The abiotic conditions that shape phytoplankton changes in
Lake Markermeer: An analysis of changes over 27 years in
the phytoplankton of a shallow turbid lake.
Donker, Bart
Department of Freshwater and Marine Ecology (FAME), Institute for Biodiversity and Ecosystem Dynamics (IBED), University of Amsterdam, Amsterdam (UVA), The Netherlands.
Abstract
Lake Markermeer in the Netherlands is a large-sized shallow freshwater delta lake, where year-long wind-is continuously inducing resuspension of fine marine sediment particles from the lake bed. Over the past years it has been reported that nutrient loads in the lake have decreased significantly. As this has been suspected to cause a shift in phytoplankton taxa, the aim of this study was to determine what underlying abiotic condition cause an increase or decrease in abundance of certain different phytoplankton taxonomic classes.For this purpose a Generalized additive model (GAM) model was made, using long-term data from between 1992 to 2016, where to influence of inorganic nitrogen, phosphate, silica and suspended matter (turbidity, amount of resuspension) concentrations and residence time of Lake Markermeer were tested for six different
phytoplankton classes.
According to this model, suspended matter seemed to highly influence Bacillariophyceae biovolumes in the lake. Suspended matter concentrations between 50 and 200 mg/L (suspended matter ranges between 5.5 and 283 mg/L), more than doubled Bacillariophyceae predicted biovolumes, compared to concentrations at 10 mg/L or higher than 250 mg/L.
The model also suggests suspended matter biovolumes to influence Chlorophyceae biovolumes, showing a similar pattern to Bacillariophyceae, where the highest biovolumes were predicted at this mid-range of 50 mg/L to 200 mg/L suspended matter concentrations. For the Cyanophyceae residence time (residence time ranges between (363 days and 580 days) was found to influence the predicted biovolumes significantly, with an increase in residence time predicting an increase in predicted Cyanophyceae biovolumes, where predicted biovolumes more than tripled when comparing the biovolumes of a residence time of 400 days, compared to those at 580 days.
Overall suspended matter seemed to be the most influential abiotic condition in Lake Markermeer to influence phytoplankton biovolumes, and no significant effect of nutrient levels on phytoplankton biovolumes has been found. This suggests that the changes in nutrient conditions of the water column between 1992 and 2016 have not significantly impacted the phytoplankton biovolumes in Lake Markermeer.
Correspondence: trab12d@live.nl (author) h.g.vandergeest@uva.nl j.m.h.verspagen@uva.nl
Introduction
Understanding long-term ecological processes has been an important challenge in ecology, and this is getting increasingly important in the light of the current fast changing world and increasingly anthropogenic impacts (Cardinale et al., 2012). Phytoplankton communities are an important basis for many ecological processes occurring in the water. They are the base of the food web as the primary producers, providing oxygen to the water and food to the zooplankton, and are an important driver in nutrient cycling and energy flow (Boyce at al., 2010; Findlay et al., 2015). Phytoplankton can also be used as an indicator of the chemical conditions and trophic status of water bodies, as phytoplankton respond directly to changes in nutrient conditions or environmental factors through fluctuations in their relative biomasses and community structure. These quick responses are made possible by their short life cycle, allowing them to quickly and easily adapt in short periods of time to their environment (Reynolds, 2006). They can however also be harmful to the environment, as many toxic species are able to quickly bloom under the right circumstances and repel many species important to their environments (Hallegraeff et al., 2003). Because of these reasons phytoplankton communities are often used as an ecosystem state indicator (Barinova & Krupa, 2017; Drozdenko et al., 2017).
Lake Markermeer is a large-sized shallow freshwater delta lake in the Netherlands, which used to be connected to the North Sea, until the completion of the Afsluitdijk in 1932. Later on it was also disconnected form the river IJssel in 1975 when the Houtribdijk (a dyke) was completed, and over time lacked natural land-water transition zones, as the area around the lake consist currently for a large part of dykes, rocks and roads. The lake is currently mainly receiving water inflow from Lake IJsselmeer (predominantly Rhine River water) by sluices connecting the two lakes, rainfall, polder drainage (Bonte & Zwolsman, 2010) and the river Eem (Noodhuis, 2010). Water discharge usually goes either through sluices or letting the water flow from Lake Markermeer to Lake IJsselmeer, this is by far the most important route since 2004 when the capacity of the water flow between lake IJsselmeer and Lake Markermeer was increased, making this connection the most important inflow in summer times and most important outflow during winter (Noordhuis, 2010). Alternatively, water can be drained to the Noordzeekanaal channel. The retention time of the water in Lake Markermeer has slightly increased over the years, during the year 2000 it was estimated to be between 10 to 15 months (beheersverslag Rijkswaterstaat, 2000-2001), while it currently is reported to be 18 months (van Riel et al., 2019). Reason for this increase in water residence time is for a large part of the policies how the sluices built around Lake Markermeer are operated. For an overview of the area see figure 1.
Figure1. Map of the area around Lake Markermeer. The sample location (Markermeer midden) is indicated with a purple diamond.
In Lake Markermeer the following phytoplankton classes are usually commonly found: Bacillariophyceae, Chlorophyceae, Chrysophyceae, Cryptophyceae, Cyanophyceae, Dinophyceae, Euglenophyceae, Haptophyceae and Tribophyceae. Bacillariophyceae (diatoms) are a major group of algae, commonly found all over the world, they are mainly phototroph, with a few heterotrophic species and are usually one of the important primary producers, especially in rich waters. A unique feature of diatom anatomy is that they are surrounded by a cell wall made of silica. Chlorophyceae (one of the classes of green algae) are another large group and usually important freshwater algae, and responsible for a large part of the primary production. They are usually (dark) green colored, due to having many chlorophyll a and b pigments. Chrysophyceae (golden algae) are another large group of algae, also commonly found in freshwater, and are recognizable by their usually a yellowish color. They are usually flagellates, but also contain amoeba like species. They are often used in studies of food web dynamics in oligotrophic freshwater systems and assaying environmental damage, resulting from
eutrophication (Sandgren et al., 1995). Cryptophyceae are distinguishable by the presence of their
characteristic ejectosomes, a membrane-bound structure consisting of two spiral ribbon-like structures allowing them to quickly move in a zig-zag like fashion away from danger. They are photosynthetic, usually containing chlorophyll a and c, and come in many different colors.
