Mowing Submerged Macrophytes in Shallow Lakes with Alternative Stable States: Battling the Good Guys?

Our modeling study shows that mowing can result in a temporal reduction of plant cover for a range of nutrient loadings and mowing intensities. These reductions of plant cover can reduce nuisance for up to several weeks, especially when the mowing intensity is fairly high and the external nutrient loading is low or moderate. Our model thus indicates that mowing can facilitate multiusage of shallow lake ecosystems. At lower intensities mowing also reduces the summer average plant cover, but this may be not sufficient to actually reduce nuisance as the remaining cover still exceeded the threshold level, which we fixed at 40%. Our model analyses indicate that it becomes more difficult to design a convenient mowing strategy when the external nutrient loading is high; the attraction of the alternative equilibrium is so strong that a rather small reduction in plant volume may be sufficient to trigger a shift to phytoplankton dominance. Interestingly, our results elucidate that a reduction of external nutrient loading alone is not an effective measure to drive back nuisance caused by aquatic plants (Fig. 4), which emphasizes the need for mowing. Because the risk of inducing a regime shift by mowing increases with external nutrient loading, the successfulness of mowing to reduce macrophyte nuisance goes hand in hand with the reduction of external nutrient loading. At what percentage of cover lake users perceive plants as a nuisance will vary between lakes and types of users. We took 40% because this number is frequently used in the gray literature, mostly in relation to growth of largemouth bass—a popular target species for sport fisheries (e.g. Gettys et al. 2014). However, we can hypothesize that when lake users already perceive nuisance at a lower plant cover (<40%) it will become increasingly difficult, or even impossible, to manage the vegetation successfully by harvesting while maintaining clear water. Vice versa, if local lake users would be more tolerant to the aquatic vegetation and perceive nuisance at higher percent cover (>40%), it will be more easy to reduce nuisance and maintain a clear-water ecosystem, especially when the external nutrient loading is not close to the critical nutrient loading level (Fig. 4). Hence, before designing a management scheme it is important to identify which stakeholders need to be served and at what percentage of plant cover they actually perceive plants as a nuisance.

Our model analyses indicate that the highest reductions of plant biomass can be achieved by mowing in mid-summer, while mowing in late summer appears to be least recommendable. The latter is not just because the peak of the holiday season (and thus recreational usage) is in mid-summer, but also because the risk of inducing a regime shift increases when mowing is conducted later in the growing season. In our model, mowing late in the growing season provides the aquatic plants with little opportunity to regain biomass before the growing season ends. This is in line with Engel (1990), who observed slow regrowth after mowing in July compared to mowing in June, and ascribed this to declining day length and water temperature. Consequently, in the following spring the macrophytes may start the competition with phytoplankton on their back foot, which eases a shift to phytoplankton dominance (Scheffer 2004, p. 280). Mowing too early in the growing season also bears a certain risk of triggering a regime shift, as our study showed, particularly when the external nutrient loading is high. We hypothesize that this is because the inter-specific competition with phytoplankton in early June is still rather strong, and setting back the submerged macrophytes favors phytoplankton growth. At the peak of the growing season, on the other hand, the intra-specific competition among macrophytes becomes more controlling, and mowing reliefs this intraspecific competition. Hence, the net growth rate of the macrophytes directly after mowing relates positively to mowing intensity (e.g. Fig. 2: the net growth rate after 30 and 60% mowing is 0.017 and 0.022 day⁻¹, respectively). This compensatory growth is not sufficient however to compensate for the entire loss of biomass, as plant cover does not recover to pre-harvesting levels (Fig. 2).

The effect of timing on the impact of mowing may be different in field situations, particularly when the macrophyte community comprises growth forms that—unlike e.g. Elodea canadensis—produce overwintering organs (Scheffer 2004, p. 279). Hence, in case of propagule forming macrophyte species such as several Potamogeton and Myriophyllum species, these propagules may have already been formed when harvesting takes place late in the growing season, wherefore the impact on the next growing season is much smaller. Harvesting earlier in the season would then be an effective way to reduce the potential for macrophyte plant growth in the succeeding year, as that would prevent the formation of propagules (Wade 1990). Interestingly, a reduction of plant volume in the succeeding year is generally considered as a positive result of harvesting (e.g. Dall’Armellina et al. 1996), while our modeling study hints that this strategy is not without risks when lakes have alternative stable states and the external nutrient loading is high.

