Is the Cyanobacterial Bloom Composition Shifting Due to Climate Forcing or Nutrient Changes? Example of a Shallow Eutrophic Reservoir

Cyanobacterial blooms in eutrophic freshwater is a global threat to the functioning of ecosystems, human health and the economy. Parties responsible for the ecosystems and human health increasingly demand reliable predictions of cyanobacterial development to support necessary decisions. Long-term data series help with identifying environmental drivers of cyanobacterial developments in the context of climatic and anthropogenic pressure. Here, we analyzed 13 years of eutrophication and climatic data of a shallow temperate reservoir showing a high interannual variability of cyanobacterial development and composition, which is a less occurring and/or less described phenomenon compared to recurrant monospecific blooms. While between 2007–2012 Planktothrix agardhii dominated the cyanobacterial community, it shifted towards Microcystis sp. and then Dolichospermum sp. afterwards (2013–2019). The shift to Microcystis sp. dominance was mainly influenced by generally calmer and warmer conditions. The later shift to Dolichospermum sp. was driven by droughts influencing, amongst others, the N-load, as P remained unchanged over the time period. Both, climatic pressure and N-limitation contributed to the high variability of cyanobacterial blooms and may lead to a new equilibrium. The further reduction of P-load in parallel to the decreasing N-load is important to suppress cyanobacterial blooms and ameliorate ecosystem health.


Introduction
Recurrent and persistent mass developments (blooms) of cyanobacteria are one of the main outcomes of eutrophication of freshwater ecosystems, that is the natural increase in organic matter production and accumulation in an aquatic ecosystem, accelerated during the last decades by human activities [1][2][3][4]. Natural control of cyanobacterial blooms, e.g., by zooplankton, is limited by their low palatability owed to long filaments or colony formation and their low nutritional value, besides their toxicity [5]. Cyanobacterial dominance continuously dominates for example Lake Taihu, China [22,23] or the Aguieira reservoir in Portugal [24], while Planktothrix agardhii perennially dominates shallow eutrophic water bodies in lowland areas of the Netherlands and Northern Germany [25,26]. Compared to large monospecific blooms, interannual variation amongst cyanobacterial dominance needs better understanding as it can be a good indicator for changes in climatic and anthropogenic pressure. Interannual variations may indicate a shift towards a new state of the composition of the bloom.
The investigated shallow reservoir has received its eutrophication with the incoming rivers during the past decades. It shows a high variability of cyanobacterial development between years, for which the driving factors are not yet identified. Aims were therefore 1) to characterize the seasonal dynamic of the nutrient pattern as well as the phytoplankton and zooplankton succession over a one-year period, and 2) to identify the driving factors of cyanobacterial development over 13 years of monitoring, testing the hypothesis that nutrients, phyto-and zooplankton in the reservoir differ along the transect from entrance to outlet (H1), and that even in this short time series, climatic forcing changes the dynamics of cyanobacterial blooms (H2).

