2004. Global hot spots of biological invasions: evaluating options for ballast–
water management. Proceedings of the Royal Society B, Biological sciences 271: 575–580.
Dupuy C, Gall SL, Hartmann HJ, Bréret M. 1999. Retention of ciliates and flagellates by the oyster Crassostrea gigas in French Atlantic coastal ponds: protists as a trophic link between bacterioplankton and benthic suspension-feeders. Marine Ecology Progress Series 177: 165–
175.
Eklund U, Håkansson M, Haamer J (1977) En undersökning om förutsättningarna för ostronodling vid svenska västkusten. Chalmers Tekniska Högskola och Göteborgs Universitet, publ. No. B 83, Göteborg. 35 pp (in Swedish)
27
Fabioux C, Huvet A, Souchu PL, Pennec ML, Pouvreau S. 2005. Temperature and photoperiod drive Crassostrea gigas reproductive internal clock. Aquaculture 250: 458–470.
© FAO 2005-2021. Cultured Aquatic Species Information Programme. Crassostrea gigas. Cultured Aquatic Species Information Programme. Text by Helm, M.M. In: FAO Fisheries
Division [online]. Rome. Updated. [Cited 29 April 2021].
Faust E, Andre C, Meurling S, Kochmann J. 2017. Origin and route of establishment of the invasive Pacific oyster Crassostrea gigas in Scandinavia. doi 10.3354/meps12219.
García L. M, Mårdberg M, Nilén S, Schlyter P. 2018. Effects of salinity levels on the distribution and size of the Pacific oyster (Crassostrea gigas) along the west coast of Skåne, Sweden. Project, Water Management BIOR66, Lund University.
Gosling E. 2003. Bivalve molluscs: Biology, Ecology and Culture. Blackwell Publishing, Oxford (2003)
Grizel H, Héral M. 1991. Introduction into France of the Japanese oyster (Crassostrea gigas). ICES Journal of Marine Science 47: 399–403.
Grosholz E. 2002. Ecological and evolutionary consequences of coastal invasions. Trends in Ecology
& Evolution 17: 22–27.
Guisan A, Thuiller W. 2005. Predicting species distribution: offering more than simple habitat models. Ecology Letters 8: 993–1009
Katsanevakis S, Zenetos A, Belchior C, Cardoso AC. 2013. Invading European Seas: Assessing pathways of introduction of marine aliens. Ocean & Coastal Management 76: 64–74.
Kawecki TJ. 2008. Adaptation to Marginal Habitats. Annual review of ecology, evolution, and systematics 39: 321–342
King NG, Wilmes SB, Smyth D, Tinker J, Robins PE, Thorpe J, Jones L, Malham SK. 2021. Climate change accelerates range expansion of the invasive non-native species, the Pacific oyster, Crassostrea gigas. ICES Journal of Marine Science 78: 70–81.
Kobayashi M, Hofmann EE, Powell EN, Klinck JM, Kusaka K. 1997. A population dynamics model for the Japanese oyster, Crassostrea gigas. Aquaculture 149: 285–321.
Kolar CS, Lodge DM. 2001. Progress in invasion biology: predicting invaders. Trends Ecol Evol 16: 199–204, doi 10.1016/S0169-5347(01)02101-2
Krassoi FR, Brown KR, Bishop MJ, Kelaher BP, Summerhayes S. 2008. Condition-specific
competition allows coexistence of competitively superior exotic oysters with native oysters.
Journal of Animal Ecology 77: 5–15.
Laugen A, Fredricsson Y, Hollander J, Lindegarth S, Strand Å. In prep. Depth-dependent winter mortality and consequences for site specific population survival of the Pacific oyster (Magallana gigas) at its northern range limit.
Laugen A, Hollander J, Obst M, Strand Å. 2015. The Pacific Oyster (Crassostrea gigas) invasion in Scandinavian coastal waters: impact on local ecosystem services.
Le Gall J-L, Raillard O. 1988. Influence de la température sur la physiologie de l’huître Crassostrea.
Océanis 14: 603–608.
