Notice: Wiley Online Library will be unavailable on Saturday 30th July 2016 from 08:00-11:00 BST / 03:00-06:00 EST / 15:00-18:00 SGT for essential maintenance. Apologies for the inconvenience.
1A number of studies have shown that spring biological events have advanced in recent decades, and concluded that these changes in phenology are driven by climatic change. Freshwater lakes are sensitive indicators of climate change, where direct effects of climate on physical processes can affect the seasonal timing of planktonic communities. However, many lake ecosystems have also experienced long-term changes in other ecological pressures that could affect phenology.
2In this study, long-term (1955–2003) physical, chemical and biological data from Windermere (UK) were analysed in order to assess the relative effects of a number of coincident pressures on the phenology of two spring diatom taxa. The analysis provides a detailed case study, highlighting the species-specific drivers that affect the phenology of dominant members of the phytoplankton community.
3The results showed that, whilst the spring peak biomass of one taxon (Cyclotella) appeared to be advancing as a result of earlier thermal stratification, the advancement of the other (Asterionella) was closely linked with both progressive nutrient enrichment and lake warming. Furthermore, nutrient enrichment explained more variation in phenology than water temperature. Both taxa also reached their peak abundance earlier when the over wintering biomass at the end of the previous year was higher.
4Patterns of change in phenology and ecological pressures were markedly nonlinear in time, as were the effects of some drivers of seasonal timing. This highlighted a need to relax the restriction of linearity in our analyses of biological seasonality.
5Synthesis. Phenological shifts may be brought about by local processes, such as eutrophication, as well as by climate change. Even in the same ecosystem different mechanisms may alter the phenology of different species.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
Despite the apparent simplicity of identifying the timing of the peak spring phytoplankton population, or indeed of any other phenological indicator, the underlying mechanisms that drive these dynamics can be complex. The timing of the peak in phytoplankton abundance represents a pivot point between replication rates and various additive loss processes, such as sedimentation, flushing and grazing pressure (Reynolds 2006). Prior to the peak abundance, the total loss rate is less than the replication rate that is determined by temperature, light and nutrient availability. After the peak, loss rates exceed the replication rate so that the net rate of population change is negative. The timing of the population maximum can, therefore, be determined by an increase in the loss rate relative to the rate of replication, or by a decline in the replication rate relative to the rates of the various loss processes acting upon the population, or both.
Direct climate effects on the phenology of these populations could act by altering the replication rate, thus modifying the length of time needed to reach a given population size and the length of time for which replication exceeds loss rates. Increases in water temperature may directly affect the replication rate (Reynolds, 2006), even under conditions of nutrient limitation (Rhee & Gotham 1981). Direct climate effects might also act to affect the light climate experienced by phytoplankton. For the diatoms that dominate spring blooms in temperate lakes, the replication rate is initially light-limited (Neale et al. 1991a,b; Maberly et al. 1994). Earlier thermal stratification, in response to climatic change, would shorten the period of deep mixing, thus improving the average light conditions experienced by the algae (Huisman et al. 1999; Huisman & Sommeijer 2002), and increasing the light limited replication rate. Provided replication progresses to a set maximum population size, the effect of earlier thermal stratification would be to reduce the time needed for this population size to be reached, bringing about an advancement of the spring population maximum.
A changing climate could also have a direct effect on the timing of the spring peak population by modifying the total loss rate acting on the population, relative to the replication rate. The onset of thermal stratification increases rates of sedimentation for diatoms (Huisman & Sommeijer 2002) as well as improving the light-limited replication rate of the algae by improving the light regime. If this event brought about high sinking losses relative to the rate of replication, we might hypothesize that earlier stratification would cause the sedimentation of spring populations to occur earlier and thus the timing of the peak population density would be advanced. However, if the result of advanced thermal stratification was only a modest increase in sedimentation losses relative to the replication rate then this event would be expected merely to slow the attainment of the maximum population size and might actually delay the spring peak. It is, therefore, possible that the direction of change of diatom phenology in response to advanced stratification would depend upon the extent to which loss processes increased, introducing the potential for nonlinearity in this relationship.
Despite the considerable attention that has been focussed on climatic change as a driver of long-term change in phytoplankton phenology, through its effects on lake physical structure, many lakes have also been subjected to coincident changes in nutrient status. Long-term changes in the abundance and composition of planktonic communities are clearly the result of ecological responses to concomitant changes in both resource supply and physical structure (George et al. 1990; Anneville et al. 2004; Anneville et al. 2005). However, the role of changing resource supply in altering phenology has rarely been considered.
It has been argued that, on theoretical grounds, phenological changes in spring diatom populations could be driven by long-term changes in the availability of key chemical resources (Reynolds 1990, 1997a) and that such mechanisms could produce similar phenological shifts to those that are frequently interpreted with respect to climate variation. During the early spring, phytoplankton can deplete the available phosphorus until concentrations become insufficient to support their light limited replication rate and population growth becomes phosphorus limited. Growth continues, albeit at a lower rate, until a maximum attainable population size is reached. For some diatom populations it has been shown that this, in turn, is determined by the supportive capacity of the available silica, with populations declining when silica is depleted (Lund 1950a,b).
