Modelling of hydrogen peroxide to combat toxic algae blooms

(1)

Internship report on the

modelling of hydrogen

peroxide to combat toxic

algae blooms

Name: Darryl Holsboer

University: University of Amsterdam Internship host: Deltares

Examiner: dr. Jolanda Verspagen Assessor: dr. Petra Visser

Daily supervisor: dr. Anouk Blauw Advisors: dr. Miguel Dionisio Pires & dr. Tineke Troost

(2)

1

1. Thematic summary

Cyanobacterial blooms are a growing problem worldwide due to their negative impact on

ecosystem functioning and water quality. Hydrogen peroxide (H2O2) has been implemented in the last decade as a cost effective and environmentally friendly method, which may be used to selectively kill cyanobacteria. In The Netherlands this method has been performed with varying degrees of success. Thus, the main aim of this internship has been to develop a model that can predict for a wide variety of field conditions the optimal application timing and dosage of hydrogen peroxide in order to most effectively combat toxic cyanobacterial blooms. This model was developed based on lab and field data provided by the University of Amsterdam. The lab data was used to set up and calibrate the model, and the field data was used for validation. The degradation of hydrogen peroxide (H2O2) and the decrease in photosynthetic yield (proxy for phytoplankton mortality) in dark and light conditions were the primary modelled processes. Additionally, a sensitivity analysis was performed with 10 scenarios based on important conditions such as: the target concentration H2O2, the solar irradiance, turbidity, presence of eukaryotic phytoplankton, stratified or mixed water columns and the delayed degradation of H2O2. Degradation of H2O2 and mortality of cyanobacteria could be modelled relatively well in the lab. Field data proved to be more difficult to predict, as many field conditions could not be considered. Still, the sensitivity analysis indicated that turbidity, mixing depth, the presence of eukaryotic plankton and to a lesser degree the delay in H2O2

degradation determine the effectiveness of H2O2in this model. These results suggest that treatment in early bloom stages would be most effective.

(5)

4

2. Description host institute

Deltares is an applied non-profit independent research institute active in the fields of subsurface, infrastructure and water. It operates both nationally and internationally and is home to a wide range of nationalities. Currently over 800 people are employed by Deltares; located in Delft and Utrecht. The organization was established in 2008 as a merger of GeoDelft, Delft Hydraulics and parts of TNO and the Dutch ministry of water and infrastructure.

Deltares aims to utilize the Dutch expertise on deltas and water management to develop innovative and sustainable solutions for global problems related to water and soil. In the coming years Deltares strives to become both nationally and internationally recognized as a ‘Triple A’ institute in these domains (Deltares, 2018). The primary focus is on coastal areas, river catchments and deltas, since these areas are increasingly being settled by people due to their economic potential. They are however also very susceptible to the effects of climate change. Innovation and technology are key in tackling these complex densely populated systems, which opens new opportunities for Deltares and the Netherlands as a whole (Deltares, 2012).

The organisation is divided in six main units (Fig. 1). During the internship I was part of the unit inland water systems. This unit can be further divided in six departments, of which I was part of the Freshwater Ecology and Water Quality department (Fig. 1).

Figure 1: Deltares organisation chart

My activities were primarily related to the modelling in R of hydrogen peroxide effectiveness in combatting toxic cyanobacterial blooms. Data fitting, model calibration, model validation and sensitivity analysis were the most important steps undertaken.

I moreover did some additional literature research to fill in some knowledge gaps and I performed data exploration and management in excel and Microsoft access on the datasets provided by the University of Amsterdam and Dutch water boards. There were plans for a short lab experiment with staining of cells using a flow cytometer to determine the actual mortality of cyanobacteria, but due to time constraints it was decided to instead use the photosynthetic yield (Fv/Fm) as a proxy for the

(6)

5 mortality. Lastly, a report was written which is not only used to present my findings but also to document how the model has been developed and what choices and assumptions were made.

3. Personal reflection

My internship was part of a collaboration between the University of Amsterdam and Deltares Delft on the effectiveness of H2O2 in treating toxic cyanobacterial blooms.

I started this internship under the supervision of dr. Anouk Blauw with three main objectives. First, I wanted to apply the theoretical knowledge that I had gathered over the years at the University of Amsterdam. Second, I wanted to improve my programming skills and become more comfortable using these skills. Third, I wanted to gain some valuable work experience.

During my internship I learned to be more independent, to cooperate and more importantly to ask questions. Especially in the beginning I frequently discussed with the University of Amsterdam and contacted other employees of Deltares for help when my daily supervisor was not available. While there was a clear path I had to follow, I was free to provide my own input where necessary. I moreover further improved upon my presentation skills, especially regarding the bi weekly progress meetings, where I was required to keep my daily supervisor and advisors up to date. I also attended a couple of department meetings where I presented my work to fellow department colleagues.

In terms of programming I learned to create a simple numerical model from scratch in R studio, which I had not done before. The programming courses I had followed at the University of

Amsterdam, which consisted mainly of Matlab with some R studio, fortunately proved to be enough of a basis. The difference in programming language between Matlab and R studio did not pose too much of a problem.

The hardest part of the internship as expected was to translate lab experiments to field situations. Keeping the model simple, but not too simple was a key learning objective in this process. Working with a limited amount of data and coming up with smart and efficient solutions was thus important. On my own this would have been a difficult task, but with close collaboration with my daily

supervisor I managed to overcome or simplify most of the complexities in the model.

Ultimately, I learned a lot and I think I did achieve most of my objectives. I feel like I contributed to Deltares by developing this model and I of course hope this model will be used and improved upon in the future. I am moreover a lot more comfortable with my programming skills and would even consider working in modelling related jobs. Joining the meetings also gave me more insight in the inner workings of the company. I think working at Deltares showed me that there are many fields where theoretical knowledge in earth sciences and environmental management may be applied.

(7)

6

4. Introduction

Toxic algae or cyanobacterial blooms are a common and increasing global problem occurring most frequently in eutrophic ecosystems. The effects of these blooms can significantly impact ecosystem services such as providing drinking water, recreation and agriculture (Lürling et al., 2014). Blooms can negatively impact water quality by increasing turbidity and producing malodors; they can alter ecosystems by causing anoxia and the subsequent death of other aquatic organisms. Cyanobacteria also produce a wide range of toxins, of which microcystins are the most common (Lürling et al., 2014). These toxins can badly affect the nervous, dermal, digestive, endocrine and hepatopancreatic systems in humans and mammals (Paerl, 2014).

Cyanobacteria have adapted to a wide range of conditions and environments because of their extensive evolutionary history (Paerl, 2014). This evolutionary genetic diversity has permitted cyanobacteria to outcompete other phytoplankton species under eutrophic conditions.

This is because some cyanobacteria can fixate atmospheric nitrogen and are thus able to grow in nitrogen limited environments, moreover in presence of elevated levels of phosphate and iron bloom development can occur (Huisman et al., 2018).

Cyanobacteria are buoyant and can float to the surface, due to hollow gas filled cell compartments called gas vesicles. Floating cyanobacteria can block light from reaching the lower depths, thereby stifling the growth of competing less harmful eukaryotic phytoplankton. Some species can even migrate up and down the water column such as the Planktothrix rubescens enabling a wider potential area where nutrients may be exploited (Huisman et al., 2018).

Cyanobacterial toxins may act as prevention against grazing and may help with reducing oxidative stress by binding on cyanobacteria specific proteins and enzymes (Huisman et al., 2018).