28
Mahr T. 2017. Plotting partial pooling in mixed-effects models. WWW document 22 June 2017:
https://www.tjmahr.com/plotting-partial-pooling-in-mixed-effects-models/. Accessed 1 May 2021.
Mann R. 1979. Some biochemical and physiological aspects of growth and gametogenesis in Crassostrea gigas and Ostrea edulis grown at sustained elevated temperatures*. Journal of the Marine Biological Association of the United Kingdom 59: 95–110.
Mann R, Burreson E, Baker P. 1991. The decline of the Virginia oyster fishery in Chesapeake Bay Considerations for introduction of a non-endemic species, Crassostrea gigas (Thunberg, 1793). Journal of Shellfish Research 10: 379–388.
Meyers LA, Bull JJ. 2002. Fighting change with change: adaptive variation in an uncertain world.
Trends in Ecology & Evolution 17: 551–557.
Molnar JL, Gamboa RL, Revenga C, Spalding MD. 2008. Assessing the global threat of invasive species to marine biodiversity. Frontiers in Ecology and the Environment 6: 485–492.
Muranaka MS, Lannan JE. 1984. Broodstock management of Crassostrea gigas: Environmental influences on broodstock conditioning. Aquaculture 39: 217–228.
Persson M, Karlson, B, Zuberovic Muratovic A, Simonsson M, Bergkvist, P, Renborg E (2020) L 2020 nr 24: Kontrollprogrammet för tvåskaliga blötdjur, Årsrapport 2014-2019.
Livsmedelsverkets rapportserie. Livsmedelsverket, Uppsala
Piersma T, van Gils JA. 2010. The Flexible phenotype: a body-centred integration of ecology, physiology, and behaviour. Oxford University Press, Oxford
Quayle DB. 1988. Pacific oyster culture in British Columbia. Department of Fisheries and Oceans R Core Team. 2021. R: A language and environment for statistical computing. R Foundation for
Statistical Computing, Vienna, Austria. URL https://www.R-project.org/
Reise K. 1998. Pacific oysters invade mussel beds in the European Wadden Sea. Senckenbergiana maritima 28: 167–175.
Reise K, Buschbaum C, Büttger H, Rick J, Wegner KM. 2017a. Invasion trajectory of Pacific oysters in the northern Wadden Sea. Marine Biology 164: 68.
Reise K, Buschbaum C, Büttger H, Wegner M. 2017b. Invading oysters and native mussels: From hostile takeover to compatible bedfellows. Ecosphere 8: e01949.
Ruesink JL, Lenihan HS, Trimble AC, Heiman KW, Micheli F, Byers JE, Kay MC. 2005.
Introduction of non-native oysters: ecosystem effects and restoration implications. Annual Review of Ecology, Evolution, and Systematics 36: 643–689.
Sexton JP, McIntyre PJ, Angert AL, Rice KJ. 2009. Annual Review of Ecology, Evolution, and Systematics 40:1, 415-436
Soberón J, Peterson AT. 2005. Interpretation of Models of Fundamental Ecological Niches and Species’ Distributional Areas. doi 10.17161/bi.v2i0.4.
Strand Å, Blanda E, Bodvin T, Davids JK, Jensen LF, Holm-Hansen TH, Jelmert A, Lindegarth S, Mortensen S, Moy FE, Nielsen P, Norling PC, Nyberg C, Christensen HT, Vismann B, Holm MW, Hansen BW, Dolmer P. 2012. Impact of an icy winter on the Pacific oyster (Crassostrea gigas Thunberg, 1793) populations in Scandinavia. 433-440, doi 10.3391/ai.2012.7.3.014.
29
Strand Å, Waenerlund A, Lindegarth S. 2011. High tolerance of the Pacific Oyster (Crassostrea gigas, Thunberg) to low temperatures. Journal of Shellfish Research 30: 733–735.
Troost K. 2010. Causes and effects of a highly successful marine invasion: Case-study of the
introduced Pacific oyster Crassostrea gigas in continental NW European estuaries. Journal of Sea Research 64: 145–165.