Cyanophyceae (blue green-algae) are a photosynthetic class and can be found in almost every aquatic habitat. They are mostly known for their ability to form extensive blooms in both freshwater and marine environments, and often produce a range of toxins, making them usually harmful for many different animal species, and are therefore often combatted by many government agencies (Olding et al., 2000). Dinophyceae (a class of
dinoflagellates) are a class containing phototrophic, heterotrophic and mixotrophic species. Often Dynophyceae are motile cells are biflagellated unicells although some species also form into chains. One of their unique properties is their unique nucleus form (sometimes called a dinokaryon) (Fukuda & Suzaki, 2015),
chromosomes in this dinokaryon are permanently condensed throughout the whole cell cycle, being attached to the nuclear envelope, and show a cholesteric liquid crystal organization. Just like Cyanophyceae, they can form harmful algal-blooms, negatively affecting the health of many animal species, and therefor generally unwanted. Euglenophyceae are unicellular flagellates, most commonly found in freshwater. A majority of the Euglenophyceae species are mixotroph, but the class contains a few phototrophic and obligatory heterotrophic species. Euglenophyceae do not have a cell wall, but are instead covered by a pellicle, a structure composed of proteinaceous strips underneath the cell membrane, supported by microtubules. These groups of
phytoplankton are studied mainly for their ecological importance as indicators of water pollution due to their often mixotrophic nature and their ability to tolerate extreme amounts of nutrients (Lackey 1968; Arguelles et al., 2014). Haptophyceae can be recognized by two flagellate and haptonema, a pin-like microtubule-based structure used for feeding for among other things, feeding (Kawachi & Inouye, 1995), attachment and gliding on a substrate (Manton, 1967). They are phototrophic species, consisting for a large part of the coccolithophores, species covered in an exoskeleton of calcareous. The Haptophyceae class mostly contains marine species, with a few freshwater species known. Lastly Tribophyceae (yellow-green algae) are an important group of the
heterokont algae, consisting mostly of freshwater species. They are phototrophic and occur often under dystrophic or mesotrophic conditions, showing their highest diversity in acidic waters enriched with dissolved organic matter (Salmaso et al., 2009).
Because Lake Markermeer has been disconnected from the river IJssel with the construction of the Houtrib dike, the lake is characterised by low phosphorous concentrations (less than 0.5 μM phosphate available in the water column) (Noordhuis, 2010; Rijkswaterstaat Waterinfo , measurements between 2008 and 2017).
Phosphorous concentrations declined after the year 2004 and kept declining afterwards, although a cause for this has not been identified yet (Noordhuis, 2010; van Riel et al., 2019). Nitrogen concentrations also kept declining in the lake since 1970 till today (Noordhuis, 2010; Rijkswaterstaat Waterinfo, measurements between 1987 and 2017), an effect caused by stricter European and Dutch laws to restrict water pollution and newly implemented active de-eutrophication policies (Noordhuis, 2010), eventually resulting in a large decline of nutrient availability in the lake, compared to the its state in the 1990’s (Noordhuis 2010; van Riel et al., 2019). Sulphate concentrations are only being monitored from the year 2005 onwards, and usually fluctuate between
78 and 105 mg/L, no upwards or downwards trend was reported between the years 2005 and 2009 (Bonte, 2009), a trend that seemed to have continued after the year 2009 (Rijkswaterstaat Waterinfo), although the latter has not been statically tested. An increase in sulphate concentrations has previously been connected to the growth of Cyanophyceae species, especially during winter or summer months (Kumar & Oommen, 2011). The pH of the lake is slightly high and fluctuating stably between 8.3 and 8.5 in the winter, while fluctuating between 9.0 and 9.3 at summer (Noordhuis, 2010). This has a negative effect on nutrient availability in the lake, as many nutrients precipitate under high pH conditions, making them unavailable in the water column
(Noordhuis, 2010).
Due to year-long wind-induced resuspension of fine marine sediment particles from the lake bed, most of the time Lake Markermeer suffers from low light penetration levels, however most of the nutrients are made available by mineralization processes in the sediment at the lakebed (Brinkmann et al., 2019). Because of this neither the very top of the lake where light is available, or the benthic zone where dissolved nutrients are available seem to be provide the conditions necessary for algal growth. Recently it was suggested that nutrients can be made available in the eutrophic zone of the lake, in the form of aggregates between from dissolved nutrients and organic solutes driving upwards in the lake, called lake snow. Nutrients from these aggregates can be released by hydrologic enzymes form microbes (Brinkmann et al., 2019), bridging the gap between light and nutrient availability.
Currently it is unknown what the consequences of the fast changing conditions in nutrient availability at the lake are for the phytoplankton composition. Long term analysis (periods over 10 years) of changes in the phytoplankton compositions of shallow lakes were done in the Mediterranean region (Villena & Romo, 2003; Vadrucci et al., 2017), Germany (Köhler, 2000), Estonia (Nõges et al., 2010; Cremona et al., 2018), China (Deng et al 2014). However in the temperate oceanic climate no such analysis has been performed up until now. As both climate and human induced changes affect phytoplankton in different ways in different environments and in different lake types, the effects found in previous works might not be comparable to those relevant for Lake Markermeer (Jeppesen et al., 2005).
The aim of this project was to determine the effect of nutrient and abiotic conditions over time on the algal community of the shallow Lake Markermeer. To gain an understanding of these consequences, datasets of abiotic and nutrient conditions (from here this is only referred to as abiotic conditions) of Lake Markermeer will be extracted from the Waterinfo database, a publicly available dataset from Rijkswaterstaat containing a variety of measurements of the water conditions in the Netherlands, and compared to taxonomic algal data from Lake Markermeer between 1992 and 2015. From this, changes over time of phytoplankton class abundances and abiotic conditions will be analysed. Afterwards, a model will be made which can predict the phytoplankton abundance, grouped by taxonomic composition, based on the abiotic conditions of the lake. From this model, the major drivers behind the phytoplankton abundances can be identified, as well as the way in which they influence over single phytoplankton class.
The changes in the condition of the lake can cause various changes in the abundance of different phytoplankton groups.
Phosphorous and nitrogen concentrations have previously been positively correlated to an increase in
abundance of Cyanophyceae, Chlorophyceae and Euglenophyceae species at shallow lakes in China and Estonia. (Lv et al., 2009; Nõges et al., 2008), and therefore the same pattern is expected for Lake Markermeer. A decline in cyanobacteria species is expected for an increase in sulphate concentrations (Kumar & Oommen, 2011). The low iron concentrations are however expected to benefit cyanobacteria species, as it has previously been reported that lake communities shifts from a green algae community structure to a cyanobacteria one, when iron concentration ranges between 0.1 to 1.0 mg/L.(Morton & Lee, 1974; Pollingher et al., 1995; Xing & Liu, 2011), concentrations close to those of Lake Markermeer, where iron concentrations fluctuate between 1.0 mg/L and 2.0 mg/L most of the time (Waterinfo).
Lastly, the large water residence time of the lake is also thought to be especially advantageous to cyanobacteria species (Olding et al., 2000). This is because long water residence time is required to provide stable water column conditions necessary for Cyanophyceae growth, even in the presence of high nutrient levels. Suspended matter concentration is expected to be one of the most important variables in determining phytoplankton composition, and is expected to affect all phytoplankton classes, due to the hypothesis that a large part of nutrients in Lake Markermeer is available from this suspended matter (Brinkmann et al., 2019) and the influence of turbidity on light availability of the lake. No specific expectations were had for the influence of suspended matter on any of the phytoplankton groups, as the specific effects of suspended matter on the biovolumes of phytoplankton classes has not been researched yet for Lake Markermeer.