Multiple methods to combat toxic algae blooms are available and have been successful in

combatting cyanobacteria such as: copper sulfate (Van Hullebusch et al., 2002), flock and lock (de Magalhães et al., 2017; Waajen et al., 2016), ultrasonication (Ahn et al., 2007), flushing and artificial mixing (Visser et al., 1996; Verspagen et al., 2006). However, most of these are either expensive, environmentally unfriendly or do not degrade cyanotoxins (Matthijs et al., 2016, Weenink et al., 2015; Barrington et al., 2015). Hydrogen peroxide is a possible solution that could help with these negative effects, since it is relatively inexpensive, degrades into water and oxygen in a couple of days, selectively kills cyanobacteria by targeting the photosystems and could even help reduce released concentrations of cyanotoxins (Barrington et al., 2015; Mikula et al., 2012).

In the Netherlands, experimentation with H2O2 in the field started in 2009 (Matthijs et al., 2012). In this initial experiment a 2mg/l H2O2 successfully reduced cyanobacteria concentrations by 99% within the first couple of days with minimal effects on remaining aquatic organisms. A similar decline was visible for cyanotoxins, but with a 2-day lag. Whether a treatment succeeds or fails is very dependent on specific field conditions. Since this initial experiment H2O2 has been applied over the years to varying degrees of success.

In order to improve the success rate of treatments and to better understand the underlying mechanisms a model has been developed in order to predict for a range of field conditions the optimal application timing and dosage of hydrogen peroxide in order to most effectively combat toxic algae blooms. For the field conditions the following parameters were considered: cloudiness,

(8)

7 turbidity, concentration H2O2, time of H2O2 degradation and the composition cyanobacteria and eukaryotic phytoplankton.

5. Theoretical framework

5.1. H2O2 as a selective algicide

H2O2 can be applied as a selective measure to combat cyanobacterial blooms. This is because in contrast to eukaryotic phytoplankton species, cyanobacteria utilize different reactions during photosynthesis. The eukaryotic phytoplankton species use the Mehler-reaction which produces H2O2. This toxic chemical is thereafter neutralized by catalase and peroxidase enzymes. In

cyanobacteria this reaction is replaced with a Mehler-like reaction. The key difference is the absence of H2O2 production during photosynthesis, instead H2O is produced by Flavodiiron proteins

(Allahverdiyeva et al., 2014). Because there is no H2O2 produced it is believed that cyanobacteria have a lower natural defense to H2O2. This is also supported by Weenink et al. (2015), according to whom cyanobacteria have less enzymes that neutralize H2O2 than eukaryotic phytoplankton species.

5.2. Toxic versus nontoxic strains

Cyanobacteria have some defense mechanism against H2O2, though the effectiveness differs between species and strains of cyanobacteria. According to a study by Schuurmans et al. (2018), cyanobacteria that produce microcystins may be more vulnerable to H2O2 than other species of cyanobacteria depending on the H2O2 concentration. Under natural conditions (1 – 50 µgram H2O2/l) toxic strains of cyanobacteria appear to be better protected against H2O2. During treatments

however these concentrations are much greater and can reach as high as 120 mg/l in some specific cases (Huo et al., 2015). At these concentrations H2O2 becomes toxic to more than just

cyanobacteria, however even at slightly elevated concentrations in the order of two to three times the natural level, toxic strains lose their advantage to non-toxic strains, who appear to be better protected to oxidative stress (Schuurmans et al., 2018). It is thus far unknown if this holds true for other species of cyanobacteria, since there are many possible mechanisms that cyanobacteria utilize to manage oxidative stress (Schuurmans et al., 2018).

5.3. H

2

O

2

effectiveness

The degradation rate of hydrogen peroxide (H2O2) is dependent on the available oxidizable compounds. In a study conducted by Tao et al., (2009) cellular respiration and enzymatic

degradation of H2O2 was compared with and without added glucose. Without glucose measured O2 concentrations decreased slowly and did not increase after addition of catalase. With glucose measured O2 decreased faster, but also increased after catalase was added. This is because catalase is an enzyme that catalyzes the degradation of H2O2 into H2O and O2. With glucose more O2 is produced than is consumed, which indicates that more H2O2 is degraded when more glucose is available.

(9)

8 Light intensity does not appear to alter the rate of H2O2 degradation in lab conditions (Piel et al., unpublished). However, light intensity in combination with application of H2O2 is suggested to strongly influence the reduction of Fv/Fm in cyanobacteria. Fv/Fm is the ratio between variable fluorescence of chlorophyll and minimum fluorescence of chlorophyll and is a widely used ratio to measure stress in plants and cyanobacteria (Matthijs et al., 2012). It is hypothesized that a higher light intensity leads to the generation of more oxygen radicals due to the photo fenton reaction (Ruppert et al., 1993; Piel et al, unpublished). This in turn can inflict more damage on the

photosystems of cyanobacteria. Depending on the concentration H2O2 and light intensity this could imply an order of magnitude greater decrease of Fv/Fm under light conditions compared to dark conditions (Drábková et al., 2007).

6. Data description

Lab data provided by the University of Amsterdam was utilized for the calibration and validation steps of the model in this report and was retrieved from an experiment studying the effectiveness of hydrogen peroxide (H2O2) at certain concentrations and light intensities (Fig.2). Concentrations in this experiment were tested for 0, 1, 2, 4, 6 and 10 mg/l H2O2 and light intensities for 0,15,50,100 and 150 µmol photons m-2_s-1_{. For every configuration of concentration and light intensity a sample} was prepared and for every sample a duplicate was prepared, which in total yielded 60 samples. The samples were cultivated in chemostats.

This initial dataset contained the decrease in H2O2 concentration (mg/l), the decrease in Fv/Fm over time and general data on cell abundance and cell size measured by a CASY cell counter and PAM analyser, which was used in the experiment to measure the Fv/Fm. This general data consisted of cell counts (cells/ml), biovolume (fl/µl), mean diameter (µm) and maximum diameter (µm).

The temporal accuracy of this dataset is 24 hours with measurements every 30 minutes in the first 4 hours and a final measurement was taken at 24 hours.

Figure 2: Set up of the lab experiment on the effectivity of H2O2 in killing cyanobacteria (Piel et al, unpublished). For validation two field datasets were used from the University of Amsterdam obtained from H2O2 treatments of the Klinkenbergerplas and Oosterduinse meer. Fv/Fm and concentrations H2O2 were measured during the treatment over a period of roughly 24 hours and at different depths (0 m, 2.5 m, and 5 m below the surface and an integrated sample over the top 6 m).

These field datasets also included the species composition (Fig.3 & Fig. 5) of cyanobacteria and eukaryotic phytoplankton measured from May to September (Fig. 4 & Fig.6), Secchi depth for the Klinkenberger plas and hydrolab data for both lakes. The Secchi depth is measured with a Secchi

(10)

9 disk. The depth at which the Secchi disk loses its visibility is used as a measure of transparency (Preisendorfer, 1986). The hydrolab is a multiparameter probe used for measuring water quality. Secchi depth and hydrolab measurements were measured for the months May to August. The hydrolab measured: depth, temperature, salinity, photosynthetically active radiation (PAR),

chlorophyll fluorescence (in volt), dissolved oxygen and specific conductance over nearly the entire depth of both lakes.

Additionally, a monitoring dataset with chemical and physical properties of Dutch water bodies was available from Dutch water boards, however this dataset was not used in the development of this model.

Klinkenberger plas

Figure 3: Dynamics in biovolume of cyanobacteria and eukaryotic phytoplankton in Klinkenberger Plas from June 15 2017 to June 18 2017.