In our model study we harvested fractions of the macrophytes to levels that may be unfeasible in real field situations. For example, there are practical reasons which frustrate harvesting large quantities of aquatic plants, as it is a labor-intensive and expensive activity. A simple calculation learns that for our modeled (circular) lake with a diameter of 1000 m, when receiving 1.2 mg P m⁻² day⁻¹, a harvesting intensity of 80% implies removing more than 650 tons of fresh biomass in just a short time span, assuming a fresh-weight:dry-weight ratio of 10 (e.g. Boiché et al. 2011; Dorenbosch and Bakker 2011; Online Resource 3). Secondly, local laws and regulations, such as the Dutch flora and fauna law, may impose restrictions on harvesting intensity and timing. Plants provide habitat and food for many species and it has been reported that significant amounts of fish and macroinvertebrates are removed along with the plants during harvesting (Engel 1990), which may include protected species. Furthermore, removing large quantities of plants may conflict with the protection of waterbirds that feed on the plants or the fauna living in macrophyte beds. A third reason is that in a field situation it will always be difficult to estimate the amount of aquatic plants that should be present to safeguard a clear water state, forcing lake managers to take a conservative approach when designing their plans. Hence, even though a submerged plant cover as low as 20% may coincide with good water quality (e.g. Portielje and Van der Molen 1998; Yanran et al. 2012), Hilt et al. (2006) advise to take 50% vegetation cover as a rule of thumb, and suggest that remaining stands after harvesting should still cover 50% of the lake. Also the Dutch authorities advise to remove maximally 50% of the plant cover, and even suggests to mow only 10% in case of native plant species (Rijkswaterstaat 2012). In our study we used Elodea sps. as model macrophytes, which are invasive in Europe. Our results show that part of the macrophytes should be retained under mowing management to prevent phytoplankton blooms under more eutrophic conditions. Implicitly this suggests that non-native macrophytes may be able to fulfill some of the ecosystem functions of native submerged macrophytes, in this case maintaining water clarity (Carpenter and Lodge 1986). This is in line with recent findings that non-native macrophytes may fulfill ecosystem functions similarly to their native counterparts and that their effectiveness depends rather on species traits than their origin (Grutters et al. 2015, 2016). Hence in management, complete removal of non-native macrophytes may be counterproductive for the ecosystem, if there are no native macrophytes to fill the empty place (Hussner et al. 2016).

From our analyses it appears that harvesting 10% of the standing crop has only a marginal effect on reducing nuisance. This situation may change however when it is possible and desirable to spatially divide functions over the lake area. By harvesting in such a way that only certain patches are cleared, it may become possible to reduce nuisance locally e.g. in a zone designated for swimming or a channel for navigation. The model we used (PCLake) is not spatially explicit and is therefore not suited to evaluate the effect of a local disturbance by harvesting, as it is intended to provide a general indication of the harvesting pressure the lake can withstand. There is only little known about the effect of spatial heterogeneity on alternative stable states in shallow lakes. Theoretical studies suggest that the potential of local disturbances to instigate an ecosystem-wide regime shift increases with interconnectedness (dispersion) within the system (Van Nes and Scheffer 2005), and decreases with spatial heterogeneity (Van de Leemput et al. 2015). These studies thus suggest that alternative stable states are unlikely to persist side by side in lakes which are very homogenous. This means that local mowing becomes risky as over-harvesting has catastrophic consequences for the entire lake, albeit the regime shift may be gradual (Bel et al. 2012; Van de Leemput et al. 2015). When lakes do exhibit spatial heterogeneity e.g. in terms of depth, fetch or sediment composition, the response to a local perturbation becomes much more difficult to predict (Van de Leemput et al. 2015), but this heterogeneity can potentially lead to coexistence of contrasting states. The latter would create opportunities for localized harvesting practices. A follow up step is to couple the ecological modules of PCLake to 2D-hydrodynamic models to analyze harvesting in a spatial hydrodynamic context. This development is still in its infancy however (e.g. Van Gerven et al. 2015).