Nutrients' Concentration
Nutrients' concentrations were very similar at the entrance, in the middle and lower basin of the reservoir during the 2018-2019 seasonal cycle ( Figure 1). No significant difference was observed between the three stations in terms of nutrient concentrations but nitrogen and phosphorus concentrations revealed different patterns of seasonal variability. Total dissolved nitrogen and nitrate-nitrite were highly correlated (Spearman correlations, r S > 0.92, p < 0.001) and reached maximum concentrations in winter. Particulate N was low, but still corresponded to the particulate P concentrations during that period. Using the theoretical composition of the phytoplankton with the Redfield ratio and the ratios given by Reynolds (2006), we find that 150 µg of P corresponds to 1.05 mg N/L (mass ratio of 47:7:1 for C:N:P) and are therefore consistent with particulate N concentrations of 1.25 mg N/L. During periods of high concentrations in N, dissolved nitrogen reached almost 100% of the total N from January to May, as particulate N remained below LOQ for several sampling dates. During that period, nitrate and nitrite accounted for 65% of the total dissolved nitrogen.
Particulate phosphorus, total dissolved phosphorus and phosphate maximum concentrations were recorded during summer, and both particulate P and TDP were correlated (Spearman correlations, r S = 0.65, p < 0.001). Phosphorus concentrations of all fractions were very low from January to May, but reached almost 180 µg L −1 for particulate P and 125 µg L −1 for TDP at the end of summer, in September 2018. From June to December, total phosphorus was mainly composed of particulate phosphorus for two thirds and TDP for one third. The dissolved fraction accounted on average for 49 ± 24%, 40 ± 10% and 40 ± 13% at the entrance, middle and in the lower basin of the reservoir, respectively. Among this dissolved fraction, more than half was composed of phosphate (mean = 53 ± 10%, 59 ± 23% and 53 ± 12% at the entrance, middle and in the lower basin of the reservoir, respectively). N/P Redfield ratios were low at the entrance of the reservoir (<10 with mean = 6 ± 4), while there were closer to 16 in the middle (>10 with mean = 12.6 ± 1.6) and the lower basin (>10 with mean = 14.9 ± 3.5; Table 1).  Table 1. N/P Redfield ratios of the particulate matter at the entrance, middle and in the lower basin of the reservoir, respectively. N/P Redfield ratios were calculated only for the July-November period, as during the rest of the year, there was no particulate nitrogen and/or phosphorus concentrations were below limit of quantification.

Phytoplankton
The relative proportions of the different groups of phytoplankton and the total abundance of cyanobacteria were very similar at the entrance, middle and in the lower basin, showing no significant difference (Figure 2A). The relative contribution of phytoplankton is calculated from cell number, biovolume being unfortunately unknown, as counting was performed at the genus level. Cyanobacteria dominated planktonic community from June to December 2018 and from June to at least August 2019. There were no cyanobacteria in winter and spring, and their concentrations reached 1,500,000 cells mL −1 in late summer (September 2018). A second bloom of cyanobacteria was observed during autumn 2018 in the middle and the lower basin, reaching 600,000 and 1,200,000 cells mL −1 , respectively, in November. Cyanobacterial density was one order of magnitude lower during summer 2019. Phytoplankton eukaryotic community ranged from 200 cells mL −1 in winter to 72,000 cells mL −1 in early summer (July 2018). Chlorophytes were the most abundant in this community almost all yearlong, except in spring (March-April) during which diatoms or in february, where euglenophytes were dominating. In January, there was a co-domnance of chlorophytes with Chrysophytes at the entrance but with diatoms in the middle and lower basin. No statistical differences were observed between abundances of the three stations, except for Polyarthra (rotifer), copepods and Daphnia, for which the entrance present a lower abundance than the lower basin (p < 0.05).