(11)

10 Figure 4: Physical parameters in Klinkenberger Plas on June 15 2017. Depth profiles of (a) the density of water and (b) chlorophyll fluorescence at 7:38 in the morning.

The density was calculated only from temperature, since no salinity was measured in this lake (Appendix B2 & B3) with the following simplified equation Eq. (1):

𝜌(𝑇) = 𝜌 × (1 − 𝛾(𝑇 − 4°𝐶)2_{) + 𝑆 (Eq. 1)}

B

A

(12)

11 Where 𝜌 = density (1000 kg m-3₎

𝑇 = temperature in ° C

𝛾 = temperature dependent change in density (6.493 x 10-6_K-1₎ 𝑆 = salinity in (g/l)

In the hydrolab and species composition data of the Klinkenbergerplas (Fig. 3 & Fig. 4) three main observations could be made. First, the concentrations of cyanobacteria appeared to be much greater than the concentrations of eukaryotic phytoplankton. Roughly 99% of the measured phytoplankton consisted of cyanobacteria. Second, in the upper 2m the density was relatively constant, suggesting the presence of a mixed layer depth. Third, two chlorophyll peaks could be observed in the data at a depth of around 2m and 5m below the surface.

Oosterduinse meer

Figure 5: Dynamics in biovolume of cyanobacteria and eukaryotic phytoplankton in Oosterduinse meer from August 7 2018 to August 8 2018.

(13)

12 Figure 6: Physical parameters in Oosterduinse meer on August 7 2018. Depth profiles of (a) the density of water and (b) chlorophyll fluorescence at 6:30 in the morning.

A

(14)

13 In the hydrolab and species composition data for the Oosterduinse meer the conditions were

different compared to the Klinkenbergplas (Fig.5 & Fig.6). One reason was because of the increased presence of eukaryotes, which for the Oosterduinse meer made up roughly 40% of the total

concentration phytoplankton.

Furthermore, only one chlorophyll peak was present at around 1.8 m below the surface. A similarity with the dataset of the Klinkenberger plas could also be observed, since the pycnocline, the point at which the density changes, appeared to be located at roughly the same depth (1.6m). The key difference being that this change in density appears to be related to salinity as opposed to temperature (Appendix C2 & C3).

7. Methodology

7.1. Approach

The study was divided into five phases (Fig. 7). In the first phase a basic model was set up using scientific literature and data of lab experiments. In the second phase the model was calibrated and validated with the same lab data used to set up the model. Modelled values were compared to measured values for the H2O2 concentrations and Fv/Fm. In the third phase the model was validated with field data; the results of this validation were first produced from the lab model but later updated to better emulate the field observations with an updated model in phase four. The third and fourth phases were repeated until the results could not be improved any more. In phase 5 a sensitivity analysis was performed to test the model with different field conditions namely: cloudiness, turbidity, concentration H2O2, time of H2O2 degradation and the fraction of cyanobacteria to find the most optimal application timing.

Figure 7: Modelling approach visualized

7.2. Phase 1: Setting up the process parameters of a basic model

7.2.1. Determining H

2

O

2

degradation rate

In the lab data duplicate samples were averaged before analysis. Degradation speeds were

subsequently derived by initially plotting H2O2 concentrations over time to get a feel for the data in excel. Trend lines were then used to retrieve the best fit. Degradation rates of H2O2 were measured

(15)

14 in the presence and absence of cyanobacteria in the lab. This model used the degradation rate of H2O2 in the presence of cyanobacteria.

For the time series a time period of 24 hours was taken with a timestep of 0.1 hours (6min) and for data fitting an exponential model was deemed to be the best model to describe the process. This is also in accordance with results from Richard et al. (2007). From the underlying equation the degradation rate was derived.

In R studio Version 1.2.1335, the data was subsequently plotted and fitted with an exponential model similarly to excel (Eq. 1). This was performed with a custom exponential function and the standard NLS function from the stats package in R. The “NLS” function used the custom exponential function as an input and calculated the nonlinear (weighted) least- squares estimates, which was thereafter used to retrieve a best fit by using the “coef” function. With the curve function the best fit could then be visualized in a plot.

𝐾

_𝐻2𝑂2

= 𝑎 + 𝑒

𝑏∗𝐶 (1) Where 𝑎 is the intercept, 𝑏 is the slope, 𝐾𝐻2𝑂2 is the degradation rate of H2O2 and C is the initial

concentration H2O2

In the lab every initial concentration of H2O2 decayed at a different rate, which necessitated one specific function that could explain this variation. At higher concentrations there appeared to be a saturation point where the decay did not decrease any further. The decay rates were thus plotted against the initial concentration and a Michaelis-Menten equation was determined to be the best fitting model, since the H2O2 degradation appears to increase asymptotically towards a maximum value with increasing initial concentration H2O2 (Eq.2). It should be noted that this is only valid for the range of 1 to 10 mg/l H2O2. No lab data was available for concentrations below 1mg/l. This is not a problem, since lower concentrations are not used in field treatments and the degradation rate in this model is based only on the initial concentration H2O2.

𝑟 =

𝑉

𝑚𝑎𝑥

∗ 𝐶

𝐶 + 𝐾

(2)

Where C = initial concentration H2O2 r= degradation rate of H2O2 Vmax= maximum decay rate of H2O2

K= half-saturation constant, i.e. the concentration at which the degradation rate is half the maximum rate.

(16)

15 The Michaelis-Menten equation is often used in biochemistry to describe enzymatic processes (Berg et al., 2002). It has moreover already been used in studies analyzing the degradation of H2O2 (e.g., Henzler & Steudle, 2000).

Conversely, no relation was found between light intensity and H2O2 degradation speed, even though degradation under different light intensities was tested in the lab. Therefore, in order to prevent unnecessary complexity, the effect of light intensity on the degradation of H2O2 was not included in the model.

7.2.2. Incorporating H

2

O

2

degradation rate in the model

In the model H2O2 degradation was predicted over a period of 24 hours with a timestep of 0.1 hours/6 minutes. The H2O2 degradation rate and its coefficients were calculated from Eq. (2) and inserted in the data frame as:

H2O2

_model

= H2O2

_model

[t − 1] + (H2O2

_model

[t − 1 ∗ 𝑟 ∗ dt)

(3) Where t = timestep

dt = interval

H2O2model = the predicted concentration H2O2

The new concentration of H2O2 is predicted based on the concentration of the previous timestep and on the concentration of the previous timestep times the H2O2 degradation and the timestep Eq. (3).

The degradation rate of H2O2 was assumed to be constant and based on the initial concentration.

7.2.3. Determining cyanobacteria mortality

The mortality of cyanobacteria was derived from Fv/Fm lab data. Fv/Fm can also be described as cyanobacteria vitality. However, for this model it was assumed that a decrease in Fv/Fm is the same as cyanobacteria mortality. This assumption is valid as long as the measured photosynthetic yield reaches zero, since at that point all cyanobacteria are deceased. If the Fv/Fm is higher than zero, the cyanobacteria may still be able to recover.

Duplicate samples were available in the dataset and were averaged before analysis. Additionally, a control group was present, which represented the change in Fv/Fm without added H2O2. An absolute and relative Fv/Fm could thus be used.

It was decided to use the absolute measurements, since the results could then more easily be compared to the field data where no control group is available. Fortunately, the change in Fv/Fm over time in the control group did not appear to be affected much compared to the effectiveness of H2O2 at higher concentrations and light intensities.

Finally, the initial Fv/Fm values for the samples in darkness were left out, as these samples were moved from light to dark conditions during the lab experiment and were not yet adapted to the new conditions (personal communication T. Piel, October 18, 2019).