Based on this reasoning, it has been hypothesized that long-term changes in phosphorus availability can cause phenological shifts by influencing the length of time for which spring diatom populations could grow rapidly at light limited rates, free from the growth restraint of phosphorus limitation (Reynolds 1990, 1997a). The corollary is that when phosphorus is more plentiful spring diatom populations would reach their silica-determined maximum (Lund 1950a,b) earlier, due to an extension of the time-period for which high replication rates are permitted under light-limited growth. Within this theoretical framework, variations in silica supply would also affect phenology by altering the maximum population size. If we assume constant net rates of increase, when silica concentrations are lower the reduced maximum population size would be reached earlier in the year. Therefore the timing of the peak population size is determined primarily by a decline in replication rates relative to loss rates, because of the limitation of population growth by insufficient silica concentrations. The time taken to reach this point is itself regulated by the availability of bio-available phosphorus. If there have been long-term changes in the supply of bio-available phosphorus and silica, then this reasoning provides an additional mechanism for advancing spring diatom phenology, in contrast to the often cited affects of climate acting via water temperature and stratification.
There are a number of other possible mechanisms that might affect phytoplankton phenology and that are not a result of direct climate effects on rates of replication and loss. We might expect that the starting population from which spring growth initiates would be of importance to the development of peak population densities. Higher over-wintering populations allow a fixed population maximum to be reached earlier, assuming constant rates of increase. The likely significance of inter-annual variations in over-wintering populations to spring dynamics has already been suggested (Gerten & Adrian 2000; Reynolds & Irish 2000), though not quantitatively evaluated with reference to other potential drivers of phenology.
Assuming that a given diatom species can be ingested by herbivorous zooplankton, we might expect that an increase in grazing losses would have a similar effect on phenology to the establishment of a stable thermocline. A dramatic increase in rates of grazing would cause total loss rates to become higher than replication rates and cause a decline in spring populations (Lampert & Sommer 1997). If the biomass of herbivorous zooplankton increased over time then the total loss rate would also increase. For a nutrient limited spring population with a decreasing replication rate, we might expect that losses would exceed the declining replication rate earlier in the year when zooplankton grazing is higher and cause an advancement of the spring peak population. If the spring zooplankton peak advanced, the spring diatom peak could advance because grazing losses supersede the replication rate before nutrient limitation occurs. However, as was the case for thermal stratification, if the inter-annual increase in grazing was modest then this may merely slow net rates of increase and delay the attainment of the spring peak population. Once again, nonlinearity might be expected in the relationship between grazing pressure and diatom phenology.
Though this list of potential drivers of phytoplankton phenology is not exhaustive, it is clear that there is a significant attribution problem: put simply, are the long-term trends that we observe a function of the direct impacts of changing climate on the lake environment alone or are they the result of non-climate drivers such as changes in the availability of key chemical resources, over-wintering populations and grazing pressure? As yet, no study has attempted to disentangle the effects of these drivers on spring phytoplankton phenology.
Previous studies have shown that patterns of long-term change in phytoplankton dynamics and phenology can be nonlinear (George et al. 1990; Maberly et al. 1994; Winder & Schindler 2004b). However, many investigations into the potential drivers of this change attempt to interpret possible underlying mechanisms using statistical methods that assume linear relationships between response and predictor variables. It is acknowledged that nonlinearities are to be expected in the responses of populations and communities to climatic change (Stenseth & Mysterud 2002). The previous theoretical consideration of the processes affecting diatom dynamics makes it clear that nonlinearities might also be expected in the responses that drive the phenology of these organisms. There is a need to assess whether the assumption of linearity in temporal trends and in ecological relationships is adequate for studies that attribute long-term changes in freshwater ecosystems to variations in specific environmental parameters.
We present a detailed case study of long-term changes in the timing of two numerically abundant and functionally distinct organisms. We test for the possibility that the drivers of changing phenology are not direct affects of a changing climate on lake physical structure, and that the species-specificity of these changes results from differences in the relative importance of underlying drivers. Flexible statistical models are used to evaluate the effects of temporal change in water column structure, trophic state, over-wintering populations and grazing upon the phenology of two diatom genera. We test the hypotheses that (i) functionally distinct species will show different patterns of inter-annual variation in spring phenology; and (ii) differences in phenological trends can be brought about by differences in the underlying drivers of seasonality, some of which are not direct effects of climate on lake physical structure.