Overall though, it is expected that the abundance for every taxonomic group will be explained by multiple different abiotic conditions, and not for a large part by a single condition, as phytoplankton abundances are expected to be dependent on a combination of many different factors (Vrieze, 2012).
Methods
Phytoplankton and lake abiotic condition data collection and preparation
For the phytoplankton abundance data, the revised taxonomic overview of algal species of Lake Markermeer between 1992 and 2065 of the MWTL program (the standard monitoring program) of Rijkswaterstaat was used. This data contains samples taken from the middle of the lake (Markermeer midden; see figure 1) of which the phytoplankton was determined to either species or genus level. This dataset was later revised by Vonk et al. (2019) to harmonise the taxonomy and add bio volumes wherever they were missing. If no species data was available and only higher taxonomical data was available, biovolumes were estimated based on the most abundant appearances. From this data only the biovolumes of the taxonomic classes were extracted and the sum of the biovolumes for each class was averaged by month. If a certain class was not included in the dataset for that month, it was not included in the analysis for that month. Residence time data from the lake was obtained by van der Geest et al., (2018). All other data for the abiotic conditions of the lake used in this study, was obtained from Waterinfo, the database from Rijkswaterstaat containing all measurement data of water conditions in the Netherlands (https://waterinfo.rws.nl/). All abiotic data was averaged by month for later analysis as well. Outliers were defined as values that are at least 3 times the interquartile distance away from the 1.5 interquartile range point and removed from our data.
For an list of all abiotic conditions taken up for data analysis and phytoplankton classes observed in Lake Markermeer according to the dataset of Vonk et al., see Table 1.
Table 1. List of phytoplankton classes and abiotic conditions used for data analysis. Phytoplankton classes Abiotic variables
Bacillariophyceae Phosphorus concentration Chlorophyceae Phosphate concentration Chrysophyceae Total nitrogen concentration Cryptophyceae
Inorganic nitrogen concentration Cyanophyceae Iron concentration Dinophyceae Sulphate concentration Euglenophyceae
Suspended matter concentration
Haptophyceae Surface water temperature Tribophyceae Surface water pH
Other/ undefined classes Water residence time
Data exploration and removal of abiotic conditions not to be taken up in the model
Phytoplankton groups observed in less than 50 months over the 27 year period were deemed too small of a sample size and removed. All abiotic condition for which data was only available for 2005 and onwards, were removed as well, as to few data points remained to do accurate analysis with.
Before further data exploration, I looked at which factors change over time and could therefore have an actual meaningful influence on the phytoplankton community. Any of the variables that showed no development over time, was excluded for further analysis. Finally, a Spearman correlation test was performed between all abiotic variables to gain a general understanding how these variables were correlated with each other. Although collinearity is of non-concern using a Generalized additive model (GAM) model, instead, concurvity effects can make the model unreliable, total phosphorus and total nitrogen concentrations would be removed for further data analysis, to primitively prevent concurvity from occurring. Concurvity can be seen as a generalization of collinearity and occurs when a smooth term in a model could be approximated by one or more of the other smooth terms in the model (Wood, 2019).
Making the Generalized additive model (GAM) model
For our model we want to find out which abiotic conditions influence the abundance of certain phytoplankton groups, and how these abiotic conditions influence the abundances. Because of the seasonal dynamics in phytoplankton abundances and the fact that the samples are taken over a long period of time (27 years) an additional time factor has to be accounted for in addition to our abiotic predictors in the model. Meaning our model has the following form:
Phytoplankton class abundance =
function (Abiotic conditions, a selection of the continuous factors from the list of abiotic conditions) +
function (Time, consisting of a categorical variable month and a continuous variable year after first sample).
A GAM was made for every taxonomic class, to model the effects of the abiotic conditions on the abundance of each class respectively. A GAM model was selected, for its ability to correct for categorical variables, in this case seasonality/ month effects, and its flexibility to model non-linear relationships. To select which abiotic
conditions would be used to model and predict class abundances the most accurately, a stepwise backwards elimination selection method, based on Aikaike Information Criterion (AIC) score was performed in R using the package “gam” (not to be confused with the package used to make the GAM models) by Trevor Hastie (last updated July 3, 2019). This delta in AIC scores was preferred over the BIC scores (Bayesian information
criterion), as they estimate the prediction power of the model, without punishing less parsimonious models as hard as the BIC score would do. From this the abiotic variables that result in the lowest AIC score, thus best fitting model, were selected to be used as predictor variables in the GAM model. Because of this selection method, it is possible that some predictors get included in the model, even though they do not have a
significant effect, according to the Wald test statistic. When a predictor is still selected, but has a p-value >0.05 it can mean that the factor still has a predictive value, but there is also a change for it to be noise. Whether the
predictor is noise or has actually a slight effect on the outcome of the model must be manually interpreted. The categorical variable “month” which is included to account for seasonal effects, and continuous variable “year after first sample”, were forced to be included in the model. These time factors correct for any other effect besides the selected predictors, so only the influence of the abiotic conditions are examined. The GAM model was made using the R package “mgcv” by Simon Wood (last updated November 9, 2019). For each GAM model, the Generalized cross-validation (GCV) smoothing method was used, as these resulted in the lowest AIC scores. After making the model, the concurvity of the data used in the model is checked. Because the GAM model can self-correct concurvity effects, only concurvity relationships of 0.8 or higher are excluded from the models (Wood, 2007). When these concurvity effects occur, multiple variations of the model are made, a model version for all predictors included originally, and version without the year correction factor, as these concurvity effects occur due to this year factor including the changes in abiotic conditions over time. This means however that the alternative model does not contain the correction all other models do. For this study this alternative model was preferred however, because it is believed to be a better indicator for which abiotic conditions can potentially shape the phytoplankton class biovolumes, even though the exact biovolumes of these classes may be predicted incorrectly (Wood, 2019).
From these models, the most influential variables, the specific pattern in which they influence predicted biovolumes and adjusted R2 scores, a measure for the proportion the model is able to explain biovolumes over random change, have been assessed.
Results
Removal of variables used as candidates for the GAM model
The classes Chrysophyceae, Dinophyceae and Euglenophyceae were not further researched, as they were observed in less than 30 month over our 27 year period and therefor deemed too small of a sample size. Data for iron and sulphate concentration were removed from our dataset as potential predictors, as they were only available for 2005 and onwards, meaning we had to few data points to do accurate analysis with (Appendix 1). pH and temperature of the surface water were also removed as a potential predictor in the model, as the pH and temperature did not seem to have shown development over the years (Appendix 1).
According to the Spearman correlation test, all abiotic variables show collinearity. For this reason I removed total nitrogen concentration and total phosphorus concentration from our list of candidate abiotic conditions, as these factors were (unsurprisingly) collinearly related with inorganic nitrogen concentration and total phosphorus concentrations (Appendix 3) and are less ecologically relevant as the inorganic nitrogen and phosphate are the most readily available forms for uptake by phytoplankton. After the removal of these abiotic variables, the following candidate predictors were left to make the model with: Phosphate concentration, inorganic nitrogen concentration, sulphate concentration, suspended matter concentration and water residence time.