(17)

16 Like the degradation of H2O2, a constant decay rate for cyanobacteria mortality was estimated. For all initial H2O2 concentrations an exponential model was used for data fitting Eq. (4). The data for 0 and 1 mg/l appeared to follow a linear trend and were therefore excluded from the fitting process.

𝑀 = 𝑎 + 𝑒

𝑏∗𝐶 (4) Where a= intercept and b= slope

In R studio the function glm/lm was utilized to fit linear models. The coef() function was used to derive the coefficients and the abline() function to fit a straight line through the input data. For exponential models the steps were identical to the steps in section 7.2.1. These steps were subsequently replicated for the different light intensities.

Similarly, to the degradation of H2O2, there were multiple decay rates dependent on both the initial concentration H2O2 and on the light intensity. In order to obtain one equation of the mortality rate, the effect of H2O2 concentration and the effect of light intensity had to be separated. The effect of H2O2 concentration could be derived from mortality in darkness by plotting the mortality rate against the initial concentration H2O2. The resulting data spread was subsequently fitted with a linear model as Eq. (5):

𝑀

_𝑑

= 𝑎 ∗ 𝐶 + 𝑏

(5) Where a = slope, b= intercept and 𝑀𝑑 = the mortality rate in darkness

Thereafter, the effect of light intensity was derived from the calculated mortality rates. These mortality rates were plotted against the light intensity and were fitted with a PI curve or linear model.

The PI curve or photosynthesis- irradiance curve is a mathematical model first proposed by Blackman (1905) to describe the relationship between light and photosynthesis. In a comparative study by Jassby & Platt (1976) eight variations of this model were tested. It was concluded that before photo inhibition occurs the data is most consistently fitted with a hyperbolic tangent Eq. (6). For subsequent steps the data for 1 mg/l was omitted, since it was not able to be fitted with a PI curve.

(6) Where 𝑙 = light intensity

𝑀𝑙 = mortality rate in light

𝐴𝑙 = mortality rate at optimal light intensity

𝐵𝑙= initial slope

This equation however only considers the variation in light intensity. To add the variation in H2O2, coefficients 𝐴𝑙

and 𝐵𝑙 were plotted against the initial concentration H2O2. The resulting equations

Eq. (7) & Eq. (8) were described as:

(18)

17 (7) Where 𝐴𝐴

and BA are coefficients to calculate 𝐴𝑙

(8) Where 𝐴𝐵

and BB are coefficients to calculate 𝐵𝑙

The 𝐵𝑙 coefficient was fitted with a PI curve, whereas the 𝐴𝑙 coefficient was fitted with a linear

model.

7.2.4. Incorporating cyanobacteria mortality in the model

The same equations used for data fitting are used in the model. The decrease in Fv/Fm was used as a proxy for mortality. The mortality 𝑀 was calculated from Eq. (5 & 6) and subsequently used to calculate the predicted concentration of cyanobacteria over time. This was added to the model as:

𝐶𝑜𝑛𝑐𝐴𝑙𝑔 = 𝐶𝑜𝑛𝑐𝐴𝑙𝑔[𝑡 − 1] + (𝐶𝑜𝑛𝑐𝐴𝑙𝑔[𝑡 − 1]

∗ 𝑀[𝑡] ∗ 𝑑𝑡)

(9)

Where 𝐶𝑜𝑛𝑐𝐴𝑙𝑔 = the predicted concentration of cyanobacteria and 𝑀 = 𝑀𝑑 if 𝑙 = 0.

If 𝑙 > 0 then 𝑀 = 𝑀𝑙.

This separation was necessary, since mortality may still occur in darkness if the concentration of H2O2 is high enough. There was moreover a discontinuity present between the decay rates in darkness and in light.

In order to combine the mortality in darkness with the mortality in light Eq. (6) was adapted as Eq. (10):

𝑀

_𝑙

= 𝑀

_𝑑

[𝑡] + 𝐴

_𝑙

× tanh(

𝐵𝑙×𝑙[𝑡−1]

𝐴_𝑙

)

(10)

The coefficients of 𝑀𝑙 were kept constant over the duration of the treatment, since the degradation

of H2O2 in the lab was explained by a constant degradation rate. Light intensity was either kept constant in case of calibration with lab data or made variable in case of validation with field data.

𝐴

_𝑙

= 𝐴

_𝐴

∗ 𝐶 + B

_A

(19)

18

7.3. Phase 2: Validation of the lab model

Validation of the model was performed by plotting measured concentrations of H2O2 and Fv/Fm, against the modelled concentrations of H2O2 and Fv/Fm.

7.4. Phase 3: Implementing modifications for a field model

The model was initially calibrated with lab data. However, in the lab the light intensity was constant, while in the field light intensity was variable over time and over depth. Three things were required to determine light intensity in the field. 1) The law of Lambert-Beer, which was applied to determine light intensity over depth for one specific point in time. 2) In order to capture the variation in irradiance throughout the day measurements of the Royal Netherlands Meteorological institute (KNMI) were used. 3) The light extinction coefficient, which is used in the Lambert-Beer equation, was derived from hydrolab data.

7.4.1. Calculating light intensity over depth in the field using two light models

Validation of the model was performed with two datasets from the Oosterduinse meer and the Klinkenbergerplas. In order to better simulate light conditions in the field two light models were implemented for this purpose according to the law of Lambert-Beer: a model to calculate light intensity at depth Z Eq. (11), which was used for stratified lakes and a model to calculate light intensity over the mixed layer depth Zm Eq. (12), which was used for lakes with mixed layers. These models can be described as:

𝐼𝑧 = 𝐼

₀

𝑒

−𝐾𝑑𝑍₍₁₁₎

𝐼𝑚 = 𝐼

₀

(1 − 𝑒

−𝐾𝑑𝑍𝑚

)

₍₁₂₎

Where Iz is the irradiance (in μmol photons m-2_s-1_{) at depth Z (in m), Im is the irradiance at mixed} layer depth Zm, I0 is the solar irradiance at the surface, Kd is the light extinction coefficient (in m-1).

These two models were used so a distinction could be made between stratified lakes and lakes where mixing occurred.

(20)

19 Hourly measurements of the solar irradiance (in J cm-2_h-1_{) from the weather stations of Voorschoten} and Schiphol were used as the input for the irradiance at the water surface. Since, cyanobacteria do not utilize the entire solar spectrum a correction factor of 0.45 was used to convert sunlight to PAR. This factor approximates the fraction PAR of the total solar irradiance (Kirk, 1994). Thereafter, J cm-2 h-1 _{was converted to μmol photons m}-2_s-1_{, using Avogadro’s constant and with the assumption that} one Joule equals 2.77 x 1018 _{photons (Kirk, 1994).}

The KNMI data consisted of hourly measurements and could not be used directly in the model. A linear interpolation was thus applied with the ‘approx’ function to calculate solar irradiance at timesteps of 0.1 hours. This function requires vectors of the points to be interpolated and optionally a set of values that indicate where interpolation is to take place.

7.4.3. Deriving the K

d

from hydrolab data

For the Klinkenbergerplas and Oosterduinse meer depth-profiles were available on the days before during and after treatment and for multiple time points. The data included measurements of light intensity under water measured in PAR (Appendix B1 & C1) and chlorophyll measured in volt (Fig. 3 & 4.