Vilà M, Basnou C, Pyšek P, Josefsson M, Genovesi P, Gollasch S, Nentwig W, Olenin S, Roques A, Roy D, Hulme PE. 2010. How well do we understand the impacts of alien species on ecosystem services? A pan-European, cross-taxa assessment. Frontiers in Ecology and the Environment 8: 135–144.
Wallentinus I, Nyberg CD. 2007. Introduced marine organisms as habitat modifiers. Marine Pollution Bulletin 55: 323–332.
West-Eberhard MJ. 2005. Developmental plasticity and the origin of species differences. Proceedings of the National Academy of Sciences 102: 6543–6549.
Wiens JJ. 2011. The niche, biogeography and species interactions. Philosophical Transactions of the Royal Society B: Biological Sciences 366: 2336–2350.
Wrange A-L, Valero J, Harkestad LS, Strand Ø, Lindegarth S, Christensen HT, Dolmer P, Kristensen PS, Mortensen S. 2010. Massive settlements of the Pacific oyster, Crassostrea gigas, in Scandinavia. Biological Invasions 12: 1145–1152.
References to R packages used:
Dorai-Raj A. 2014. binom: Binomial Confidence Intervals For Several Parameterizations. R package version 1.1-1. URL: https://CRAN.R-project.org/package=binom
Fox J, Weisberg S. 2019. An {R} Companion to Applied Regression, Third Edition. Thousand Oaks CA: Sage. URL: https://socialsciences.mcmaster.ca/jfox/Books/Companion/
Kassambara A. 2020. ggpubr: 'ggplot2' Based Publication Ready Plots. R package version 0.4.0.
URL: https://CRAN.R-project.org/package=ggpubr
Slowikowski K. 2021. ggrepel: Automatically Position Non-Overlapping Text Labels with 'ggplot2'.
R package version 0.9.1. URL: https://CRAN.R-project.org/package=ggrepel
Wickham et al., 2019. Welcome to the tidyverse. Journal of Open Source Software, 4(43), 1686, URL: https://doi.org/10.21105/joss.01686
30
Appendix 1
The following R code was used for the data analysis and plots featured in this thesis.
Loading libraries.
library(tidyverse) library(binom) library(ggrepel) library(ggpubr) library(car) library (lme4)
Importing the data.
oysurv2021 <- read_csv2("oyster_data_2021.csv") Adding column of survival and omitting NAs.
oysurv2021 <- oysurv2021 %>%
as_tibble() %>%
mutate(survival = live/(live+dead)) %>%
filter(survival!= "NaN") %>%
select(-year)
Converting variables to the right format.
oysurv2021$transect <- as.factor(oysurv2021$transect) oysurv2021$square <- as.factor(oysurv2021$square) oysurv2021$depth <- as.numeric(oysurv2021$depth)
Making Figure 7. Wrangling data for bar plot of mean survival across sites.
oysterbar <- oysurv2021 %>%
group_by(site) %>%
summarise(
live = sum(live), dead = sum(dead) )
oysterCI <- binom.confint(oysterbar$live, (oysterbar$live+oysterbar$dead), conf.level = 0.95, methods = "exact")
oysterbar2 <- cbind(oysterbar, oysterCI) %>%
left_join(oysterbar, by = "site") survival <- ggplot(oysterbar2) +
geom_bar( aes(x=reorder(site,-mean), y=mean), stat="identity", fill="gray60", alpha=0.5) +
geom_errorbar( aes(x=site, ymin=lower, ymax=upper), width=0.4, colour="orange", alpha=0.9, size=1.3) +
labs(x=" Site", y="Mean survival probability")
Making Figure 8. Data wrangling for survival vs depth.
survival_vs_depth <- oysurv2021 %>%
select(site, depth, live, dead) %>%
group_by(site) %>%
31 summarise(
live = sum(live), dead = sum(dead),
mean_depth = mean(depth), survival = live/(live+dead), max_depth = max(depth), min_depth = min(depth) )
Plotting survival vs mean depth.