Changes in abiotic conditions of Lake Markermeer over time
From 1992 untill 2008, the inorganic nitrogen concentration declined, but after 2008 the inorganic nitrogen concentration increased slightly. The phosphate concentration increased from 1992 to 1998, and declined from 1998 to 2004, after which it stabilised at low concentrations until 2016. Silica concentrations have been slightly increasing over the years, although they showed a slight decrease between the end of 2001 till the end of 2005,
after which it started to continue slightly increasing again. Both suspended matter concentrations, as well as residence time show continuous fluctuations, where residence time also showed a clear increase over time, whereas suspended matter is highly variable but remains relatively constant over time(figure 2). The abiotic data not considered as potential predictors can be seen in Appendix1.
Figure 2. Changes in abiotic conditions in Lake Markermeer from 1992 to 2016. Monthly averaged values of (a) Inorganic nitrogen concentration, (b) Phosphate concentration, (c) Silica concentration, (d) Suspended matter concentration and (e) Residence time if the water. The blue line represents the the locally estimated scatterplot smoothing (loess), with the
Changes in phytoplankton class abundances over time
From 1992 until 2002, Bacillariophyceae biovolumes slightly increased, but after 2002 they slightly decreased until 2008, after which it stabilized until 2016. Chlorophyceae biovolumes increased from 1992 until 2002, and declined from 2002 until 2016. Cryptophyceae biovolumes increased from 1992 to 1998, and declined from 1998 to 2004, after which it very slightly increased until 2016. Cyanophyceae show a slight decline between 1992 and 1998, after which it stayed stable until 2003, from which it started to slightly increase until 2012, where Cyanophyceae biovolumes stabilised again. Haptophyceae biovolumes are slightly increasing from 1993, until 2016. Lastly, Tribophyceae are slightly increasing between 1995 and 2015, but stabilised in the period between 1999 and 2005.
Figure 3. Changes in phytoplankton biovolumes per class in Lake Markermeer from 1992 to 2016. Monthly averaged values of (a) Bacillariophyceae, (b) Chlorophyceae, (c) Cryptophyceae, (d) Cyanophyceae, (e) Haptophyceae and (f) Tribophyceae. The blue line represents the the locally estimated scatterplot smoothing (loess), with the 95% confidence
interval in grey, calculated using the function loess of the base package “stats” in R. Y-axis scales differ per phytoplankton group.
Phytoplankton seasonality
A strong seasonal pattern is visible for all taxonomic classes (figure 4). For all phytoplankton classes, except for the Cyanophyceae and Chlorophyceae, the fourth month April has the highest median biovolume values. For Cyanophyceae the highest value is reached in August, while for Chlorophyceae this was in June.
All classes, except for Cyanophyceae and Chlorophyceae, show a strong decline in biovolumes during the colder months, fall and winter, with a strong increase in biovolumes in spring. The Cyanophyceae show a decline between February and May, followed by a strong increase come June. Chlorophyceae show relative stability over the year, with a slight decrease in biovolumes during the summer months, July, August and September.
Figure 4. Boxplots of seasonal patterns for different phytoplankton classes in Lake Markermeer. Data for (a) Bacillariophyceae , (b) Chlorophyceae, (c) Cryptophyceae, (d) Cyanophyceae, (e) Haptophyceae and (f) Tribophyceae.
On the x-axis are the months represent by number (1=January, 12=December). Y-axis scales differ per phytoplankton group. Seasonality patterns are based on analysis from 1992 to 2016.
The adjusted R2 scores from our models reveal that between 17.9% and 45.1%, or 36.9% if we exclude the taxonomic classes that show concurvity effects, of the abundance of phytoplankton classes can be explained using these GAM models (Appendix 2). The models for Cyanophyceae and Chlorophyceae show the best explanatory power, by having a significant predictor and the highest R2 scores, 0.369 and 0.268 respectively. For almost every taxonomic class, suspended matter was included in the most powerful model, only for Cryptophyceae and Haptophyceae, suspended matter was not included. For Bacillariophyceae, suspended matter seems the best and only explanatory predictor (p<0.05) from our list of predictors.
For Cryptophyceae, no explanatory predictors were selected in the backwards elimination selection, and only the general year predictor variable was included in the GAM model.
For the Tribophyceae class, although silica and suspended matter were selected as predictors for the model with the lowest AIC, no variables with significant explanatory power have been found.
For the Haptophyceae class the same thing occurred, although the variable silica was selected to be taken up in the model, it had no significance in the model.
For both the Cyanophyceae and Chlorophyceae class, concurvity effects were shown for residence time, meaning that the reliability of the model is uncertain, as an alternative model had to be made excluding the variable for year, and thus not correcting for changes in abiotic conditions over the years, which have not been taken up in my dataset. For Cyanophyceae residence time is the only significant predictor (p=0.001) when all four factors (thus including the general year variable) of the original model were considered. Residence time stays a significant predictor (p<0.001) as well when the year factor gets removed from the model, thus
suggesting that residence time indeed is a good explanatory variable for Cyanophyceae biovolumes, and not an artefact due to concurvity effects.
For Chlorophyceae both silica (p<0.05) and suspended matter (p<0.001) were significant predictors for the original model (including the general year variable). However in the alternative model silica no longer retained its significance (p>0.05), indicating it might possibly have been significant in the original model due to
concurvity effects. Suspended matter retained its significance in the alternative model, thus suggesting that suspended matter is a reliable predictor for Chlorophyceae biovolumes, whereas the reliability of silica as predictor remains uncertain.
GAM model output
Suspended matter concentrations significantly impacted the abundance of Bacillariophyceae. Looking at the output of the model a pattern is visible where Bacillariophyceae abundance is highest between suspended matter concentrations between 50 and 200 mg/L. At very low concentrations (<50 mg/L) or very high concentrations (>220 mg/L) Bacillariophyceae abundance decreases drastically from around 2 mm3/L to biovolumes lower than 1 mm3/L.
For the alternative model for the Chlorophyceae class, a similar pattern is visible. Although inorganic nitrogen, silica and residence time are included in the model with the lowest AIC score, it is only suspended matter concentration that significantly impacted the abundance of Chlorophyceae. Similar to the Bacillariophyceae class, at low or very high amounts of suspended matter, low abundances (are predicted according to the model, where high abundance is expected between the 50 and 200 mg/L suspended matter concentrations (The outcome of the original version of the model with the year factor included is available in Appendix 4). For the Cryptophyceae class no significant parameters for the model were found.
The alternative model for the Cyanophyceae showed a significant importance for residence time as predictor for abundance. The abundance seems to be predicted to increase with an increasing residence time overall
according to the model.