It was not possible to derive the Kd from just the measured PAR, since the cyanobacteria were

unevenly spread over the depth of the lake. The light intensity was therefore predicted and

calibrated by comparing the measured chlorophyll profile and a specific extinction coefficient to the observed light intensity in PAR as Eq. (13):

𝐼𝑧

_{𝑝𝑟𝑒𝑑}

= 𝐼

₀

𝑒

−𝐾𝑐𝐾𝑠𝑝𝑍

₍₁₃₎

Where 𝐾𝑠𝑝 is the specific extinction coefficient and 𝐾𝑐 is the measured chlorophyll over depth in

volt.

Measurements were not available for every depth and point in time, which required the use of linear interpolation to estimate the missing values. For depth, duplicate measurements were present in the data. These were averaged before interpolation with the ‘aggregate’ function. This function groups by columns and then calculates a summary statistic such as the mean over the entire dataset. Linear interpolation was subsequently performed for chlorophyll over depth and PAR over depth. No clear interval was present in the original data. The resolution was around 0.01 m and as such an interval of 0.1 m was chosen, which was a compromise between accuracy and data frame size. The calculation of the Kd was performed in two steps namely: calibration of the predicted light

intensity to derive a specific extinction coefficient and interpolation of the chlorophyll-depth profiles.

Initially, a Kd was derived by multiplying the chlorophyll-depth profile with a specific extinction

constant Eq.(14). This Kd was then put into Eq. (15) to calculate the predicted light profile. The

predicted light profile was then plotted against the measured PAR profile and could subsequently be calibrated, by changing the specific extinction constant. These steps were repeated for every

available hydrolab depth profile on the days of the treatment. The specific extinction constant was calibrated to best fit all hydrolab PAR-depth profiles.

(21)

20

𝐼𝑧

_{𝑝𝑟𝑒𝑑}

= 𝐼

₀

𝑒

−𝐾𝑑𝑍

₍₁₅₎

Where 𝐼𝑧𝑝𝑟𝑒𝑑 is the predicted light intensity at depth Z (in μmol photons m-2 s-1).

Finally, an interpolation over time was performed for the newly calibrated Kd-depth profile, since

hydrolab measurements were not available for every hour much less every 6 minutes. This was accomplished with chlorophyll- depth profiles from before, during and after the treatment. These steps were performed for the Klinkenbergerplas and the Oosterduinse meer.

7.4.4. Using the predicted K

d

to calculate the light intensity over time and depth

The predicted light intensity under water was calculated over depth and time using the calculated Kd-depth profiles in section 7.3.3. and with Eq. (11) and Eq. (12) as:

𝐼𝑧𝑝𝑟𝑒𝑑𝑖𝑐𝑡[𝑡, 𝑑] = 𝐼𝑧𝑝𝑟𝑒𝑑𝑖𝑐𝑡[𝑡, 𝑑 − 1] × 𝑒𝐾𝑑×𝑑𝑡 (16)

𝐼𝑚𝑝𝑟𝑒𝑑𝑖𝑐𝑡[𝑡, 𝑑] = 𝐼𝑚𝑝𝑟𝑒𝑑𝑖𝑐𝑡[𝑡, 𝑑 − 1] × (1 − (𝑒𝐾𝑑×𝑑𝑡)/(−𝐾𝑑× 𝑑) (17)

Where 𝑡 was the timestep, 𝑑 was the model iteration over depth and 𝑑𝑡 was the depth interval. Interpolated KNMI data was used as the input for light intensity calculated at the surface.

The measured light intensities of the KNMI did not always correspond with the measured PAR under water, which is explained by the difference in sensors namely: flat versus spherical. A conversion factor of four based on the conversion from a sphere to a circle was therefore used, since the lab calibration is based on light intensities measured by the same spherical sensor.

For the layers below the surface the Lambert-Beer equations were used. The resulting predicted light intensity could then be used to calculate the predicted concentration of cyanobacteria with Eq. (9).

7.4.5. Application of H

2

O

2

in the field

In the field H2O2 did not decrease as expected, since the method of application and field conditions determine when and how quickly H2O2 is degraded. This could not be adequately predicted by the lab model. Thus, instead of calculating H2O2 concentrations the field model uses actual measured H2O2 concentrations.

7.5. Phase 4: Recalibrating and updating the model

7.5.1. Changing the mortality rate from constant to variable

In the field measured concentrations H2O2 fluctuated over time and depth. To improve the results mortality rates were calculated not from the initial concentration H2O2 but recalculated every time step based on the actual measured concentration of H2O2.

(22)

21

7.5.2. Species composition

Besides conditions related to light availability, the effect of species composition was also included in the model for the field validation. The fraction cyanobacteria of the total phytoplankton population was used as proxy for this effect, since not enough data was available to more accurately predict this effect. This fraction was then added to Eq. (9) as Eq. (18):

𝐶𝑜𝑛𝑐𝐴𝑙𝑔 = 𝐶𝑜𝑛𝑐𝐴𝑙𝑔[𝑡 − 1] + (𝐶𝑜𝑛𝑐𝐴𝑙𝑔[𝑡 − 1]

∗ 𝑀[𝑡] ∗ 𝑑𝑡 ∗ 𝐹

_𝑎𝑙𝑔

)

(18)

Where 𝐹𝑎𝑙𝑔

=

the fraction cyanobacteria of the total phytoplankton population

7.6. Phase 5: Implementing modifications for the sensitivity analysis

The final aim of the study was to select optimal field conditions for the application of H2O2, so a minimal dosage of H2O2 could be used to effectively combat cyanobacterial blooms. A sensitivity analysis was applied for this purpose and some additional modifications were thus made to the model. Implementation details are given below.

1) Instead of observed weather data, the predicted solar irradiance was used, based on latitude, time of year and cloudiness, so that the model could be applied anywhere in the world.

2) The turbidity was calculated using the Secchi depth, instead of hydrolab data which is generally not available.

3) The possibility to delay the degradation of H2O2.

7.6.1 Calculation of the theoretical solar irradiance at the water surface

The theoretical solar irradiance 𝐸𝑡 was calculated based on the implementation in DELWAQ and is

described as Eq. (19):

𝐸

_𝑇

=

𝐼0

𝜋 𝑅2

𝑅2

(sin 𝛿 sin 𝜙 + cos 𝛿 cos 𝜙 cos 𝜔)

(19)

Where:

𝛿 = 0.0691 −

0.399921 × cos(1 × 𝜂 × 𝑑) −

0.006758 × cos(2 × 𝜂 × 𝑑) −

0.002697 × cos(3 × 𝜂 × 𝑑) +

0.070257 × sin(1 × 𝜂 × 𝑑) +

(23)

22

0.000907 × sin(2 × 𝜂 × 𝑑) +

0.001480 × sin(3 × 𝜂 × 𝑑)

𝜂 =

2𝜋

366 𝑅

2

𝑅

2

= 1 + 0.033 cos(𝜂𝑑)

𝜔 = |12 − ℎ| ×

𝜋

12

And:

𝐸

_𝑇

=

solar irradiance maximum at time t (in Wm-2₎

𝐼

0

=

solar constant ~ 1367 (in Wm-2)

𝑑 =

day number (in d) 𝑅2

𝑅2

=

Relative difference (-)

𝛿 =

angle between sun and the earth surface at day d (in rad)

𝜔 =

angle between sun and earth surface at hour h (in rad)

ℎ =

hour of the day (in h)

𝜙 =

latitude (in rad)

7.6.2 Calculation of the theoretical solar irradiance over time and over depth

Eq. (19) only calculates the solar irradiance for one hour and only for the water surface, thus Eq. (11) and Eq. (19) were turned into a function (Max_I_0_fun) to calculate solar irradiance over time and depth. A function was also made with Eq. (12) for the Im model.

