We used a well-established water quality model (W2) to demonstrate the influence of water withdrawal strategies and light extinction on the growth of the cyanobacteria P. rubescens in Rappbode Reservoir, the largest drinking water reservoir in Germany. Although the occurrence (Ernst et al., 2009), physiology (Selmeczy et al., 2016) and the toxicity of P. rubescens (Viaggiu et al., 2004) have been well studied, little is known about active strategies to control blooms of this species. This is a major gap in current research as this highly specialized cyanophyte can proliferate in nutrient-poor water bodies, which often serve as drinking water sources. Indeed, existing research on cyanobacterial blooms focus largely on surface blooming and scum forming algae in eutrophic waters like Lake Taihu (Huang et al., 2020) or recent Lake Erie (Pirasteh et al., 2020). But research has only superficially addressed subsurface blooms like those typically formed by P. rubescens. This study advances our knowledge in this respect by adding three novelties: (i) providing an ecosystem model for the prediction of subsurface blooms, (ii) generate concrete reservoir operation strategies to mitigate subsurface blooms in Germany’s largest drinking water reservoir, and (iii) identify key environmental factors (light) that affect their occurrence. Needless to say, other environmental factors like nutrient supply and distribution are similarly important and require more research in future.
Note that P. rubescens has a low specific chlorophyll a content meaning that its biomass is not as low as suggested by the small chlorophyll values. Moreover, Rappbode is an oligotrophic water body and the formation of such a cyanobacteria population is outstanding and alerting the context of drinking water provisioning. Meanwhile, P. rubescens can contain high microcystin toxin quota, so only a small amount of algae can severely harm the human health and lake ecosystem. Survey data from German water bodies indicated the total microcystin content in P. rubescens ranges from 2000–5000 ug/g dry weight (Chorus and Welker, 2021). So the biomass of 4 ug/L (i.e. mean concentration in scenario R) corresponds to a total microcystin concentration of 1.44–3.6 ug/l, which is higher than the threshold for the drinking water sources from the WHO guideline (1 ug/l, see WHO (2004)). Considering its harmful effect on water quality, consequently, it is beneficial for the reservoir operators to suppress the growth of P. rubescens.
4.1. Selective water withdrawal and its influence on growth of P. rubescens
Selective water withdrawal is widely used in stratified reservoirs worldwide (Deng et al., 2011) for optimizing variables that collectively influence water quality (e.g., water temperature, dissolved oxygen concentration, turbidity) within the reservoir or for implementing natural downstream temperature regimes as a component of environmental flows (Olden and Naiman, 2010). Various types of infrastructure are adapted to achieve selective withdrawal, such as multi-level offtake towers, temperature-controlled curtains, floating outlets, pivoted pipes or stop-lot gates (Ren et al., 2020) and each infrastructure component has specific options and restrictions that influence its application. Outlet towers, for example, can only be used to withdraw water at specific depths while pivoted pipes can be freely moved in the vertical direction to take out water over a continuous depth range. Most previous studies of selective withdrawal have focused on temperature dynamics in the downstream river (Zheng et al., 2017; Weber et al., 2017) or within the reservoir (Mi et al., 2019; Çalışkan and Elçi, 2009). Although the same strategy is in principle also applicable for biogeochemical variables like algal biomass, nutrient concentrations and pathogens, such applications had been rarely used or modelled (Zhang et al., 2013; Feldbauer et al., 2020). Our study fills this gap and its practical value should help stakeholders optimize their management strategies and mitigate water quality problems arising from the occurrence of P. rubescens.
The location of peak P. rubescens biomass in the water column corresponds to a physiologically adjustable depth of neutral buoyancy, and may vary in different water bodies (Maltese et al., 2012) and over time due to changing light and nutrient gradients (Walsby et al., 2004). It is advantageous if the withdrawal facility can precisely follow the depth of the P. rubescens peak in order to maximize its potential for the algae removal. We therefore designed withdrawal depth by distance under the water surface, instead of the commonly absolute elevation. This is meaningful since the outtake depth follows the algal peak even in case of water level changes due to hypolimnetic withdrawal of raw water. Moreover, we believe that the selective withdrawal strategy, tested in this study, could remove other harmful or unwanted constituents (e.g., high concentrations of dissolved organic carbon or heavy metals) in reservoirs, as long as its occurrence is restricted to a narrow depth range and the offtake can directly flush the layer containing the constituents out of the water column (Rigosi and Rueda, 2012).