For both the Haptophyceae and Tribophyceae the only significant predictor was the year correction factor, indicating that there are abiotic factors that changed over the year shaping the abundance of these two classes, but they were not one of the five candidate predictors.
a
Figure 5. Plots showing the combined effect of the linear and nonparametric contributions for each selected abiotic factor to the phytoplankton class abundance in Lake Markermeer according to a GAM model. (a) Bacillariophyceae model, (b) Haptophyceae model, (c) alternative Chlorophyceae model (year variable excluded due to concurvity), (d) Cryptophyceae model, (e) alternative Cyanophyceae model (year variable excluded due to concurvity) and (f) Tribophyceae
model. The blue line represents the predicted abundance value whereas the grey area is the 95% confidence interval (only shown for abundances above 0 mm3/L). Abiotic factors that significantly impacted the abundance are indicated by a *.
Discussion
This study assesses the effects of changing abiotic conditions on the phytoplankton community in Lake Markermeer. The results suggest that phytoplankton of different taxonomic classes respond to the changing abiotic and nutrient conditions in Lake Markermeer.Two abiotic conditions with a significant effect on some of the phytoplankton classes have been found.
Residence time had a major influence on Cyanophyceae abundance, where longer residence times result in a higher biovolumes of Cyanophyceae species. Suspended matter had a major influence on Bacillariophyceae and Chlorophyceae biovolumes, whereas for both species suspended matter concentrations between 50 mg/L and 200 mg/L seemed the most advantages for high biovolumes of these species.
Residence time significantly impacted the Cyanophyceae biovolumes for Lake Markermeer.
In agreement with the hypothesis that an increase in residence time would benefit the Cyanophyceae
abundance, indeed a trend was visible where for higher residence times more Cyanophyceae were predicted by the model. Longer residence times are thought to be beneficial for Cyanophyceae dominance, because long water residence time is required to provide stable water column conditions necessary for Cyanophyceae growth, even in the presence of high nutrient levels (Olding et al., 2000). ). One hypothesis why Cyanophyceae need higher residence time, and thus a more stable water column is their inability to adapt and reproduce in rapidly changing environments (Paerl 1988; Olding et al., 2000).
A similar relationship between an increase in Cyanophyceae abundance and in increase in residence time is also shown in lakes where residence times are way lower than over a year, as is the case in Lake Markermeer. So is in the lakes in the Norfolk Broads national park in the UK a similar pattern shown on a way lower time scale, where shallow water bodies with residence times of less than 11 days showed significantly lower Cyanophyceae cell densities than at water bodies higher where the residence times exceeded 25 days (Moss & Balls, 1989). One important difference between the lakes in the Norfolk Broads and Lake Markermeer is the strong year-long wind induced resuspension of Lake Markermeer, potentially explaining the difference in time scale of this effect.
Suspended matter concentration is the most important abiotic condition to drive the phytoplankton abundances by far, it was selected to be taken up in the best fitting model, for four out of the six taxonomic groups (Bacillariophyceae, Chlorophyceae, Cyanophyceae and Tribophyceae), where it was a significant predictor for two taxonomic classes: Bacillariophyceae and Chlorophyceae. In both of these cases, it could be concluded that for Lake Markermeer, very low concentrations of <50 mg/L or very high concentrations of >200 mg/L were disadvantage for the abundance of Bacillariophyceae and Chlorophyceae. This result was in
agreement with the hypothesis that suspended matter concentrations are an important driver in the
abundance of the phytoplankton groups in Lake Markermeer. This is a possible conformation that indeed lake snow is highly important in nutrient supply for the photosynthetic algae community (Brinkmann et al., 2019),
providing aggregates of dissolved nutrients from the lakebed in the eutrophic zone of the lake, allowing microalgal growth and providing organic compounds to the rest of the phytoplankton community in the lake. This theory of lake snow is especially well supported, because the two important classes for primary producers Bacillariophyceae and Chlorophyceae showed the most convincing dependence on suspended matter
conditions.
Another possible reason for suspended matter concentration to be an important driving factor can be the filtering of the light. Light becomes less intense and different wavelengths get filtered stronger or lower in the lake based on the amount and nature of the particles in the water (Stomp et al., 2007). For the Chlorophyceae and green Cyanophyceae the filtering of this light can be an important explanation of potential dominance of these species in the case of a high concentration of suspended matter consisting of dissolved organic matter and particulate organic matter. This is because these high concentrations can result in filtering of light in such a way that mostly red light is available on the bottom of the lake, meaning that species, such as Chlorophyceae and green Cyanophyceae, who can strongly absorb the available red light, can become the dominant species in the lake, as was shown before in Lake Groote Moost, another shallow peat lake in the Netherlands (Stomp et al., 2007). As high amounts of suspended matter also indicate a large effect of mixing in the water column, an alternative explanation of increasing Bacillariophyceae biovolumes under higher suspended matter amounts can be the especially well adjustments of Bacillariophyceae to strong mixing and lower light irradiance in the lake, allowing them to grow faster than other species around them (Flöder et al. 2002). Under very high suspended matter concentrations (200 mg/L), all phytoplankton showed a decline in biomass, indicating that these high amounts of suspended matter take up to much light for it to be beneficial for the primary production at the lake.
Although Silica concentrations were not taken up in any model where it had a significantly effect to explain any of the phytoplankton class biovolumes, it still was often selected to be taken up in the model, meaning it is potentially possible to deduce that although it does not play a significant role in determining the shape of the abundance in one of the taxonomic groups tested in Lake Markermeer, silica plays a role in facilitating slight increases or decreases in phytoplankton biovolumes.
This might potentially be an indicator of a slightly limiting or inhabiting effect of silica concentrations shortages or surpluses, but there is also a change for this to be noise. It is difficult to make any strong statements
regarding the impact of these silica concentrations, as adding silica to the model improves its predictive capabilities, but it has not significantly changed any of the patterns of the shown prediction of any of the models. I argue here, that silica is not a nutrient which determines phytoplankton abundances in Lake Markermeer, as it has not been taken up in the model for Bacillariophyceae, whereas an increase in silica concentrations previously have shown a growth in Bacillariophyceae dominance, regardless of season. This increase is most likely due to a high inherent growth rate under non silica-limiting conditions (Egge et al., 1992). The dissolved inorganic nitrogen concentrations have only been taken up in a model once, namely the
alternative model for Chlorophyceae. This might mean the same as it did for silica, that inorganic nitrogen has a slightly limiting effect at relatively low or high concentrations, or a stimulating effect at mid-range
concentrations for Lake Markermeer of around 0.4 or 1.2 mg/L. However, as it has been only taken up in any of the models once (none of which significant), it can almost certainly be deduced that inorganic nitrogen has had no effect on the abundance of any of the phytoplankton groups in the lake.
Lastly, phosphate concentrations were never selected to be taken up in the model for any of the six
phytoplankton classes. This result that the decrease in phosphate concentrations is in sharp contrast to earlier findings, where a decrease in total phosphorus showed a significant increase in Dinophyceae (Jeppesen et al., 2010) in a study researching re-oligotrophication in 14 lakes in Denmark (Jeppesen et al., 2010), as well as an decrease in non- heterocystous Cyanophyceae, Cryptophyceae and Chlorophyceae, a change in species composition often related to oligotrophication (Reynolds, 1984; Reynolds, 1997). This same pattern has also been shown in more oligotrophication studies (Dokulil & Padisák, 1994; Cronberg, 1999; Köhler et al., 2000). One noteworthy difference between this study and the other studies, is the use of phosphate concentrations instead of total phosphorus concentrations, possibly create a difference in interpretation, as phosphate is always a readily available for uptake, where total phosphorus also contains plant and animal fragments suspended in lake water.