Zooplankton
Zooplanktonic community abundance and composition was different between the entrance of the reservoir and middle-lower basin, apart from a common pattern of higher density during summers than winter ( Figure 2B). Abundances of the rotifer Polyathra were significantly higher in the lower basin compared with the entrance (p < 0.05). The Cladocera Daphnia and copepods had also higher abundances in both the middle and lower basins compared with the entrance (p < 0.05). At the seasonal scale, the rotifer species Keratella strongly dominated the micro-zooplankton community during both summers at the entrance of the reservoir, reaching 6000 and 12,000 individuals L −1 in July 2018 and August 2019, respectively. In the middle and in the lower basin, Pompholyx and Polyarthra species accompanied Keratella, reaching a total maximum concentration of 3000 and 6000 individuals L −1 in July 2018 and August 2019, respectively. Concerning meso-zooplankton, the Cladoceran Bosmina genus dominated the community in the entire reservoir, together with copepods in a smaller proportion ( Figure 2C). The total concentration of meso-zooplankton reached 1200 and 1500 individuals L -1 in the middle and the lower basin during summer 2018, while they were 4 to 5 times lower at the entrance of the reservoir (300 individuals L −1 ). During summer 2019, Daphnia developed in the middle and lower basin, dominating at 51% the community in July 2019 in the lower basin. Concentration of Nauplii was quite homogenous in the whole reservoir, ranging between few individuals in winter to 800 individuals L −1 in summer.  Table 2). GAMs were applied successfully to temperature, nitrate concentrations, residence time and flow showing their significant changes over the tested time period (Figure 3, Figure S1 and Table 3). Even during this relatively short time period with respect to long term data, a slight increase of temperature was observed ( Figure 3A, Figure 4A) During summers 2010 and 2013-2019, either warm temperature (≥20 days above 20 • C) or high light intensity (≥10 days above 2800 J m −2 ) or dry months (less than 1.5 million m 3 water entering into the reservoir) were recorded ( Table 2). Before 2013 (except 2009), a strong and frequent wind was measured in the summers, with at least 19 days of wind with an average daily speed greater than 4 m s −1 ( Figure 4E, Table 2). After 2013, only 2019 was classified as a windy summer. GAM could not be adjusted on wind due to the high daily variability and explained less than 4% of the deviance (not shown).
The biggest differences between the years of the study concerned the amount of water entering the reservoir during the hydrological year, ranging from 19 to 155 million m 3 per year (Table 2). 2012, 2017 and 2019 were the driest hydrological years, having received less than 40 million m 3 . This increased the residence time and lowered the level, and water outflow stopped from July onwards ( Figure 3B,C, Figure 4C Table 2). Table 2. Hydrological and meteorological characterisation of summer periods between 2007 and 2019. "Annual river inflow" was measured between November of the previous year and October, while "Summer river inflow" was measured between June and September. Dates of downstream flow stop (out of the lake) are indicated and depends on the water level (overflow). The date of the autumn increase in river inflow is also indicated in the last column.   Table 3. Summary of the GAM results performed on the abiotic parameters, with their smoothing functions. Plots are shown in Figure S1. Adj., adjusted.

GAMs Model
Smoothing Functions

Nutrients
A strong seasonal pattern was observed for nitrates ( Figures 3E and 4F), with maximum concentrations in winter (11.5 mg N-NO 3 − L −1 ) and minimum ones in summer. Owing to the low data points (n = 283), GAM was not able to discouple the seasonal pattern from the interannual one, and we smooth only one independent variable, the time ( Figure 3D). Nitrate concentrations slightly decreased over the studied period, and since 2009 concentrations were at the LOD at the end of summer periods, except in 2014 ( Figures 3D and 4F). Contrastingly, no seasonal pattern nor long-term tendency was observed for total phosphorus for the 2007-2019 period ( Figure 4G). Total phosphorus concentration remained high with on average 100 ± 67 µg P-PO 4 3− L −1 over the period studied.

Cyanobacterial Blooms
The cyanobacterial community varied strongly interannually, even including years without blooms, 2010, 2014 and 2015 ( Figure 5). Planktothrix agardhii was the most abundant species, dominating 6 of the 13 monitored years and reaching a biomass of at least 35-55 mm 3

Coupling Blooming Species with Environmental Parameters
To link the cyanobacteria species composition depending on time and environmental parameters, a canonical correspondence analysis has been performed on 195 sampling dates performed during summer. The final CCA included four environmental variables and time, which explain 10.2% of the total variability in the species composition ( Figure 6). The CCA discriminated the 12 cyanobacteria species of which 6 formed a bloom. The two first axes of the analysis presented here represent 7.5% of the total variability in species blooming (total inertia) and 73.9% of the constrained inertia. The first axis correlated with time (r = 0.92), hence when samples are grouped by sampling dates, old dates appear on the left while recent ones are on the right (Figure 6a). The first axis correlated moreover to air temperature (r = 0.35), residence time (r = 0.3) and light (r = 0.25) (all towards the positive side). The first axis is also explained by Planktothrix (23.7%) and Gomphosphaeria (20.3%) on the negative side and Dolichospermum (14.2%), Aphanothece (13.9%), Lemmermaniella (8.3%) and Aphanocapsa (5.6%) on the positive side.
The second axis negatively correlates with air temperature (r = −0.73), residence time (r = −0.50) and light (r = −0.29), and positively with flow (r = 0.31) ( Figure 6). Sampling dates with low flows thus coincide with high temperature, residence time and light. Air temperature and residence time thus contribute to both axes. The second axis is explained by Microcystis (28.2%) and Aphanizomenon (24.3%) on the negative side (bottom part − low flow), and by Planktothrix (18.2%), Lemmermaniella (11.7%), Coelomoron (7.0%) and Limnothrix (3.5%) on the positive side (upper part-high flow). All blooming species thus contributed at least to one of the two axes, while Planktothrix contributes negatively to both of them, in opposition with air temperature and residence time.
It should be noticed that nutrients were not included in the CCA analyses because they were measured less frequently and not necessarily at the same times than the phytoplankton community, moreover P concentrations did not change during the period of the data series. Dolichospermum sp. however seems to benefit below concentrations of 2 mg L −1 of nitrogen (Figure 6c): blooms of Dolichospermum were indeed only observed at very low nitrate concentrations.