Generally, not all cut plant biomass is removed from the lake due to inefficiency of the harvesting equipment (Hussner et al. 2016). The fragments that are not collected start to decompose in the water column, thereby releasing nutrients and contributing to the depletion of oxygen which in turn can stimulate internal nutrient loading from the sediment (Hilt et al. 2006). Additionally, cutting machinery may cause resuspension of sediments, which may reduce transparency and stimulate nutrient recycling. These side-effects of mowing are expected to be detrimental to ecosystem functioning (Rijkswaterstaat 2012), but it is difficult to quantify their true importance in the field. For simplicity reasons, we did not consider the effect of collateral disturbance in our main analyses. Yet, we did briefly look into their relative importance (presented as Online Resource 2), which revealed that, for the modeled circumstances and assumptions, the effect of collateral damage is marginal. This finding is in line with Carpenter and Gasith (1978) who reported short lived or insignificant effects on the littoral environment after clearing a 0.2 ha patch. Only when a regime shift has already been initiated, our model shows that the collateral effects of mowing stimulate the upheaval (Online Resource 2). However, we did not consider all potential side effects of mowing invasive aquatic macrophytes. For example, a factor we did not consider in this study is that many nuisance species (including Elodea spp.) spread by vegetative fragmentation (Hilt et al. 2006; Redekop et al. 2016). Mowing can stimulate dispersal of non-native nuisance species when fragments are produced that easily ride with the flow and settle at new places (Abernethy et al. 1996; Zehnsdorf et al. 2015). Especially when surrounding lakes or waterways are still free of these exotics, the vegetative dispersal capacity of the nuisance species that is being managed should be taken into consideration (Zehnsdorf et al. 2015). Recently, Hussner et al. (2016) reviewed how management aimed at the reduction or eradication of invasive aquatic plants can impact other (native) species present in the ecosystem. Interestingly, these effects can be both positive and negative. For example, Dawson et al. (1991) reported a case where 30 macroinvertebrate individuals were removed per gram dry weight of cut aquatic plants, while Bickel and Closs (2009) showed that total invertebrate biomass and abundance was significantly higher in the areas where mowing took place compared to the untreated macrophyte beds. Moreover, while Engel (1990) reported that up to 450 fish were removed per 100 kg fresh weight of cut aquatic plants, the potential for improving growth and size structure of fishes by reducing macrophyte density has long been recognized (e.g. Wiley et al. 1984; Olson et al. 1998). Furthermore, vegetation is a major food source for many waterfowl species aquatic and it is known that herbivorous birds such as coots (Fulica) can have a large impact on vegetation density (Van Altena et al. 2016). Interestingly, this trophic interaction may give rise to an interaction effect between mowing and herbivory. Hence, if a large quantity of vegetation is removed by means of mechanical mowing, this may either cause waterfowl to leave, but it may also cause birds to put extra pressure on the remaining vegetation, potentially triggering a critical regime shift to the turbid state (Van Altena et al. 2016).

Because there are nutrients stored in plant tissue, as well as in material attached to the plant surface such as periphyton and calcite incrustations, the removal of submerged macrophytes may help to remediate the detrimental effects of eutrophication, both in the lake where the plants are removed from, and in downstream aquatic ecosystems (Carpenter and Adams 1977). Our modeling scenarios indicate that the highest amount of P is extracted from the system when both the external P loading and the mowing intensity are high. The relative removal however, which tells more about the capacity to actually prevent further enrichment of the system via harvesting, increases with decreasing nutrient loading, maximally reaching 58% in our analyses. While it should be noted that periphyton and calcite incrustations are not explicitly modeled by PCLake, we find these numbers to be grossly in line with estimations presented in the literature. For example, for a eutrophic lake with 30% plant cover, Carpenter and Adams (1977) estimated that a relative removal of 37% of the P loading could be established if all plants would be harvested. Conyers and Cooke (1983) reported that a relative removal of 44% could be reached in a mesotrophic lake with 43% plant cover. Moreover, Wile (1978) presented a case where harvesting operations resulted in the removal of 560 kg P, and estimated the relative removal to be 47%. It is important to realize however that these numbers refer to the relative removal of P from the system as a whole, and not solely from the water column (Burton et al. 1978). Although many rooted macrophytes species are well capable of assimilating nutrients directly from the water column through their shoots (Madsen and Cedergreen 2002; Angelstein and Schubert 2008; Christiansen et al. 2016), at least part of their nutrients may be obtained from the sediment, especially in systems where large amounts of P are available in the sediment. As a consequence, removal of plant biomass does not axiomatically offset the external loading of P into the water column, and hence the incoming nutrients may perpetuate eutrophication. We postulate that the effect of harvesting rooted macrophytes on ecosystem functioning is highest when the macrophytes take up most of their nutrient directly from the water column. Furthermore, while harvesting alone may not be able to completely offset the incoming nutrient fluxes (Burton et al. 1978), we argue that the usefulness may be enhanced by the joint application of complementary management measures, such the application of phosphorus adsorbing natural soil and modified clay.

Harvested plant tissue can potentially serve as a source of nutrients, instead of only being waste material. The use of aquatic plant biomass to fertilize agriculture fields is an old practice (Roger and Watanabe 1984), which is still carried out in many parts of mainly the developing world. Recently harvesting aquatic plant biomass has been put forward as a way to close the P cycle (Quilliam et al. 2015). Although excessive growth of macrophytes indicates a local surplus of nutrients, P is a scarce element in many places, leading to phosphate starvation in crops, and global phosphate sources are declining rapidly (Cordell et al. 2009; Childers et al. 2011). The recovery of valuable P thus has the potential to increase the viability of harvesting as a management measure, which is otherwise a costly procedure (Hilt et al. 2006). Currently there is no agreement on how to maximize P uptake and removal by macrophytes (Quilliam et al. 2015). Our model results suggest that it is beneficial for lake managers to reduce the external nutrient loading as much as possible, as that will reduce the possibility of harvesting triggering an unwanted regime shift to a state without macrophytes, and increase the relative removal of P.