Data originate from an ongoing long-term monitoring programme in the pelagic zone of the North Basin of Windermere, UK (54°20′ N, 2°57′ W). The basin covers an area of 8.1 km2 and has a mean depth of 25 m (maximum depth 64 m). The ecology of Windermere has been summarized in Reynolds & Irish (2000). The analysis focuses on the period 1955–2003, during which time the same physical, chemical and biological data have been collected at 1–2-weekly intervals using consistent methods. Data collected during periods of weekly sampling were filtered onto a 2-weekly scale, in order to ensure the same level of sampling effort throughout the study period, and remove biases caused by this variation in protocol.
Integrated surface water samples for the determination of key chemical and biological properties were collected using a weighted plastic tube (Lund 1949). The length of the tube varied over the time period between 5 m (pre 1962), 10 m (1962–64) and 7 m (post-1964) (George et al. 2000). Concentrations of soluble reactive phosphorus (SRP) and silica were determined according to Mackereth et al. (1978). The concentration of chlorophyll a was determined spectrophotometrically according to Talling (1974). Phytoplankton species were identified and enumerated according to Lund et al. (1958).
On each occasion crustacean zooplankton were collected by hauling a net (mesh size 120 µm, mouth diameter 0.3 m) through the water column. Samples were initially fixed with a small quantity of 70% ethanol, before being preserved in 4% formaldehyde. After a period of sedimentation, the volume of zooplankton in each sample was calculated from the diameter of the sample storage vessel and the height of the sedimented zooplankton material.
Temperature data were collected over the deep point of the lake at various depths, typically closely spaced in the summer to resolve effectively the thermocline, but with more spacing in the hibernal months, when the lake is fully mixed. Measurements were taken with a thermistor in the 1950s, a Mackereth oxygen electrode in the 1960s and 1970s and a Yellow Springs Instruments probe since the 1980s (George et al. 2000). To avoid any inherent bias in the data, the raw data were linearly interpolated vertically and then linearly interpolated through time to give temperatures on a 1-m daily grid (see Jones et al. 2008). These interpolated data were used to calculate daily Schmidt stability (Hutchinson 1957), S, given by,
where g is gravity, A0 the surface area of the lake, z the depth, Az the cross-sectional area at depth z, ρmax the maximum density of freshwater, ρz the density of water at depth z, zm the maximum depth of the lake and zg, the centre of gravity of the lake, given by,
where V is the volume of the lake. Depths, areas and volumes of the lake are given in Ramsbottom (1976).
response and predictor variables
The present analysis focuses on the phenology of Asterionella formosa (Hass.) and Cyclotella spp. These taxa were chosen since they are morphologically distinct and, therefore, likely to be distinct in their ecological function. Cyclotella spp. were analysed at the genus level due to the presence of unidentified species in the long-term records, though the data suggest that the assemblage was dominated by Cyclotella comensis Grunow and Cyclotella pseudostelligera Hustedt. In order to place the analysis of these taxa into a wider ecological context, we estimated their contribution to the aggregate diatom biovolume on each sampling date from January to June, and calculated the timing of the spring peak in diatom biovolume. Since biovolume measurements do not exist for the phytoplankton in the Windermere data set we based these calculations on published cellular biovolumes, many of which were derived from Lake District samples (Conley & Kilham 1989; Reynolds 2006). Long-term change in the timing of the phytoplankton peaks was analysed with respect to temporal changes in nutrient concentrations, over-wintering populations, potential grazing pressure, water temperature, sunlight and water column stability. The precise predictor variables used in the analysis are outlined below (see also Table 1).
Table 1. Summary of the final GAM and GLM for the timing of the spring peaks of Asterionella formosa and Cyclotella spp.
Asterionella formosa excl. 1961, 1962, 1966, 1970
Models were produced by stepwise selection based upon AICc values. The degrees of freedom for each smoother term in the GAM is denoted as ‘d.f.’ and the slope coefficient of the GLM is denoted as ‘b’. Statistical significance is judged by F-tests. ‘Drop cont.’ is the drop contribution; the residual deviance after excluding each variable from the model. ‘SS’ denotes Schmidt stability.
The AICc is calculated as AIC + (2k(k + 1))/(n – k – 1) according to Burnham & Anderson (2004) where k = number of model parameters, n = number of observations and AIC = −2log (Likelihood) + 2 × d.f.
Winter mean concentrations of SRP and silica were used as indicators of nutrient availability to the spring diatom assemblage (Talling & Heaney 1988). Here ‘winter’ was defined as the first 10 weeks of the year following George et al. (2000). Long-term change in the potential grazing pressure was quantified using the mean settled volume of zooplankton collected in the standard haul over the period preceding the spring population peak, i.e. from the start of the year until the timing of the spring population peak. In order to consider explicitly time-lagged effects created by the effects of over-wintering abundance on spring dynamics, the mean abundance of each taxon during the last 10 weeks of the previous year was also included in the analysis. This definition is based upon the quintile concept used in George (2000) and George et al. (2000). These data were log-transformed due to the presence of a minority of large, and potentially influential, values in the data set.