Discussion
A low spatial variability but a high interannual variability in cyanobacteria biomass and composition have been revealed by this study. The seasonal cycle characterized this shallow reservoir as relatively homogeneous, with a similar evolution of nutrient concentrations and phytoplankton abundances, but different patterns for zooplankton along the increasing distance and depth from the entrance to the lower basin. These findings reject our first hypothesis assuming differeces concerning most parameters, but it could be accepted for the zooplankton dynamics after further investigations.
Maximum concentrations of phosphorus and nitrogen were both high, but with strong opposite seasonal patterns: the highest phosphorus concentrations occurred in summer when nitrogen was the lowest. Phosphorus concentrations of the lake were 3 times higher in the reservoir in summer 2018 compared to stations measured in headwaters of the incoming river Yvel [27]. As both nutrients usually enter during the wet winter months [28], their opposing seasonal pattern could indicate either enhanced uptake or denitrification for N, whereas the P concentrations exceeding those of the incoming water could have been released from the sediments during summer, as known from other lakes [20,29,30].
In summer, the Redfield N/P ratios in the total fraction were always below 20, confirming a deficiency of bioavailable N in the water column during this period [31]. This result is in line with an analyse of 369 German lakes concluding that N limitation seems to predominate during summer in shallow polymictic lakes [29].
Phytoplankton and zooplankton succession and densities were similar to many eutrophic water bodies. Total abundances of meso-zooplankton appeared high, but remained in agreement with other shallow eutrophic lakes [32][33][34]. Small zooplanktonic taxa seemed to dominate at the detriment of larger cladoceans, e.g., Daphnia sp., as known from eutrophic temperate lakes of both Europe [34] and the USA [35]. As sampling size was low in this study, more research focusing on zooplankton dynamics in that reservoir would be beneficial.
An interesting element in the seasonal cycle was the interannual variability between the two summers: while the bloom of cyanobacteria reached an exceptional high value of 1.5 million cells mL −1 in 2018, exceeding the limit allowing bathing by a factor of 15 in France and other countries [36], it was one order of magnitude inferior in 2019. At the same time, in the lower basin the zooplanktonic community switched from a dominance of Polyarthra and Bosmina during 2018 to a dominance of Keratella, Pompholyx and Daphnia in 2019, which could have been provoked by decreasing cyanobacteria in the phytoplankton community. Some zooplankton taxa such as the raptorial rotifer Polyarthra, or to a lesser extent the cladocerans Bosmina, are highly selective feeders avoiding cyanobacteria compared to filter feeders like Daphnia that thrive better in the absence of cyanobacteria [37][38][39][40]. A zooplankton community can also be top-down controlled by zooplanktivorous fish preferring large zooplankton such as Daphnia [41]. Despite the Lac-au-Duc reservoir being a frequented fishing site, the lack of published data on the fish compartment does not allow us to discuss their potential role in structuring the planktonic community. warmer summers connected to an increase of water residence time, a slight decrease of wind and a decrease of N sources, which in total confirms our second hypothesis (climate forces as main drivers). Before 2013, with the exception of 2009, summers were indeed characterized by more windy days. Blooms of Planktothrix agardhii dominated during that first period, as this species tolerates low average insolation in turbid waters of polymictic lakes [26,41]. P. agardhii also grows best at temperatures between 10-20 • C [42], and may become disadvantaged by warmer temperatures. Additionally, at the regional scale in Brittany, P. agardhii was the dominant taxa of the freshwater cyanobacterial community between 2004 and 2011 [43].
Surprisingly, none of the measured microcystin congeners was detected during Planktothrix agardhii blooms despite Planktothrix sp. being able to form toxic blooms in temperate freshwater ecosystems [44,45]. Variation of toxin production can be explained by (i) presence/absence of mcy genes necessary for their synthesis and (ii) individual variation within mcy genotypes with inactivation or regulation at the level of genes expression [46,47]. Natural Planktothrix agardhii blooming populations can be composed of microcystin producing (mcy) and non-producing (non-mcy) genotypes, and their proportion can vary considerably [47][48][49]. Moreover, it was demonstrated on non-mcy strains of P. agardhii isolated from nine European freshwater bodies to have lost more than 90% of their mcy genes during evolutionary processes [50]. Based on the analyse of 138 Planktothrix strains from three continents, the variable spatial distribution of mcy and non-mcy genotypes was suggested to depend on ecophysiological adaptation [51]. In laboratory experiments, non-mcy strains of P. agardhii seem to have better fitness than mcy strains under non-limiting conditions [52]. Similar results were obtained for Microcystis: non-mcy strains dominated under optimal growth conditions [53]. The authors of these studies then hypothesized that when cyanobacteria grow under favourable environmental conditions, the cost of producing microcystins becomes too high compared to the advantages it can bring [52]. Based on these results, it is tempting to hypothesize that non-limiting growth conditions concerning P and N may have favoured the selection of non-mcy Planktothrix agardhii strains in the Lac-au-Duc reservoir. It would be interesting to verify the presence of mcy genes, especially since microcystin detections were demonstrated to negatively correlate to Planktothrix biovolume in several Brittany lakes [43], suggesting that non-mcy P. agardhii strains are prevalent at the regional scale. From the composition of the cyanobacterial bloom, also other toxins, such as anatoxin-a and cylindrospermopsin could have been expected [44,54], they were however never detected during the monitoring between 2007-2019.