Model analysis of the long-term effects of harvesting suggests that harvesting can potentially be used to prevent nutrient over-enrichment by increasing the resilience of the system to external loading, that is, by increasing the CNL_eu (Fig. 7). It is important to note however that in this analysis harvesting was executed every year, and that we started off with a clear and oligotrophic lake—in the domain where no alternative state is apparent. Because of the latter, almost all macrophytes can be removed at the start of the analysis without risking a shift to the alternative state, as there simply is none. In turn, the removal of macrophytes prevents the accumulation of nutrients in the system, postponing the formation of an alternative equilibrium and hence increasing the CNL_eu. This implies that the history of the lake is an important factor to consider when designing a mowing strategy. If nutrients have been able to accumulate in the lake prior to the mowing activities, as in our first analyses, the resilience of the lake to perturbations such as mowing may have already decreased and fairly small fractions of macrophyte removal may be enough to instigate a regime shift (see Online Resource 4 for an illustrative example). Thus, based on the long term mowing scenarios we argue two points. The first is that phytoremediation can be a worthwhile measure to prevent a lake from becoming eutrophic when it is still oligotrophic and its capacity to withstand perturbations is still high. The second is that it is much more difficult to use phytoremediation to impoverish a lake when it is already eutrophic, even though the absolute removal of nutrients is high, because the capacity to withstand perturbation is much reduced. As many vegetated lakes in NW Europe have only recently recovered from the turbid state, and their sediments are likely to be saturated with nutrients, mowing schemes should be designed with great care.

An important question is how the results of this theoretical exercise should be interpreted by managers and can be useful in contemporary ecosystem management. Our point of departure is that every water system is unique (n = 1), but that there are general mechanisms that are key to the ecological functioning of every lake. PCLake has been developed to include the most important biotic and abiotic processes and lake characteristics (Janse et al. 2008). Moreover, to strive for generality, the model has been calibrated with data from >40 lakes with the aim to get the best overall fit (Janse et al. 2010). Hence, PCLake provides a coherent framework to investigate the effect of mowing within an ecosystem context with alternative stable states, allowing us to focus on important aspects of mowing, such as the intensity and timing, while keeping other factors constant. An important purpose of such analysis is to provide scientists and managers with working hypothesis about the way ecosystems function, and to contribute to the development of theory. The insights that are obtained by simulations cannot easily be derived from any other type of study, as the analyses would be too costly or unethical to do in natural systems. As such, these insights complement the insights obtained by alternative approaches, such as lab experiments and field observations (Peck 2004; Scheffer 2004, p. 313). PCLake is one of the very few integrated ecosystem models available for this kind of simulations (Janssen et al. 2015)

On one hand, we argue that the insights obtained by our simulations are widely applicable, as the model is built up of many general prevailing processes and principles. For example, although Elodea is used as a model species, we expect that, at least in qualitative terms, the response of other yet similar submerged angiosperms, such as Lagarosiphon major or Egeria densa, will be comparable. Moreover, lake characteristics have been chosen such that the model describes a hypothetical lake that is representative for many small and shallow lakes in the temperate zone. Interestingly, PCLake has even shown to be useful outside the temperate zone (e.g. Mellios et al. 2015; Kong et al. 2016).

On the other hand we acknowledge that the results of PCLake are highly dependent on the lake characteristics modeled. For example, Janse et al. (2008) showed that the resilience of shallow lakes decreases with increasing depth and fetch, implying that in larger and deeper lakes mowing can more readily result in a regime shift to the phytoplankton-dominated turbid state. Also the choice for a specific threshold level where lake users perceive plants as nuisance has implications for our results. When for a given case study, these controlling factors deviate too much from the settings used in this study, the calculations presented here should be redone for the new setting. Please note that such limitations apply to any type of experiment focusing on few independent variables.

Of course, it is conceivable that the ecology of certain lakes may differ fundamentally from the system currently portrayed by PCLake. For example, Blindow et al. (2014) distinguish between a charophyte-dominated clear water state and an angiosperm-dominated clear water state, and report on notable differences in the strengths of the exerted positive feedback loops with water clarity. Effects may be even more profound when a certain process has a strong effect on the functioning of a specific lake, but is not covered by the model. For example, grass carp (Ctenopharyngodon idella) has a strong trophic interaction with aquatic plants (Hussner et al. 2016), however this fish species is currently not included in PCLake. In all cases, a customized PCLake study, whereby the model is adapted, calibrated and validated for a specific case, will provide the most accurate predictions which can be readily employed in ecosystem management (e.g. Witteveen+Bos 2010; Nielsen et al. 2014; Trolle et al. 2014; Kong et al. 2016). The present study provides a clear example of how to set up a model analysis with PCLake to evaluate the effect of mowing on shallow lake ecosystem functioning.