This function in the first loop calculates for every timestep on a specific day the solar irradiance at the water surface with Eq. (19). In the second loop the function calculates for a specific timestep the light intensity over depth with Eq. (11 & 12) and then in the third loop the light intensity over depth is calculated for all timesteps. The day number and Secchi depth were used as the input arguments.

(24)

23

7.6.3 Calculation of the turbidity

In the absence of hydrolab data the Kd may be derived from the Secchi depth, which is an easy

method to determine water transparency in cases where hydrolab data is not available. The Secchi depth is related to Kd according to Eq (20), which is based on the universal equation proposed by

Poole & Atkins (1929).

Kd

= (1.7/𝑆𝑒𝑐𝑐ℎ𝑖 𝑑𝑒𝑝𝑡ℎ)

(20)

7.6.4 Calculation of the effect of cloudiness

Field measurements of solar irradiance at the water surface were required to correct the maximum solar irradiance, which are not always available. In order to still model the effects of cloudiness, a cloud modification factor (CMF) was derived from observed weather data, which is the ratio of solar irradiance in cloudy and cloudless conditions based on the cloud coverage (in oktas). An okta is a measure of cloud coverage and ranges from 0 to 8. A cloudiness of 0 octa indicates cloudless conditions, while a cloudiness of 8 octa indicates an overcast sky (Bilbao et al., 2016; Alados et al, 2000). KNMI data from the Voorschoten and Schiphol weather stations were used and June to September in 2017 was chosen as the time period, since the summer months are when cyanobacterial blooms usually occur.

To calculate the CMF the theoretical solar irradiance was first calculated based on Eq. (19). The KNMI measured solar irradiance was then divided by the theoretical solar irradiance to derive the initial CMFs. One specific time was chosen to remove the effects of the sun angle. This was

accomplished by filtering the KNMI data in R studio with the group_by and filter functions. Grouping was performed by hour and filtering by cloudiness class and for 12:00 pm.

The filtered ratios were thereafter plotted against the cloudiness classes (0 -8). Power and

polynomial functions are both used to describe the effect of cloudiness (Bilbao et al, 2016; Alados et al., 2000). Ultimately, a power model was chosen, since there were no perceivable differences between the two.

The CMF was incorporated directly into the first loop of Max_I_0_fun as:

𝐶𝑀𝐹 = 1 − 𝐴

𝑐𝑚𝑓

× 𝑐𝑙𝑜𝑢𝑑

𝐵𝑐𝑚𝑓

(21)

𝐸

_𝑇𝐶

= 𝐸

_𝑇

× 𝐶𝑀𝐹

(22)

Where:

𝐸

𝑇𝐶

=

cloudiness corrected solar irradiance at time t (in Wm-2)

𝐶𝑀𝐹

= cloud modification factor (-)

𝐴

_𝑐𝑚𝑓 = intercept

𝐵

_𝑐𝑚𝑓 = slope

(25)

24

Cloud

= cloudiness (in octas)

7.6.5 Calculation of the delayed degradation

The last modification to the model function of the sensitivity analysis was the addition of the delayed degradation of H2O2. By keeping the concentration constant during a specified amount of time, the degradation can be started later to better simulate alternative H2O2 application methods. This resulted in four different model functions: Iz and Im without delayed H2O2 degradation and with delayed H2O2 degradation (Appendix F).

7.7. Phase 5: Comparison of scenarios for the sensitivity analysis

In the sensitivity analysis six scenarios were calculated and compared (Table 1). In every scenario the cyanobacteria mortality is visualized over time and over a depth of 6m. This value was chosen, because in the field data the maximum measured depth was integrated over 6m depth. The depth interval was set at 0.1m.

For the Iz model, which is used in this report to model mortality in stratified lakes, the first scenario presents field conditions of the Klinkenbergerplas as a reference. The subsequent scenarios present differing effectiveness of H2O2 in reducing the concentration cyanobacteria. The parameters used in the sensitivity to determine effectiveness are: concentration H2O2, delay in degradation H2O2, solar irradiance (Io), degree of transparency (Secchi depth), cloudiness and fraction cyanobacteria of the total phytoplankton population.

For the Im model, which is used in this report to model mortality in lakes with mixing, a different approach was taken and five different parameters were tested with three different values (Table 2). In scenario 1, the effect of delayed degradation of H2O2 is compared. In scenario 2, a day in May, June and August are compared. In scenario 3, cloudy versus cloudless days are compared. In scenario 4, three different Secchi depths and in scenario 5 three different mixing depths are compared.

Table 1:Sensitivity analysis scenarios for the Iz model

Scenario H2O2 Delay Io / cloud Secchi depth Fraction cyano Iz 1: 4.5 mg/l 6 hours Field conditions 0.6m 1

Iz 2: 4.5 mg/l none Max / 0 oktas 6m 1

Iz 3: 4.5 mg/l none Max / 0 oktas 0.5m 1

Iz 4: 4.5 mg/l none Max / 0 oktas 0.5m 0.5

Iz 5: 4.5 mg/l none 8 oktas 0.5m 0.5

(26)

25 Table 2:Sensitivity analysis scenarios for the Im model

8. Results

8.1. Phase 1: Data fitting

For the degradation of H2O2 an exponential model was used to estimate the decay rate (Fig.8). This was performed for every initial concentration of H2O2 and light intensity (Fig. 9). The fitted model coefficients were significant for every tested concentration and light intensity (Appendix D1). No clear trend was visible between the degradation of H2O2 and light intensity.

Figure 8: Change in concentration H2O2 over time for an initial concentration of 4mg/l and all lab light intensities.

Scenario H2O2 Delay Io/ cloud Secchi depth Fraction

cyano

Mixing Depth Im 1: 4.5 mg/l (0.1,1 & 6 hours) Max/ 0 octas 0.5m 1 6m

Im 2: 4.5 mg/l none Max/ 0 octas 0.5m 1 6m

Im 3: 4.5 mg/l none (0,4 & 8 octas) 0.5m 0.5 6m

Im 4: 4.5 mg/l none Max/ 0 octas (0.5,1 & 2m) 1 6m

(27)

26 Figure 9: Change in concentration H2O2 over time for initial concentrations of 1,2 4, 6 and 10 mg/l and all lab light

intensities

The Michaelis-Menten model (Fig. 10) was significant only for one of the two coefficients namely: the half-saturation constant, which determines the slope of the data fit. The other coefficient, the minimum decay rate of H2O2 was not significant (Appendix D2).

(28)

27 Figure 10: Michaelis-Menten model fitted on data of degradation rates H2O2 plotted over initial concentration H2O2

The P values of the intercept were all significant for the mortality. The slope was not significant for H2O2 concentrations of 0 mg/l and 1mg/l. For the former this was only the case for light intensities of 50, 100 and 150, while for the latter this was only for light intensities of 15, 50 and 100 µmol

photons m-2_s-1_{(Appendix D3).}

Interestingly, mortality in darkness appeared to increase more linearly with concentration compared to mortality in light (Fig. 11), which is also observed when plotting mortality rates in darkness against the initial concentration of H2O2 (Fig.12). The coefficients in this data fit were significant only for the slope (Appendix D4).

(29)

28 Figure 11: Change in photosynthetic yield for different initial H2O2 concentrations under A) 0, B ) 15, C) 50, D) 100 and E) 150 µmol photons m-2_s-1_light.