Our results also indicate that P. rubescens concentrations decreased following the increased water withdrawal volume, but the relative effectiveness in the algae removal per unit of water withdrawn, is getting weaker at high withdrawal discharge (see Fig. 4). For example, the decrease of P. rubescens concentration is much lower from scenario V2 to Vmax, than from scenario Vmin to V1. This results should be attributed to the changes of stratification intensity under different withdrawal scenarios. Here the intensity is represented by buoyancy frequency (N2) based on the density gradient in water column, which is calculated as (Read et al., 2011):
$${N}^{2}=\frac{g}{}\frac{d}{dz} \left(3\right)$$
where g is the the acceleration due to gravity, ρ is density and z is depth. The results clearly show that during summer metalimnetic N2 under scenario V2 onwards is higher than that under the first two scenarios (Vmin and V1, see Fig. S3). This indicates high rates of withdrawal increase the stratification intensity and provide a more stable density gradient, which is beneficial to P. rubescens growth. Also, high withdrawal can be disadvantageous with respect to the loss of water storage. For example, from scenario V1 to Vmax, the water level decreased by approximately 5 m for each 10 million m3 of water withdrawn (Fig. 3). From a stakeholders’ perspective, such a dramatic decrease in water level harms water security for drinking water supply, and could have negative influences on water quality due to increases in turbidity and nutrients (Zohary and Ostrovsky, 2011). Therefore, in most cases there should exist an optimal withdrawal volume for the specific research place (around 10 million m3 in our case), balancing positive and negative effects.
Withdrawing metalimnetic water at the beginning of P. rubescens growth more effectively reduced its concentration in the reservoir than later withdrawals (see Figs. 5 & 6). On the one hand, this makes sense as withdrawal is exactly at the time when growth rate is maximal, but on the other hand the accumulated amount of exported biomass should be lower than for a period with later withdrawal. We hypothesize that selective withdrawal – besides the direct effect of biomass export – also induces indirect effects on P. rubescens growth by changing the dynamics of soluble reactive phosphorus (SRP) concentration. The supply of phosphorus in the initial P. rubescens growth phase appeared to be partly recycled from diatom death. Our simulations indicated that diatoms took up a large proportion of the bioavailable phosphorus in spring before dying, with mineralization then recycling organic phosphorus to its bioavailable inorganic form, again. P. rubescens appeared to fill in a niche that exploited this recycled phosphorus. Scenarios with early withdrawal removed a substantial amount of mineralized phosphorus from the system, which was confirmed in our results: SRP concentration at depth 10–12 m, where P. rubescens was highest, was always lower in the Tearly scenarios than in the case of late withdrawal or the reference simulation (up to 2 µg L− 1 in some cases, see Fig. S4). Since phytoplankton in Rappbode Reservoir is P-limited (Wentzky et al., 2018), the reduction in phosphorus concentration in the Tearly scenarios leads to lower biomass of P. rubescens compared with the Tbloom and Tlate scenarios. What’s more, the T- scenarios showed that early withdrawal increases the diatom biomass in the surface layers (Fig. S5) which implies a stronger shading effect from diatoms on the P. rubescens residing in the metalimnion, leading to further deterioration of growth conditions for this species.
In the scenarios above, we separately checked the optimal withdrawal volume and timing to suppress the bloom of P. rubescens. To further clarify the conclusion, we coupled scenario V and scenario T together and established another 15 sub-scenarios (scenario VT) in which the optimal discharge (10 million m3) was evenly withdrawn at the metalimnion during 23 days (corresponding to the daily discharge of 5 m3 s− 1), starting from day 133 to 231 at one-week intervals. Here all the other settings are the same as before. Quite close to those from scenario T, the new results suggested that the population of P. rubescens was much lower under the strategy applied at its early growth stage than the other two stages (see Fig. S6), which further verified our previous conclusion and the robustness of the study (i.e. the optimal timing is independent of the withdrawal discharge).