The January–May mean water temperature and time at which selected Schmidt stability values were exceeded were used to indicate changes in the physical structure of the lake. Four Schmidt stability thresholds were selected: 20, 50, 150 and 300 J m−2. These corresponded to times at which the water column temperature differences (i.e. temperature at the water surface minus the bottom temperature) were approximately 1.0 °C, 1.5 °C, 3.0 °C and 5.0 °C. The decision to include a range of thresholds was based upon uncertainty a priori of the level of stability that would dramatically affect phytoplankton dynamics and upon the expectation that different stability thresholds would be influential for different phytoplankton taxa, because of interspecific variations in settling velocities (Huisman & Sommeijer 2002). The timing of these thresholds, along with the water temperature, were correlated with winter values of the North Atlantic Oscillation (NAO, Hurrell 1996) in order to evaluate the links between these physical properties and large-scale climate variation.
Since growth is light limited in the early spring in Windermere, we also included data on sunlight in the analysis. Sunlight data (in the form of minutes of bright sunshine per day) were provided by a ‘burning paper’ trace located adjacent to the North Basin of Windermere in Ambleside, Cumbria (George & Hewitt 1998). For the purposes of the analysis, these data were averaged over the first 20 weeks of each year; representing the period of growth for the diatom taxa included in the analysis.
The phenology of the two phytoplankton taxa and of aggregate diatom biovolume was quantified using two methods: the day of the year corresponding to the maximum observed abundance (Weyhenmeyer et al. 1999; Winder & Schindler 2004b) and the central tendency of the phytoplankton peak (Colebrook & Robinson 1965; Edwards & Richardson 2004). Prior to analysis, these data were examined for evidence of temporal autocorrelation using plots of Ljung–Box Q statistics (Ljung & Box 1978). Despite the presence of systematic temporal variation in these data, statistically significant temporal autocorrelation was not found. The two methods of assessing the timing of the spring peak population were compared for each species by paired t-test, in order to establish whether one method produced consistently later/earlier timings than the other. Furthermore, inter-annual variations in timing produced by the two methods were compared by Pearson's correlation.
The relationships between phytoplankton phenology and potential ecological drivers were analysed using general linear models and generalized additive models, with Gaussian errors. The latter permit nonlinear relationships by fitting smoothing functions (Hastie & Tibshirani 1990). Residuals were examined for evidence of departures from normality and homogeneity. Measures of influence (leverage and Cook's Distance) were used to establish whether the resulting models were highly dependent on extreme data points. In order to avoid overfitting by including all predictors in an additive model, each analysis was initially carried out using a general linear model with stepwise selection based upon AICc values. The use of AICc allows the selection of the model with the closest to optimal balance between goodness-of-fit and parameterization, whilst making a correction for small sample size (Burnham & Anderson 2004). The retained predictor variables were then included in an additive model if plots of residuals vs. predictor variables revealed systematic patterns. If the residuals formed systematic patterns with respect to any omitted explanatory variables these were also included in an additive model. When additive models were used, a manual stepwise selection procedure, based on comparison of AICc values, was used to exclude non-significant terms. Additive models were run with 2, 3 and 4 degrees of freedom and the optimal degree of smoothing was then decided by comparing AICc values between models. Residuals were tested for temporal autocorrelation and, where this existed, a first order autoregressive process was included in the final model. Whenever nonlinear relationships were found, their explanatory power was compared to that of a linear relationship, by carrying out an F-test on the difference in the residual sum-of-squares between a model including the nonlinear term and a model including the linear term.
temporal change in the timing of peak diatom populations
Standardized weekly means of cell densities over the period 1955–2003 show that the peak in Cyclotella abundance precedes the A. formosa peak (Fig. 1) and that population peaks of the former species are, on average, more protracted and irregular than the latter. The central tendency method produced timings that were slightly earlier than the day of maximum abundance method (A. formosa: difference between means = 3.9 days, t = 4.32, P < 0.001; Cyclotella: difference between means = 5.6 days, t = 3.46, P = 0.001). However, similar inter-annual variations in timing were found when comparing the central tendency method and the day of maximum abundance method: the two indices were well correlated for both A. formosa (Pearson's r = 0.92, P < 0.001) and Cyclotella spp. (Pearson's r = 0.81, P < 0.001). Given this similarity, only the central tendency data were used in subsequent analyses. During the study period the timing of A. formosa showed a progressive linear advancement (Fig. 2a) at a rate of approximately 0.3 days year−1 (Gaussian additive model, equivalent degrees of freedom, edf = 1.0, χ2 = 9.1, P < 0.01). The phenology of Cyclotella showed no systematic change prior to the mid 1980s. However, after this date, the timing of the peak population advanced rapidly (Fig. 2b) at a rate of approximately 1.3 days year−1 (edf = 2.4, χ2 = 11.8, P < 0.01). The nonlinear trend in Cyclotella phenology, estimated by the additive model, explained significantly more temporal variation than did a simple linear trend (F = 4.3, P < 0.05).