Since summer 2013, Microcystis and Dolichospermum became dominant or codominant, related to generally calmer conditions. Both genera are known to benefit greatly from water column stability as they can regulate their buoyancy according to their requirements in the illuminated area of stable lakes [55][56][57]. Microcystis occurrence was moreover strongly related to light intensity and temperature, which enhances the stabilization of the water column and favoured the development and persistence of Microcystis blooms in other lakes as well [16,58]. Temperature also benefits directly cyanobacterial development through their growth rate [59], and has been the most important factor driving the development of Microcystis sp. as analysed in more than 1000 lakes [60]. At the regional scale, light intensity was identified as the main climatic driver for Microcystis sp. [43]. Despite dominant in 2013 and 2015, Microcystis abundances and associated toxin production remained low in Lac-au-Duc compared to other lakes (e.g., [16,24], suggesting that the favourable conditions for its development were not fully met. The optimal growth rate for Microcystis is well above 25 • C [59], a temperature rarely reached in Lac-au-Duc or in the surrounding region [43]. We can also hypothesize that Microcystis was limited by nitrogen in 2018 and 2019, when the species dominated in early summer but disappeared thereafter. Although Microcystis is able to use diverse forms of nitrogen, N availability appears to be an essential element controlling the development of this cyanobacteria and its toxin production [61,62].
This context of N limitation in Lac au Duc could also explain the emerging occurrence of the diazotrophic Dolichospermum in recent years, as underlined by the threshold above which its occurance decreases. Nitrate concentrations progressively declined during the last decade in the river entering the reservoir, as in other rivers at the regional scale [63]. Thus, N-fixation ability provides an ecological advantage for this cyanobacterium [41].
Indeed, the first occurrence of Dolichospermum in 2012 and its largest bloom in 2017 corresponded to the driest hydrological years of the time series, that is the years with the lowest nitrogen recharge. Drought also seems to have an impact on the hydrological functioning of the reservoir: from 2015 onwards, due to the succession of rather dry hydrological years, the water level remained low, when outflow stopped from beginning of July until autumn. Water movements were doubtless strongly reduced and this phenomenon was probably exacerbated during dry summers such as in 2017 and 2018, when the inflow of water ceased already in early summer. Air temperature, residence time and light contributed to both axis of the CCA, therefore jointly with time on the first axis, indicating that both parameters tend to increase over time. The occurrence of Dolichospermum was seems partly related to a long water residence time producing a favourable environment for these buoyant cyanobacteria. These results are in accordance with those of Hayes et al. (2015) who found evidence in 42 lakes from agricultural watersheds that droughts strongly influence the system towards N limitation and induce the development of diazotrophic cyanobacteria.
We hypothesize that the lake is in the progress to shift to a new equilibrium, dominated by N 2 fixing cyanobacteria. This is potentially connected to toxin production, but evidence for a causal link between reduced N loading and diazotrophic cyanobacteria such as Dolichospermum abundance or biovolume is mixed [57]. In addition, evidence is increasing that N 2 fixation cannot always compensate significantly for the N deficiency, underlying the need to continue reducing emission of both nitrogen and phosphorus in the catchment [20,[64][65][66][67]. Thus, a further monitoring of the cyanobacterial community is recommended.
This study provides another proof for the necessity to apply a catchment wide approach to limit P and N entrance into lakes and reservoirs. While reducing N was successful, it remains difficult to reduce non-point source P, despite efforts undertaken by farmers in changing agricultural practices. Moreover, depending on the water-bodies' bathymetry and the P stocked in the sediment, the time to reach and restore less eutrophe conditions needs to be taken into consideration. Many environmental parameters can be intercorrelated (nutrient availability, residence time, surface water temperature, stratification, etc.), making it impossible to separate the individual effect of these factors on a community. The complementary approach to such environmental surveys is to use very long term data, a multiple sites approach or mesocosms experiments. The intensity (frequency, stations and parameters) of long term sampling campaigns may present shortcomings, as in our case due to the restauration attempts with applications of CuSO 4 amongst others, thus we were forced to eliminate several of the years, which may have weakened the data set.