This fact that for none of the model inorganic nitrogen was significant, together with the non-usage of
phosphate in the models, indicates that the decrease in inorganic nitrogen and phosphate concentrations in the lake are not the cause of the changing phytoplankton composition, at least when considering the six selected phytoplankton classes, of the lake. From this it is possible to conclude that the decrease in nutrients which have taken place over the last few years in Lake Markermeer have not significantly influenced phytoplankton class abundances.
In this study, a significant abiotic condition which can explain phytoplankton abundances was found for three out of the six groups, however for Haptophyceae and Tribophyceae only the year parameter came up as significant, and for Cryptophyceae no significant predictor was found at all. As the year predictor is only used to correct for any potential condition in the lake, which were not taken up in the list of candidate predictors, it has no real meaning on its own. This means that for the Cryptophyceae, Haptophyceae and Tribophyceae all five candidates: inorganic nitrogen, phosphate, silica and suspended matter concentrations, as well as residence time, had no significant impact on the biovolumes of these three phytoplankton classes in Lake Markermeer. Although good predictors for Bacillariophyceae, Chlorophyceae and Cyanophyceae were found, the abiotic conditions which can predict Cryptophyceae, Haptophyceae and Tribophyceae still remain to be found. In sharp contrast to previous knowledge and my hypothesis, even for the models where significance has been found, it was only a single abiotic condition. This is not in agreement with the common assumptions, that phytoplankton abundance is most of the time regulated by a combination multiple factors and nutrients, instead of a single one (Vrieze, 2012).
A possible explanation for not finding significant influences of the nutrient concentrations of the lake with the abundance of the different phytoplankton classes, is that the phytoplankton in Lake Markermeer is “top-down” regulated by grazing of zooplankton, instead of “bottom-up” by nutrient availability.
However, the relatively high abundance of small microalgae species could indicate that the zooplankton grazing pressure is low in lake Markermeer (Noordhuis 2010), making this an unlikely assumption.
It could also be possible that the absolute changes in phosphate and inorganic nitrogen concentrations are not large enough to have an effect on changes in the phytoplankton biovolumes. This is possibly a more likely assumption, as in other studies where an effect of changing concentrations of total/inorganic nitrogen and total phosphorus/phosphate on phytoplankton composition was observed, considered larger changes then was the case for Lake Markermeer. Deng et al. (2014) showed a difference in phytoplankton succession in the shallow lake Taihu (China), where changes in total phosphorus were observed between 0.5 mg/L and 3.0 mg/L and
changes in total nitrogen were observed between 2 mg/L and 8 mg/L, a definite larger shift in nutrients then was the case for Lake Markermeer, where total phosphorus shifted between 0.01 and 0.4 mg/L and total nitrogen shifted between 0.50 and 3.70 mg/L. However another study considering the shallow lake Albufera (Spain), already showed a decline in Cyanophyceae and Euglenophyceae species, while showing an increase in Chlorophyceae, Bacillariophyceae, Cryptophyceae and Dinophyceae species, between total phosphorus concentrations of 0.49 mg/L and 0.34 mg/L for the lake, concentrations more similar to those of Lake Markermeer (Villena & Romo, 2003).
Another possibility is that the composition of the “lake snow” is a driving factor behind phytoplankton class biovolumes in Lake Markermeer,as large parts of the nutrients phytoplankton take up are potentially coming from these aggregates.Noteworthy however, is that in these aggregates in Lake Markermeer, no phosphorus has been detected, meaning that for at least phosphate, this is not the case.
Future research of the exact composition of these aggregates would be needed to confirm whether or not, indeed lake snow composition has a major influence of the phytoplankton class abundances.
Conclusion
In summary it was found that suspended matter concentrations plays a large role in determining the
biovolumes of the phytoplankton classes in Lake Markermeer. These suspended matter concentrations impact both Bacillariophyceae and Chlorophyceae mostly, being most beneficial for both of these classes between concentrations of 50 mg/L and 200 mg/L, according to a GAM model. Residence time is mostly important in regulating the biovolumes of Cyanophyceae, where increasing residence times are. Silica concentrations can potentially be a minor factor in the regulation of different phytoplankton classes, although it is argued that any effects silica has shown in the GAM models should be interpreted as noise. No significant relationships between phytoplankton biovolumes and abiotic conditions has been found for Cryptophyceae, Haptophyceae and Tribophyceae species. No significant relationship for any of the nutrient concentrations has been found for any of the phytoplankton classes as well,
Acknowledgements
I would like to thank dr. J.M.H. Verspagen and dr. H.G. van der Geest for their supervision for this bachelor project, and invaluable input on my research method and draft of this report.
References
Akaike, H., 1973. Information theory and an extension of the maximum likelihood principle. In: Petrov, B.N. (Ed.), Proceedings of the Second International Symposium on Information Theory. Akademiai Kiado, Budapest, pp. 267– 281.
Arguelles, E., Martinez-Goss, M., Shin, W., 2014. Some Noteworthy Photosynthetic Euglenophytes from Los Baños, Laguna (Philippines) and its Vicinities. The Philippine Scientist. 51. 1-36.
Barinova, S., Krupa, E.G., 2017. Bioindication of Ecological State and Water Quality by Phytoplankton in the Shardara Reservoir, Kazakhstan. Environment and Ecology Research, 5, 73-92.
Barton K., 2016. Package ‘MuMIn’. cran.r-project. Beheersverslag 2000-2001 Rijkswaterstaat
http://publicaties.minienm.nl/documenten/beheersverslag-rijkswateren-ijsselmeergebied
Bonte, M., 2009, Drinkwaterfunctie Markermeer en verzilting Ijsselmeergebied. KWR watercycle research institute. Bonte, M., Zwolsman, J.J.G., 2010. Climate change induced salinisation of artificial lakes in the Netherlands and consequences for drinking water production. Water Research 44, 4411–4424.
Boyce, D.G., Lewis, M.R., Worm, B., 2010. Global phytoplankton decline over the past century. Nature 466, 591–596. Brinkmann, B.W., Vonk, J.A., Beusekom, S.A.M. van, Ibanez, M., Pardo, M.A. de L., Noordhuis, R., Manders, E.M.M., Verspagen, J.M.H., Geest, H.G. van der, 2019. Benthic hotspots in the pelagic zone: Light and phosphate availability alter aggregates of microalgae and suspended particles in a shallow turbid lake. Limnology and Oceanography 64, 585–596.
Cardinale, B. J., J. E. Duffy, A. Gonzalez, D. U. Hooper, C. Perrings, P. Venail, A. Narwani, G. M. Mace, D. Tilman, D. A. Wardle, A. P. Kinzig, G. C. Daily, M. Loreau, J. B. Grace, A. Larigauderie, D. S. Srivastava & S. Naeem, 2012.