Based on the approach used to estimate the total diatom biovolume, A. formosa dominated the spring diatom community. In the first 6 months of the year this species made-up, on average, approximately 46% of the total diatom biovolume. On approximately 31% of sampling occasions, primarily during the total biovolume peak, this contribution was in excess of 80% of the total biovolume. Cyclotella spp., on average, constituted 20% of the total spring diatom biovolume. On approximately 14% of sampling occasions, principally prior to the A. formosa and total biovolume peaks, this taxon comprised over 80% of the total biovolume. As a result of the dominance by A. formosa, the inter-annual variations in the timing of the spring maxima of both A. formosa and total diatom biovolume were strongly correlated (Pearson's r = 0.85, P < 0.001).
potential drivers of diatom phenology
Over the study period a number of ecological changes occurred in the North Basin of Windermere. Gaussian additive models were used to explain each potential driving variable as a function of time statistically, in order to highlight significant temporal trends. Model smoothing parameters were estimated by Generalized Cross Validation. The winter concentration of SRP increased significantly (Gaussian additive model, edf = 4.0, χ2 = 179.8, P < 0.001) from the start of the period until the early 1990s, after which no further systematic change was apparent (Fig. 3a). In contrast, there was only a slight, non-significant, long-term decrease (edf = 1.3, χ2 = 1.7, P > 0.05) in winter silica concentrations (Fig. 3b). The magnitude of over-wintering populations of A. formosa generally increased with time (Fig. 3c, edf = 4.1, χ2 = 11.3, P < 0.05). Temporal changes in over-wintering populations of Cyclotella were rather more irregular (edf = 7.6, χ2 = 22.5, P < 0.05), with no clear overall trend (Fig. 3d). The mean biovolume of zooplankton prior to each of the diatom peaks generally increased until the late 1980s, declined through the 1990s, and subsequently increased until the end of the time series (Fig. 3e,f). These temporal trends were statistically significant (both variables, edf = 6.4–6.7, χ2 ≥ 44.0, P < 0.001).
Mean water temperature increased over time (edf = 1.9, χ2 = 13.5, P < 0.01), most rapidly after the late 1980s (Fig. 4a). The day of year on which stability thresholds were surpassed generally advanced at the same time as this increase in water temperature (Fig. 4b–e). This pattern was most obvious for the day on which the 20, 50 and 300 J m−2 stabilities were surpassed (all variables, edf = 1.7–3.0, χ2 ≥ 9.1, P ≤ 0.01), and was weaker for the 150 J m−2 threshold (edf = 2.0, χ2 = 3.8, P > 0.05). Prior to this period, there was some evidence that the 20–150 J m−2 thresholds were occurring progressively later in the year. There was evidence to suggest that warmer springs and earlier stratification followed milder winters. Inter-annual variations in water temperature were correlated with the NAO winter index (Pearson's correlation with Bartlett (1946) correction for autocorrelation in both series r = 0.61, P < 0.001), as were the times at which the 20 and 300 J m−2 stability thresholds were exceeded (r = −0.34, P < 0.05 and r = −0.43, P < 0.01). The times at which the 50 and 150 J m−2 thresholds were surpassed were only weakly associated with the winter NAO (r = −0.28, P < 0.10 and r = −0.25, P < 0.10). Daily mean minutes of bright sunshine decreased from the start of the study period until the late 1980s and then increased (Fig. 4f, edf = 3.1, χ2 = 9.7, P < 0.05).
For all but one of the variables with statistically significant patterns of temporal change, the nonlinear trend estimated by the additive model explained significantly more temporal variation than did a simple linear trend (all F ≥ 3.1, all P < 0.05). Only for mean water temperature was this difference less pronounced, with the additive model explaining marginally more temporal variation than a general linear model (F = 3.9, P = 0.06).