(30)

29 Figure 12: Mortality rates in darkness plotted against the initial concentration H2O2 and fitted with a linear model.

Mortality in light was subsequently fitted by using a PI curve (Fig. 13). The P values of the coefficients were not significant for 0 mg/l and 1mg/l and not significant for the slope of 6 mg/l (Appendix D5). It was decided to leave out 1 mg/l in order to improve the fit.

(31)

30 Figure 13: Mortality rates in light plotted against the initial light intensity and fitted with a PI curve

The fits for Al and Bl (Fig. 14) were improved by removing the data of 1 mg/l. Bl Eq. (7) was significant for both coefficients, while Al Eq. (8) remained insignificant for both coefficients (Appendix D6).

(32)

31 Figure 14: Coefficients Al (Eq.7 )and Bl (Eq.8 ) of the mortality M (Eq.6) plotted against initial H2O2 concentrations. For Bl data for 1mg/l was omitted and for Al 1 mg/l and 2mg/l was omitted.

8.2. Phase 2: Lab validation

Degradation of H2O2 appears to be greatest for 1 mg/liter and 2 mg/liter and decreases considerably with higher concentrations. This effect is also apparent in the fit of the model, which worsens with increasing concentration of H2O2 (Fig. 15).

(33)

32 Figure 15: Model Validation of H2O2 degradation by comparing measured lab data with predicted model data. Performed for the initial H2O2 concentrations of 1 mg/l (A), 2 mg/l (B), 4 mg/l (C), 6 mg/l (D) and 10 mg/l (E).

Valibration of the mortality was performed both in dark and in light conditions. As expected, in darkness mortality increased with increasing concentrations of H2O2 (Fig. 16). The modelled values moreover seemed to accurately follow the lab data.

(34)

33 Figure 16: Model validation of mortality in darkness by comparing measured lab data with predicted model data.

Performed for the initial H2O2 concentrations of 1 mg/l (A), 2 mg/l (B), 4 mg/l (C), 6 mg/l (D) and 10 mg/l (E).

In light conditions mortality in general appears to increase with light intensity. This was predicted relatively well by the model. However, this trend does not appear to hold true for all concentrations; for 1 mg/l a reverse effect is visible with Fv/Fm increasing (Figure 17). This effect can be perceived for the light intensities 15, 50 and 100 µmol photons m-2_s-1_{. At a light intensity of 150 µmol photons} m-2_s-1_{this effect seems to disappear (Appendix A1, A2, A3).}

(35)

34 Figure 17: Model validation of mortality with a light intensity of 50 µmol photons m-2_s-1_{by comparing measured lab data} with predicted model data. Performed for the initial H2O2 concentrations of 1 mg/l (A), 2 mg/l (B), 4 mg/l (C), 6 mg/l (D) and 10 mg/l (E).

8.3. Phase 3: Field validation

8.3.1. Validation H

2

O

2

The predicted degradation of H2O2 was validated with two treatments datasets: Klinkenbergerplas on the 15th_{of June 2017 and Oosterduinse meer on the 8}th_{of August 2018. For the treatment on the} 19th_{of June at the Oosterduinse meer not enough data points for H2O2 were available (Appendix D).} It was therefore decided to refrain from using this dataset in the validations of the degradation of H2O2 and mortality.

(36)

35 Figure 18: Validation degradation H2O2 Klinkenbergerplas on June 15th_{2017 by comparing measured field data with} predicted model data

The degradation of H2O2 in the Klinkenbergerplas did not follow the 4.5 mg/l target set by the model and increased in the first 8 hours of the treatment (Fig. 18). After treatment stopped and H2O2 started decreasing the model overestimated the actual degradation (Appendix E). The concentrations of H2O2 moreover appeared to not be homogenously mixed entirely over the measured depth. The highest concentrations were reached at a depth of 2.5m and this also remained the case in the 24 hours following the treatment.

Delay in degradation

(37)

36 Figure 19: Validation degradation H2O2 Oosterduinse meer on August 7th_{2018 by comparing measured lab data with} predicted model data

The validation for the Oosterduinse meer fared slightly better with the model appearing to follow the measured H2O2 degradation at a depth of 2.5m. However, at the other depths H2O2 fluctuated which could not be explained by the model (Fig. 19).

8.3.2. Validation mortality

For validation of the mortality the field measured concentrations of H2O2 were used, because in the field H2O2 was injected over multiple depths, leading to variable rates of mortality over depth. Target concentrations were determined before the treatment, however in practice these only serve as guidelines. Moreover, two light models were tested, Iz (Eq. 11 ) and Im (Eq. 12 ).

(38)

37 Figure 20: Validation mortality Klinkenbergerplas on June 15th_{2017 by comparing measured field data with predicted} model data

The Iz and Im models were both run for all measured depths in the field. For the Im model the mixing depth was derived from the hydrolab data (Fig. 4A). The Im model was run for all measured depths in order to test whether the mortality is better explained by the Iz or the Im model.

The Im model was expected to yield better results for the Klinkenbergerplas, since the data appeared to suggest some presence of mixing. This can be seen in the Fv/Fm, which decreases homogenously towards zero for all depths except 5m. The treatment did not appear to be very effective at a depth of 5m, since values of Fv/Fm appeared to fluctuate (Fig. 20). Mixing thus does appear to be of importance for this treatment, since at all depths the Iz model underestimates the mortality. Still, only the measurements at a depth of 2.5m really appeared to follow the Im model.

(39)

38 Figure 21: Validation mortality Oosterduinse meer on August 7th_{2018 by comparing measured field data with predicted} model data

The Oosterduinse meer was believed to be more stratified during the treatment, since wind still conditions occurred on that day (personal communication T. Piel, October 18, 2019). This also appeared to be represented in the hydrolab measurements, where only a shallow mixing depth of approximately 1.6m was perceivable. The mixing depth was derived from the hydrolab data (Fig. 6A).The effect of this was expected to be better explained by the Iz model, however for depths of 2.5m and 5m the model underpredicted while at the surface mortality was overpredicted (Fig. 21). The Im model underestimated the mortality for all depths, but overall did approach the actual mortality better.

(40)

39

8.4. Phase 5: Sensitivity analysis

8.4.1 Cloud modification factor (CMF)

Figure 22: Cloud modification factor.

The ratios appeared to generally decrease with increasing cloudiness, which was expected (Fig. 22). Interestingly, the relation appears to be roughly linear from 0 to 6 octas. The greater decrease after 6 octas could be explained by indirect light becoming the dominant source of light, due to increased cloud coverage. The power model was chosen over the polynomial due to the higher significance of the slope coefficient (Appendix D7).

8.4.2 Iz model

In the Iz model three parameters appear to be most influential to the effectiveness of H2O2. The turbidity and the concentration or fraction of cyanobacteria seem to be most important during the entire treatment, while the effect of delayed degradation of H2O2 becomes visible later in the day. In reference scenario 1 (Fig. 23), algae are killed down to a depth of around 3m, which is higher than any scenario except scenario 2 (Fig. 24). This difference in mortality gradually changes over the course of the treatment, which is explained by the delayed degradation in the reference scenario. This effect is most apparent 24 hours after the treatment, where in other scenarios mortality stagnates, because of the lower concentrations H2O2.

(41)

40 Figure 23: Scenario 1 Iz model

Figure 24: Scenario 2 Iz model

The impact of turbidity is most perceivable between scenarios 2 and 3 (Fig. 24 & Fig. 25), where in the former case all algae are deceased over the entire visualized depth, while in the latter case complete mortality only occurs down to a depth of 2m. Yet, a photosynthetic reduction of roughly 50% is still visible for the rest of the water column.