Biodiversity loss and its impact on humanity. Nature 486: 59–67.
Cremona, F., Tuvikene, L., Haberman, J., Nõges, P., Nõges, T., 2018. Factors controlling the three-decade long rise in cyanobacteria biomass in a eutrophic shallow lake. Science of The Total Environment 621, 352–359.
Cronberg, G., 1999. Qualitative and quantitative investigations of phytoplankton in Lake Ringsjön, Scania, Sweden, in: Hansson, L.-A., Bergman, E. (Eds.), Nutrient Reduction and Biomanipulation as Tools to Improve Water Quality: The Lake Ringsjön Story.
Deng, J., Qin, B., Paerl, H.W., Zhang, Y., Wu, P., Ma, J., Chen, Y., 2014. Effects of Nutrients, Temperature and Their Interactions on Spring Phytoplankton Community Succession in Lake Taihu, China. PLoS One 9.
Dokulil, M.T., Padisák, J., 1994. Long-term compositional response of phytoplankton in a shallow, turbid environment, Neusiedlersee (Austria/Hungary) 13.
Drozdenko, T., Mikhalap, S., Nikolskaya, L., Chernova, A., 2017. Assessment of Ecological State of the Velikaya River Delta Based on Hydrochemical Indicators and Structure of Phytoplankton. Open Journal Systems.
Egge, J., Aksnes, D., 1992. Silicate as regulating nutrient in phytoplankton competition. Mar. Ecol. Prog. Ser. 83, 281– 289.
Feld, C.K., Segurado, P., Gutiérrez-Cánovas, C., 2016. Analysing the impact of multiple stressors in aquatic biomonitoring data: A ‘cookbook’ with applications in R. Science of The Total Environment 573, 1320–1339.
Findlay, H. S., Gibson, G., Kędra, M., Morata, N., Orchowska, M., Pavlov, A. K., Reigstad, M., Silyakova, A., Tremblay, J. E., Waldemar, W., Weydmann, A., and Logvinova, C., 2015. Responses in Arctic marine carbon cycle processes: Conceptual scenarios and implications for ecosystem function. Polar Research,
Flöder S., Urabe J. & Kawabata Z.-I. 2002. The influence of fluctuating light intensities on species composition and diversity of natural phytoplankton communities. Oeco-logia 133: 395–401.
Fukuda Y., Suzaki T., 2015. Unusual Features of Dinokaryon, the Enigmatic Nucleus of Dinoflagellates.
van der Geest, H., Vonk, A., & Ouboter, M., 2018. Reconstructie water- en stoffenbalans Markermeer 1976-2015. In Rapport Universiteit van Amsterdam, in opdracht van Rijkswaterstaat.
Hallegraeff ,G.M., Anderson, D.M, Cembella, A.D. , Enevoldsen H.O., 2003. Manual on Harmful Marine Microalgae. Jeppesen, E., Jensen, J.P., Søndergaard, M., 2002. Response of phytoplankton, zooplankton, and fish to re-oligotrophication: An 11 year study of 23 Danish lakes. Aquatic Ecosystem Health & Management 5, 31–43. Jeppesen, E., Søndergaard, M., Jensen, J.P., Havens, K.E., Anneville, O., Carvalho, L., Coveney, M.F., Deneke, R., Dokulil, M.T., Foy, B., Gerdeaux, D., Hampton, S.E., Hilt, S., Kangur, K., Köhler, J., Lammens, E.H.H.R., Lauridsen, T.L., Manca, M., Miracle, M.R., Moss, B., Nõges, P., Persson, G., Phillips, G., Portielje, R., Romo, S., Schelske, C.L., Straile, D., Tatrai, I., Willén, E., Winder, M., 2005. Lake responses to reduced nutrient loading – an analysis of contemporary long-term data from 35 case studies. Freshwater Biology 50, 1747–1771.
Kawachi, M. and Inouye, I., 1995. Functional roles of the haptonema and the spine scales in the feeding process of Chrysochromulina spinifera (Fournier) Pienaar et Norris (Haptophyta=Prymnesiophyta). Phycologia 34, 193-200. Köhler, J., Bernhardt, H., Hoeg, S., 2000. Long-term response ofphytoplankton to reduced nutrient load in the flusehd LakeMüggelsee (Spree system, Germany). Arch. Hydrobiol. 148,209–229
Lange, K., Townsend, C.R., Matthaei, C.D., 2016. A trait-based framework for stream algal communities. Ecology and Evolution 6, 23–36.
Kumar, N.J.I., Oommen, C., 2011. Phytoplankton composition in relation to hydrochemical properties of tropical community wetland, kanewal, gujarat, India. Appl Ecol Env Res 9, 279–292.
Lackey J.B., 1968. Ecology of Euglena. In: Buetow DE, ed. The Biology of Euglena. Vol I. General Biology and Ultrastructure. Academic Press, New York, 28-14.
Lv, L., Wu, H., Chen, M., 2009. Effects of nitrogen and phosphorus on phytoplankton composition and biomass in 15 subtropical, urban shallow lakes in Wuhan, China.
Manton, I., 1967. Further observations on the fine structure of Chrysochromulina chiton with special reference to the haptonema, ‘peculiar’ golgi structure and scale production. J. Cell Sci. 2, 265-272.
Morton, S.D. and Lee, T.H., 1974. Algal blooms-Possible effects of iron. Environmental Science and Technology, 8, 673- 674.
Moss, B., and Balls, H., 1989. Phytoplankton distribution in a floodplain lake and river system. II. Seasonal changes in the phytoplankton communities and their control by hydrology and nutrient availability. J. Plankton Res. 11: 836– 867.
Nõges, T., Laugaste, R., Nõges, P., Tõnno, I., 2008. Critical N:P ratio for cyanobacteria and N2-fixing species in the large shallow temperate lakes Peipsi and Võrtsjärv, North-East Europe. Hydrobiologia 599, 77–86.
Nõges, P., Mischke, U., Laugaste, R., Solimini, A.G., 2010. Analysis of changes over 44 years in the phytoplankton of Lake Võrtsjärv (Estonia): the effect of nutrients, climate and the investigator on phytoplankton-based water quality indices. Hydrobiologia 646, 33–48.
Noordhuis, R. 2010. Ecosysteem Ijsselmeergebied: nog altijd in ontwikkeling. Report IJG0910TD061. Rijkswaterstaat. Nwodo, U. U., E. Green, and A. I. Okoh., 2012. Bacterial exopolysaccharides: Functionality and prospects. Int. J. Olding, D.D., Hellebust, J.A., Douglas, M.S., 2000. Phytoplankton community composition in relation to water quality and water-body morphometry in urban lakes, reservoirs, and ponds. Can. J. Fish. Aquat. Sci. 57, 2163–2174.
Paerl, H.W., 1988. Growth and reproductive strategies of freshwater blue–green algae (cyanobacteria). In Growth and reproductive strategies of freshwater phytoplankton. Edited by C.D. Sandgren. Cambridge University Press, New York. pp. 261–315.