statistical modelling of long-term change
Prior to statistical modelling, data from 1 year (1974) were excluded from the A. formosa analysis due to an atypical bimodal seasonal pattern that did not represent well the general tendency for a unimodal pattern in the other years. Two years were omitted from the Cyclotella spp. analysis (1967 and 1990) since this taxon was not found in the samples from those years. A general linear model of the timing of the spring A. formosa peak, including all potential predictor variables, retained only mean winter SRP concentration and the mean over-winter cell density as statistically significant predictors. Plots of residual variation against all retained and omitted predictor variables revealed that systematic residual variation in phenology was still present with respect to the magnitude of the over-winter population. The model was therefore re-run as a Gaussian additive model (Table 1). The timing of the spring peak population advanced as a linear function of increasing winter SRP concentration and as a nonlinear function of the magnitude of the over-wintering population (Fig. 5a,b). When the effect of the over-wintering population was modelled as a nonlinear function, mean water temperature also had a significant effect on the timing of the spring peak population, with populations advancing as a linear function of increasing water temperature (Fig. 5c, Table 1). Models with 2, 3 and 4 degrees of freedom for the effect of the over-wintering population explained comparable amounts of variation in Asterionella phenology (45.9–48.0%), though the model with the degrees of freedom set at 2 produced the lowest AICc value and was therefore selected as the closest to optimal model. The nonlinear effect of this predictor explained significantly more variation in the timing of the peak population than a simple linear function (F = 11.27, P < 0.01). These three variables explained approximately 46% of the total deviance in the timing of the peak. The relative importance of these predictors was assessed by omitting each, in turn, from the final model and recording the increase in residual deviance. The increase in residual deviance was greatest when the over-wintering population term was omitted, and least when water temperature was omitted (Table 1), indicating that the latter had the weakest effect on the timing of the spring population. The timing of the spring peak also coincided with the time at which the concentration of silica fell below 0.5 mg L−1 (Fig. 6). Inter-annual variations in these events were well correlated (Pearson's correlation with Bartlett (1946) correction for autocorrelation in the silica series, r = 0.47, P = 0.001). In the absence of the 3 years for which peak population densities clearly preceded silica depletion this correlation was further strengthened (Pearson's correlation, r = 0.91, P < 0.001), as were the relationships between the timing of the spring peak, mean winter SRP concentration, water temperature and mean over-winter cell density (Table 1). These years had the lowest population densities of Asterionella in the data set (mean cell densities from January to June < 90 cells mL−1, cf. > 200 cells mL−1 for other years).
The timing of the spring peak in Cyclotella abundance was significantly related to the timing of thermal stratification (Table 1). The time at which Schmidt stability exceeded 20 J m−2 was a statistically significant predictor of timing. The magnitude of the over-winter population was also significantly related to the timing of the peak in Cyclotella spp. abundance (Table 1). Peaks in the abundance of Cyclotella spp. occurred earlier in years with earlier stratification and higher over-wintering populations, though a relatively low proportion of the total variation in timing could be explained by these factors (30.8%). Residuals from the general linear model showed no systematic variation with the retained predictor variables or with omitted variables, so additive models were not necessary. Comparison of the timing of peak Cyclotella abundance with the timing of silica depletion revealed that the peak in abundance occurred before silica concentrations fell below 0.5 mg L−1 (Fig. 6).
comparison of years with early and late peaks in asterionella abundance
In Fig. 7 we compare the spring peaks in A. formosa abundance in 1955 and 2003, along with seasonal changes in growth limiting factors. The years were chosen to represent the temporal extremes of the data set and to illustrate the seasonal events that occur in years with early and late population peaks. In both years, the collapse of the A. formosa population coincides with the depletion of silica to growth limiting concentrations (Fig. 7a,b). The net rate of increase is slower in 1955, after a long period of phosphorus depletion, than in 2003.
There is no doubt that there has been a general trend of advancement for spring biological events in recent decades (Visser & Both 2005; Menzel et al. 2006). An expanding body of literature is now highlighting the presence of such patterns in aquatic ecosystems, and particularly in plankton communities (Gerten & Adrian 2002; Edwards & Richardson 2004; Winder & Schindler 2004a). Though detailed community data are not always available, where they are it appears that the extent of the phenological shift varies between species (Gerten & Adrian 2002; Edwards & Richardson 2004). The present study confirms this and, moreover, that the pattern of inter-annual phenological change also varies between species. By considering two numerically abundant and morphologically distinct diatoms from the same lake ecosystem we have shown that a combination of shared and differing mechanisms could account for species-specific differences in these patterns of change.
The colonial diatom A. formosa is a dominant member of the spring phytoplankton community in Windermere (Reynolds & Irish 2000) and it is clear that there has been an advancement in the timing of the peak population of this species over the last 50 years. This phenological shift can be partially explained as a response to the warming of the lake over this period, but it is also apparent that the spring population peak occurs earlier in years that begin with a higher winter SRP concentration. Though the former result supports findings in the existing literature on the direct effects of climate change on lake plankton communities (Winder & Schindler 2004a,b), the latter supports the hypothesis proposed by Reynolds (1990, 1997a). This states that the spring peak population is advanced when higher initial SRP concentrations permit a longer period of rapid light-limited growth, thus permitting an earlier attainment of the silica determined maximum population size (Fig. 8a). That the central tendency of the A. formosa population almost always coincides with the time at which silica concentrations fall below the value 0.5 mg L−1 necessary to permit further cell divisions (Lund 1950a,b; Reynolds 2006) supports this interpretation. This suggests that the phenology of this species is a response not only to climatic change, but also to the local process of nutrient enrichment. Both of these ecological pressures have affected the replication rate relative to the loss processes acting upon the population. This would suggest co-limitation of replication rates, and ultimately of seasonal timing, by temperature and nutrient availability (Rhee & Gotham 1981). Since A. formosa dominate the spring diatom community, aggregate diatom biovolume shows a very similar shift in phenology to this species alone. As a result, the responses of this species to eutrophication and warming appear to be felt at higher levels of ecological organization.