(42)

41 In scenario 4 (Fig. 26) the increased presence of eukaryotic species leads to a reduction in mortality throughout the water column. Full mortality now only occurs up to around 1.8m depth and the rest of the water column is only reduced by roughly 25%. The smallest perceivable differences are between scenario 4 and 5 (Fig. 26 & Fig. 27), where a reduction of solar irradiance only appears to slightly reduce full mortality down to a depth of 1.5m, but where the rest of the water column remains relatively unchanged.

In the last scenario (Fig. 28), only the upper 1.2m is free of algae, whereas the rest of the water column is unaffected.

(43)

42 Figure 28: Scenario 6 Iz model

8.4.3 Im model

In the results of the Im model the main differences are once again caused by the turbidity and degradation timing, but in addition the effect of the mixing depth is presented.

In scenario 1 (Fig. 29), the effect of a delayed H2O2 degradation is perceivable around five hours into the treatment. This was tested with a Secchi depth of 0.5 m, under clearer conditions this difference could be larger, due to the increased availability of light.

(44)

43 Figure 30: Scenario 2 Im model

Figure 31: Scenario 3 Im model

In scenario 2 (Fig. 30) and 3 (Fig. 31), no big differences are perceivable in mortality between days and between sunny and cloudy days. The relative influence of cloudiness seems to be even smaller in the Im model than in the Iz model. This suggests that solar irradiance at the surface might not be the most important factor and is in most cases not an issue.

(45)

44 Figure 32: Scenario 4 Im model

In scenario 4 (Fig. 32), the influence of transparency was tested. Large differences in effectivity can be observed with a shallower Secchi depth indicating more turbid conditions. At a Secchi depth of 2m, the effectivity moreover appears to reach a saturation point.

Figure 33: Scenario 5 Im model

In scenario 5 (Fig. 33), different mixing depths were compared. It was expected that a greater mixing depth would increase the effectiveness of H2O2. This was especially true for the upper 2m where light availability was still high, however light decreases over depth whether there is mixing present or not. The increased light availability introduced from mixing thus appeared to have diminishing returns on the mortality for depths greater than 2m. Another interesting observation that could be made is that at a depth of 2m, mixing depth and turbidity have a similar effect on the mortality.

(46)

45

9. Discussion

9.1 Lab data

The predicted degradation of H2O2 of the model was in most cases in line with the measured degradation H2O2 in the lab. It was observed that the degradation of H2O2 decreases with increasing initial concentration of H2O2. This could be explained by the fact that at lower concentrations relatively speaking more of the H2O2 is degraded during oxidation of organic material released by cyanobacterial cells. At higher concentrations organic material is oxidized more quickly, once all compounds are oxidized, enzymatic breakdown is the dominant process in degradation, which could explain why at higher concentrations H2O2 appears to degrade more linearly (Tao et al., 2009). The predicted mortality of algae similarly showed a strong correspondence with the measured data. The deviation of the 1 mg/l H2O2 results could be explained by hormesis, which occurs when

organisms and cells are exposed to moderates amounts of stress. This mechanism is believed to be an evolutionary adaptive strategy permitting organisms to survive occasional harsh conditions. It indicates that at a low concentration an otherwise toxic agent could become beneficial (Mattson, 2008).

There does however appear to be a specific light intensity range where this is most apparent, and it is furthermore unclear if this indicates that at lower concentrations 1 mg/l is ineffective. Using photosynthetic yield as a measure of mortality is limited in that way, since only the health of the photosystems is measured, which if damaged may still be repaired (Allahverdiyeva & Aro, 2012). A better way to determine mortality would be to perform cell staining, this was however not feasible in the timespan of this internship and no field data for this was available. Cell staining is performed with a flow cytometer and is based on the integrity of the cell. During a cell membrane rupture a fluorescent dye can enter and bind on the cell nucleonic acids thereby staining the cell (Mikula et al., 2012). Ruptured cells do not live for long, as such cell staining is a good method for determining mortality. Since concentrations of H2O2 drop below 1mg/l over the course of a treatment, more research into the actual mortality of cyanobacteria at low concentrations H2O2 is required.

The effect of light color was not included in the model but does appear to have an influence on the effectivity of H2O2 as well. Especially orange light related to the presence of high amounts of DOM is suggested to increase the effectiveness of H2O2 (Luimstra et al., 2018; Luimstra et al., 2019; Piel, unpublished).

9.2 Field data H2O2

The validation results of the H2O2 showed some deviations from the actual measured data. In the field H2O2 was not distributed homogenously, because of wind, movement of the boat and

horizontal transport, which could all have affected the distribution, measurements and degradation of H2O2. It was therefore difficult to find clear patterns.

For the Klinkenbergerplas degradation rates were underestimated even when considering the timing of degradation. It is possible that this is an effect of bloom density which was not modelled, since a faster degradation rate at first could point at large amounts of organic matter being oxidized. In the lab, measured concentrations may be lower than in the field, thus if more organic matter is available the degradation rate changes as a result.

Surprisingly, for the Oosterduinse meer the model appeared to predict the degradation better, but only for a depth of 2.5m. Though in this case the measured concentrations of cyanobacteria were

(47)

46 lower and eukaryotic phytoplankton were present in almost equal concentrations, which could have explained the differences in degradation over depth.

It must be noted that degradation of H2O2 is based on lab data of a single species of cyanobacteria. In the field there are many types of organisms and organic material that could have an impact on the degradation rate of H2O2.

9.3 Field data mortality cyanobacteria

The reason for the inaccurate validations of the mortality in both lakes (Fig. 17 & 18) could have multiple reasons. In the Klinkenbergerplas especially the mortality rate at the surface deviated, which could be explained by photoinhibition occurring at the surface, which is not considered in the model.

Whereas in the Oosterduinse meer mortality was mostly underestimated. The effect of eukaryotic phytoplankton was however approximated, since lab data for this was limited. Ideally lab

experiments with just eukaryotic phytoplankton or experiments with varying ratios of cyanobacteria and eukaryotic phytoplankton would have to be conducted to more accurately predict the

effectiveness of H2O2. There is also the daily variability of eukaryotic phytoplankton that may be considered, which could be an additional underlying process.

9.4 Sensitivity analysis

The results of the sensitivity analysis were generally in accordance with what was expected, since the parameters directly influencing the light intensity under water appeared to have the strongest effect on the effectivity of H2O2 .The effect of cloudiness was nevertheless not as strong as predicted. On the one hand, this could be because it only directly influences the light intensity at the water surface. On the other hand, cloudiness was only calculated based on KNMI data from the year 2017. A better approximation could be retrieved by using the complete data from 1951 to 2019 and for different times of day.

In addition, the effect of cloud formation and sun angle was not considered. Cirrus clouds for instance might even increase the available light intensity as opposed to cloudless conditions (Kazantzidis et al., 2011). This data was however not readily available and either deemed not important enough or too complex to include in the model.

Lastly, the assessment of the sensitivity analysis was performed graphically. A statistical or mathematical method could also be applied to improve the accuracy of the analysis and to better substantiate the smaller differences (Frey & Patil, 2002).

There are more ways the model could be improved especially regarding the application and spatial distribution of H2O2 over the treatment lakes. In case of the former, the way delayed degradation of H2O2 is implemented in the model, is not entirely accurate with how it occurs in the field, where concentrations steadily increase until the target concentration is reached. In case of the latter, not much spatial data is available except for measured concentrations H2O2. The model should perhaps also introduce the capability of algae to recover if photosynthetic is to be used as the primary indicator for vitality.

Modelling of hydrogen peroxide to combat toxic algae blooms

Internship report on the