Given its main purpose as a drinking water reservoir, the water storage in Rappbode Reservoir needs to be maintained above a certain volume for water security purposes (10 million m3 as absolute minimum according to the regulation of the reservoir authority). Since the water level decreases from spring to autumn in Rappbode Reservoir (Mi et al., 2019), the amount of water potentially available for withdrawal also decreases over the season. The selective withdrawal strategy should therefore remain within a “safe operating space” (see Fig. 8) that allows substantial control of Planktothrix biomass on the one hand, and guarantees adequate water storage on the other hand.
4.2. The role of light intensity in the algae growth in Rappbode Reservoir
Light intensity is a key factor influencing algal growth in aquatic systems, particularly for subsurface populations that are strongly affected by light extinction within the surface waters (Knapp et al., 2021). In the metalimnion, PAR from spring to autumn exceeded 0.5 mol m− 2 day− 1 and was up to 1.5 mol m− 2 day− 1 under clear-water conditions (scenario L0.35). In comparison, PAR ranged from 0.1 to 0.5 mol m− 2 day− 1 in the reference simulation (\({\epsilon }_{\text{b}}\)= 0.45 m− 1), and remained below 0.2 mol m− 2 day− 1 for \({\epsilon }_{\text{b}}\) ≥ 0.55 m−1 (see Fig. S7). In our algal growth model, saturating light intensities are 14 mol m−2 day−1 and 3.2 mol m−2 day−1 for diatoms and P. rubescens, respectively (see Mi et al., 2020a), indicating substantial light limitation, especially for diatoms, in the metalimnion. Interestingly, since diatoms undergo rapid development in spring (i.e., before the growth of P. rubescens) and monopolize nutrients in their biomass (Wentzky et al., 2019), scenario L0.35 allowed a longer metalimnetic persistence of diatoms (see Fig. S8), which in turn delayed the growth and occurrence of P. rubescens (see Fig. 7). Again, this points to the importance of indirect effects among phytoplankton groups and a proper representation of community dynamics and competition among algal groups is required to capture the dynamics at a species level.
Current climate forecasts indicate that due to global warming, air temperature for the study region may increase substantially in this century (Mi et al., 2020b), which will intensify stratification duration and stability potentially supporting the dominance of P. rubescens in lakes and reservoirs (Knapp et al., 2021). Our study suggests that decreasing light intensity in the metalimnion may offset the supportive effects of warming on the P. rubescens growth. A number of studies have reported increasing DOC concentrations (i.e. brownification, see Kritzberg et al. (2020)) in surface waters. This brownification, which is observable at continental scales (Monteith et al., 2007), is coming along with higher light extinction that is interfering with primary production by reducing light availability (Karlsson et al., 2009). Light conditions in the metalimnion are strongly affected by brownification and rising DOC concentrations can effectively shrink or even close the ecological niche for P. rubescens. Several drivers can lead to the brownification including changes in climate conditions and land cover, and reduction in atmospheric acid deposition. Faster flushing rates due to increasing precipitation, for example, restrain DOC sedimentation which leads to elevated DOC. To get a better quantitative understanding about the role of brownification for P. rubescens dynamics, further model applications should take these processes into account and deliver useful information for the reservoir operators.
4.3 Limitations and future work
The water quality model used in this research systematically elucidates the influence of two important driving factors (i.e., metalimnetic water withdrawal and background light conditions) on P. rubescens growth in Rappbode Reservoir. To extend the current study, our model can be used as a template to assess the possibility to also control other harmful algae by selective withdrawal, e.g., species associated with surface blooms. For achieving this goal, further model development may be required, e.g. to include other algal groups and new features like vertical migration of phytoplankton cells associated with buoyancy control.
Additionally, although metalimnetic water withdrawal can effectively suppress P. rubescens growth, the water that is withdrawn could potentially release high levels of cyanobacteria into downstream ecosystems. As indicated by Teurlincx et al. (2019), ecological functioning of inland waters can only be comprehensively understood if we take their connections with upstream catchments and downstream receiving waters into account. Accordingly, for future studies, it is recommended to combine upstream catchment models with lake models, as well as models of downstream ecosystems, in order to upscale the research from local to regional perspective and develop system-level management strategies.