The spring sub-dominant Cyclotella spp. reach their peak earlier in the year than A. formosa, but have also shown a phenological advancement over the last 50 years. The pattern of inter-annual change in phenology differs markedly from that for A. formosa, indicating that different mechanisms are responsible for this change. The absence of an effect of SRP availability on Cyclotella spp. phenology suggests that nutrient concentrations are not responsible for the observed changes in seasonal timing. Furthermore, the peak in Cyclotella abundance is typically before silica concentrations fall below the 0.5 mg L−1 limit that adversely affects A. formosa. Experiments have shown that the growth rate of Cyclotella meneghiniana Kützing can be maximized at lower silica concentrations than A. formosa (Tilman & Kilham 1976). Though C. meneghiniana is not the dominant species of this genus in Windermere, based on a consideration of nutrients alone we might hypothesize that Cyclotella spp. should be able to persist after the A. formosa population has declined, when silica concentrations have been depleted relative to the concentration of phosphorus. The data do not support this assertion, suggesting that the peak abundance of Cyclotella spp. occurs before nutrient limitation can become the dominant factor affecting phenology.
Although less of the variability in phenology could be explained for Cyclotella spp. than for A. formosa, statistically significant effects of the timing of stratification were found. This would lend support to the hypothesis that the phenology of this species is affected by physical processes. The peak abundance of Cyclotella spp. advanced with advancing thermal stratification. However, the threshold stability values that affected Cyclotella spp. were indicative of weak thermal structure. The timing of the peak in Cyclotella spp. abundance was related to the time at which Schmidt stability exceeded 20 J m−2, approximately equivalent to a temperature difference of only 1 °C throughout the whole water column. This is consistent with previous studies that have shown that species of this genus can grow during periods of mixing, in the absence of a stable thermal structure (Viner & Kemp 1983; Lindenschmidt & Chorus 1998). The development of thermal structure, and associated reduction in turbulent mixing, will result in increased sinking losses from the Cyclotella population (Huisman & Sommeijer 2002), thus increasing rates of loss relative to the replication rate and causing a decline in the abundance of this genus. Earlier increases in sinking losses could bring about an earlier peak in Cyclotella abundance with advancing thermal stratification (Fig. 8b). The advancement in thermal stratification paralleled the warming of Windermere after the late 1980s: a trend that has been observed in other European lakes and that has been linked to variations in the North Atlantic Oscillation (George et al. 2000; Gerten & Adrian 2000). There was some evidence to suggest that higher spring water temperatures and earlier stratification in the North Basin of Windermere followed milder winters, though the weak correlations between some stability thresholds and the winter NAO indicated that spring weather conditions, as well as winter conditions, likely had an important influence on these physical properties. Irrespective of the nature of the mechanistic link between climate and lake physical structure, it appears that there is a link between climatic change and the phenology of Cyclotella, via effects on lake physical processes.
This raises the question of why the A. formosa population does not decline as a result of the same processes that cause the loss of Cyclotella spp. during this early stage of stratification. The answer may lie in the morphological differences between these taxa. A. formosa has a higher surface area to volume ratio than the C. comensis or C. pseudostelligera that dominate this genus in Windermere (Reynolds 1997b; Morabito et al. 2004; Reynolds 2006). This suggests that A. formosa has a higher form resistance to sinking and would be lost less quickly from a mixed water column. Reynolds (1984) showed that a morphologically similar Cyclotella species sank more rapidly from a mixed experimental column than A. formosa. We might expect, based upon this, that Cyclotella spp. grow to their peak abundances early in the year as a result of their dependence on intense mixing conditions. Further, in the early stages of stratification we might expect that Cyclotella spp. would suffer higher sinking losses than A. formosa, and thus decline in abundance when A. formosa can still maintain a positive net rate of growth. Since the timing of the spring population maximum can be defined as the point at which rates of loss exceed replication rates, we can expect that this functional difference between the two taxa allows A. formosa to reduce its sinking losses more effectively than Cyclotella during the earlier stages of stratification. Lindenschmidt & Chorus (1998) observed the same difference in the relative timing of these genera, with respect to the seasonal mixing pattern. The result of this functional differentiation is that Cyclotella spp. are more sensitive to inter-annual variations in the timing of thermal stratification than A. formosa.
It must be acknowledged, however, that the mechanism driving the timing of Cyclotella populations may be rather more indirect. A. formosa is the dominant vernal phytoplankton species and as such changes to its timing, both in terms of growth initiation and maximum abundance, may well have consequences to its competitors. This idea has not been tested in this paper but will be a priority for further work. The use of a ‘cardinal characteristics’ approach (Maberly et al. 1994) to define and compare different phases of the population growth of coexisting species would be valuable in attempting to validate this hypothesis.