Plot_A <- ggplot(survival_vs_depth, aes(x=mean_depth, y=survival), stat="identity") +
geom_point(size=1.5) +
geom_smooth(method="lm", color = "darkorange") + geom_label_repel(aes(label = site),
box.padding = 0.35, point.padding = 0.5,
segment.color = 'grey50') +
labs(x="Mean depth (cm)", y="Survival probability")
Plotting survival vs max depth.
Plot_B <- ggplot(survival_vs_depth, aes(x=max_depth, y=survival), stat="identity") +
geom_point(size = 1.5) +
geom_smooth(method="lm", color = "darkorange") + geom_label_repel(aes(label = site),
box.padding = 0.35, point.padding = 0.5,
segment.color = 'grey50') +
labs(x="Min depth (cm)", y="Survival probability")
Plotting survival vs min depth.
Plot_C <- ggplot(survival_vs_depth, aes(x=min_depth, y=survival), stat="identity") +
geom_point(size = 1.5) +
geom_smooth(method="lm", color = "darkorange") + geom_label_repel(aes(label = site),
box.padding = 0.35, point.padding = 0.5,
segment.color = 'grey50') +
labs(x="Max depth (cm)", y="Survival probability")
Combining the three plots with ggarrange.
survival_vs_depth <- ggarrange(Plot_A, Plot_B, Plot_C, ncol = 1, nrow = 3, labels = c('A', 'B', 'C'), align = "v")
ggsave("ggarrange mean, max, min (one column).jpg", width = 6, height = 10)
Making Figure 9. Datawrangling.
32 oysurv21 <- oysurv2021 %>%
mutate(square=as_factor(square), site = as_factor(site),
survival=live/(live+dead)) %>%
filter(survival!="NaN") %>%
mutate(logitsurv=logit(if_else(survival == 0,0.0000001,
if_else(survival == 1,0.999999,survival)))) Plotting survival vs depth in logit space.
survival_by_site <- ggplot(oysurv21, aes(y = logitsurv, x = depth)) + geom_point(size = 0.5) +
stat_smooth(method = "lm", size = .75, color="darkorange") + ylab("Survival probability (logit)") +
xlab("Depth (cm)") + facet_wrap("site")
Making Figure 10. Fitting the no pooling model.
oyster_no_pooling <- lmList(logitsurv ~ depth | site, oysurv21) %>%
coef() %>%
rownames_to_column("site") %>%
rename(Intercept = `(Intercept)`, Slope_depth = depth) %>%
add_column(Model = "No pooling")
Fitting a model on all the data pooled together.
m_pooled <- glm(logitsurv ~ depth, oysurv21, family=gaussian) oyster_pooled <- tibble(
Model = "Complete pooling", site = unique(oysurv21$site), Intercept = coef(m_pooled)[1], Slope_depth = coef(m_pooled)[2]
)
Fitting a partial pooling model.
m <- lmer(logitsurv ~ 1 + depth + (1 + depth | site), oysurv21)
# Making a dataframe with the fitted effects oyster_partial_pooling <- coef(m)[["site"]] %>%
rownames_to_column("site") %>%
as_tibble() %>%
rename(Intercept = `(Intercept)`, Slope_depth = depth) %>%
add_column(Model = "Partial pooling")
Plotting and comparing the three pooling approaches.
oyster_models <- bind_rows(oyster_pooled, oyster_no_pooling, oyster_partial_pooling) %>%
left_join(oysurv21, by = "site")
p_model_comparison <- ggplot(oyster_models) + aes(x = depth, y = logitsurv) +
33 geom_abline(
aes(intercept = Intercept, slope = Slope_depth, color = Model), size = .75
) +
geom_jitter(size = 0.5) + facet_wrap("site") +
labs(x = xlab, y = ylab) +
scale_color_brewer(palette = "Dark2") +
theme(legend.position = "top", legend.justification = "left")
Fitting the generalised linear mixed-effects model with random intercept and random slope.
summary(oymort2021.glmer <- glmer(cbind(live,dead) ~ depth + (depth|site), data=oysurv2021, binomial))