Palmer, W., Ordination Methods - an Overview., n.d. URL:
http://ordination.okstate.edu/overview.htm#Figure_14.%20https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2121141/ pdf/fem0062-0142.pdf (accessed May 2020).
Passy, S. I., 2007. Differential cell size optimization strategies produce distinct diatom richness–body size relationships in stream benthos and plankton. J. Ecol. 95: 745–754.
Pollingher, U., Kaplan, B. and Berman, T., 1995. The impact of iron and chelators on Lake Kinneret phytoplankton. Journal of Plankton Research, 17, 1977-1992.
Reynolds, C. F., 1984. The Ecology of Freshwater Phytoplankton.Cambridge University Press, Cambridge. Reynolds, C. F., 1997. Vegetation processes in the pelagic: A modelfor ecosystem theory. Excellence in Ecology, Ecology Institute,Oldendorf.
Reynolds, C. S., 2006. The ecology of phytoplankton. Cam bridge University Press, Cambridge, U. Salmaso, N., Tolotti, M., 2009. Encyclopedia of Inland Waters.
Sandgren, C.D., J.P. Smol, and Kristiansen J., 1995. Chrysophyte algae: ecology, phylogeny and
development. Cambridge University Press, New York
Stomp M., Huisman J., Stal L.J., Matthijs H.C., 2007. Colorful niches of phototrophic microorganisms shaped by vibrations of the water molecule. ISME J.1(4):271-282. doi:10.1038/ismej.2007.59
Van Riel, M.C., Vonk, J.A., Noordhuis R., and Verdonschot, P.F.M., 2019. Novel ecosystems in urbanized areas under multiple stressors: Using ecological history to detect and understand ecological processes of an engineered ecosystem (lake Markermeer). Notitie Zoetwaterecosystemen, Wageningen Environmental Research, Wageningen UR, Wageningen.
Vadrucci, M. R., Barbone, E., Ungaro, N., Romano, A. and Bucci, R., 2017. Application of taxonomic and morpho-functional properties of phytoplankton communities to water quality assessment for artificial lakes in the Mediterranean Ecoregion. J. Plankton. Res., 39, 550–563.
Vonk, J.A., Bijkerk, R, Noordhuis, R, & van der Geest, H.G., 2019. Verrijkte dataset taxonomische samenstelling van de microalgengemeenschap in het IJsselmeer en Markermeer 1994-2015. (Version 1.0) [Data set]. Zenodo. Xing, W., Liu, G., 2011. Iron Biochemistry and its environmental impacts in freshwater lakes. Fresenius Environmental Bulletin 20, 7.
Villena, M.-J., Romo, S., 2003. Phytoplankton changes in a shallow Mediterranean lake (Albufera of Valencia, Spain) after sewage diversion. Hydrobiologia 506–509, 281–287.
Vrieze, S.I., 2012. Model selection and psychological theory: A discussion of the differences between the Akaike information criterion (AIC) and the Bayesian information criterion (BIC). Psychological Methods 17, 228–243. https://doi.org/10.1037/a0027127
Ward, B.A., Dutkiewicz, S., Barton, A.D., Follows, M.J., 2011. Biophysical Aspects of Resource Acquisition and Competition in Algal Mixotrophs. The American Naturalist 178, 98–112.
Wood, S.N., 2007. Fast Stable Direct Fitting and Smoothness Selection for Generalized Additive Models. Journal of the Royal Statistical Society: Series B (Statistical Methodology) 70.3 (2008): 495–518.
Wood, S.N., 2019. Mixed GAM Computation Vehicle with Automatic Smoothness Estimation. Package “mgcv”, of the
Appendix 1
Changes in abiotic conditions in Lake Markermeer from 1992 to 2016. Monthly averaged values of (a) Total nitrogen concentration, (b) pH of the surface water, (c) Total phosphorus concentration, (d) Temperature of the surface water and
(e) Iron concentrations. The blue line represents the the locally estimated scatterplot smoothing (loess), with the 95% confidence interval in grey, calculated using the function loess of the base package “stats” in R.
Appendix 2
Summary of the GAM model for every taxonomic class. The table includes the predictors used for the model,
the P-value (obtained by the Wald test) of these predictors (significant predictors are displayed with a *), the concurvity scores (concurvity scores higher than 0.8 are displayed with a *), the adjusted R2 of the model and the AIC score for the model. The predictor year refers to the changes in class abundance over the years, which is used to catch all potential influencing factors that were not included in our predictor candidate list. The Chlorophyceae and Cyanophyceae class have multiple GAM modes, because of concurvity effects that can interfere with the reliability of the model. These classes have two versions of the model, one with both the year correction and the predictor that showed the concurvity (in both cases residence time), and an alternative
one without this year variable.
Phytoplankton class Predictors selected P value Concurvity R
2 of the
model AIC
Bacillariophyceae Suspended matter 0.006* 0.271 0.233 640.2
Year 0.287 0.107
Chlorophyceae (Residence Inorganic nitrogen 0.101 0.647 0.29 714.6
time and year inlcuded) Silica 0.29 0.481
Residence time 0.039* 0.922*
Suspended matter <0.001* 0.513
Year 0.2109 0.798
Chlorophyceae (Year Inorganic nitrogen 0.0819 0.554 0.268 716.0
factor excluded Silica 0.0988 0.384
Residence time 0.1344 0.374
Suspended matter <0.001* 0.469
Cryptophyceae Year 0.003* 0.179 -124.3
Cyanophyceae (Residence Silica 0.198 0.382 0.451 444.4
time and year included) Residence time 0.001* 0.915*
Suspended matter 0.067 0.426
Year <0.001* 0.772
Cyanophyceae (Year Silica 0.098 0.251 0.369 464.7
factor excluded) Residence time <0.001* 0.208
Suspended matter 0.105 0.340 Haptophyceae Silica 0.247 0.367 0.211 -218.7 Year <0.001* 0.114 Tribophyceae Silica 0.075 0.471 0.361 -222.2 Suspended matter 0.711 0.486 Year <0.001* 0.290
Appendix 3
Correlation matrix of all abiotic conditions in Lake Markermeer. From top to bottom: (a) Total Phosphorus concentration, (b) Suspended matter concentration, (c) Temperature of the surface water, (d) Total nitrogen concentration, (e) Total
inorganic nitrogen concentration, (f) Phosphate concentration, (g) Residence time of the water and (h) Silica concentrations. Diagonally trough the plot, histograms for every variable was shown. On the left side of the diagonal
bivariate scatterplots, with a fitted line are shown and on the right side the Spearman-correlation coefficient and significance level; for *<0.05, for **<0.01 and for ***<0.001. The base package “stats” in R is used to calculate the
Appendix 4
a
Figure 5. Plots showing the combined effect of the linear and nonparametric contributions for each selected abiotic factor to the phytoplankton class abundance in Lake Markermeer according to a GAM model. (a) Original Chlorophyceae
model (year variable included) and (b) original Cyanophyceae model (year variable included). The blue line represents the predicted abundance value whereas the grey area is the 95% confidence interval (only shown for abundances above 0