Despite the apparent differences in the drivers of phenology for these co-occurring species, there also appears to be a common influence on their dynamics: the magnitude of the over-wintering population. When population densities are higher at the end of the previous year, the subsequent spring peak tends to occur earlier. Assuming constant rates of increase, we would expect that a certain population size, or indeed carrying capacity, would be attained earlier if the starting population size is larger (Fig. 8c). Previous studies have suggested that the magnitude of over-wintering phytoplankton populations can affect spring population dynamics and phenology (Maberly et al. 1994; Gerten & Adrian 2000; Reynolds & Irish 2000) and that the magnitude of the over-winter population might be linked to climate (Gerten & Adrian 2000; Edwards et al. 2006). Further work is needed in order to establish whether climate is indeed behind these variations, so that the effect of over-wintering abundance on spring phenology actually represents a lagged effect of climate, or whether these variations are in fact a result of changing lake trophic status. This will also form the focus of a later study.
Despite the fact that both of these diatom taxa can be grazed by crustacean zooplankton (Lehman & Sandgren 1985; Vanni & Temte 1990), there was no evidence for an effect of variations in grazing pressure on the timing of their population maxima. Dependent on the taxon being considered, this would suggest that grazing losses are a small component of the overall loss rates suffered by the diatoms during the spring period, or that the variations in grazing losses are minimal compared to inter-annual variations in replication rate. For Cyclotella, it would seem that sedimentation is a more important component of the overall loss rate whilst, for A. formosa, variations in phenology are driven more by factors that affect the rate of replication than by those that affect the rate of loss.
Despite the presence of statistically significant effects of resource supply, water temperature, patterns of stratification and over-wintering populations on phenology, it is clear that there is still considerable unexplained inter-annual variation in seasonal timing. The 3 years in which the A. formosa population declines before silica is depleted demonstrate this. The years in question (1961, 1962 and 1970) showed no evidence of being exceptional with respect to any of the explanatory variables considered here but, along with 1966, had far lower population densities of A. formosa than the other years in the data set. In the absence of these years, the effects of winter SRP concentration, water temperature and over-wintering population size on phenology were considerably strengthened. Such unexplained variation in phenology is likely to originate from a variety of sources. First of all, there are likely to be as yet unidentified influences on phenology and, moreover, interactions between different driving variables. Though interactions between drivers can be included in statistical models, the additional explanatory power and complexity come with a risk of both over-parameterization and reduced tractability. There is the possibility that the rate of spring increase is affected by competitive interactions between the target species and other coexisting species for light as a result of reduced transparency under conditions of high phytoplankton biomass. Furthermore, interactions with parasites may also affect spring dynamics (Canter 1979). The influence of short-term weather variations on phenology must also be acknowledged as a source of unexplained variation within the context of this analysis (George & Hewitt 2006).
The results of this study indicated that nonlinear trends provided a significantly better description of long-term variations in the phenology of Cyclotella and of patterns of change for many of the potential driving variables considered. Furthermore, the relationship between over wintering population size and Asterionella timing was nonlinear. However, analyses of changing biological seasonality typically use methods that assume linear relationships. Given the nonlinear nature of both patterns of change and the relationships that might drive these patterns (Stenseth & Mysterud 2002) it is clear that the restriction of linearity must be relaxed in our analyses of biological seasonality.
It is recognized that long-term changes in plankton community composition and abundance are the result of both changes in climate mediated physical forcing and nutrient enrichment (George et al. 1990; Anneville et al. 2004; Anneville et al. 2005). However, most studies of phenology have emphasized the role of climate variation. The widespread nature of the spring phenological response does suggest that this is the product of a large-scale phenomenon such as climate change which, for lake phytoplankton communities, might be most apparent via effects on temperature and mixing patterns. However, it is also possible that local phenomena can produce similar phenological shifts. The present study shows that shifting phenology may also be a response to variations in both resource supply and climate, thus confirming the hypothesis proposed by Reynolds (1990, 1997a). The relative importance of large-scale and more localized phenomena can vary among species that share the same habitat. Therefore, in addition to earlier observations that the extent of the phenological shift is species-specific, it appears that the mechanism driving this change is also species-specific. The form of the temporal trend in phenology (in this case linear vs. nonlinear), along with consideration of the multiple drivers that affect these patterns, could give clues to underlying mechanisms and provide the insight needed to tackle the problem of attribution.
We are grateful to all staff from the Freshwater Biological Association and Centre for Ecology and Hydrology for collecting and processing the samples collected during the long-term monitoring programme. This work was funded by the Freshwater Biological Association and by the European Union Sixth Framework Programme integrated project Euro-limpacs (GOCE-CT-2003-505540). We thank Professor C. S. Reynolds for his comments on the manuscript.