Conclusions
To conclude, over only 13 years, two major shifts in the cyanobacterial community have been recorded: a first shift in 2013 from a mixing tolerant population of Planktothrix to a buoyant species, Microcystis, preferring more stable water conditions, influenced by generally calmer and warmer conditions. A second shift from 2017 onwards was driven simultaneously by droughts (representing the biggest change) and reduced N loadings favouring the diazothrophic Dolichospermum. Our study also points out that if P reduction is not successful, dominance of Microcystis sp. and Dolichospermum sp. potentially increase. It is tempting to predict that these buoyant cyanobacteria will replace the Planktothrix agardhii population permanently in the context of climate forcing favouring warmer and dryer conditions at the global scale if N and in particular P cannot be reduced below their requirements [45,68]. This could imply many changes in terms of potential toxin production and increased persistance of cyanobacterial populations from year to year with the growing stock of cells in sediment as both Microcystis and Dolichospermum have dormant cells, while Dolichospermum possesses in addition akinetes.
Changes can also concern curative actions as buoyant cyanobacteria are sensitive to artificial mixing for example, in contrast to Planktothrix [69]. Thus, it will be necessary to follow the evolution of future years to confirm or deny the persistent installation of Microcystis or Dolichospermum in relation to climatic variation.

Study Site
The study site, called "Lac au Duc", is one of the largest shallow water bodies (250 ha, 2.6-m depth in mean) in Brittany (western France, Figure 7). This 3-million m 3 recreational and drinking water reservoir drains an agricultural catchment (37,000 ha) with the Yvel river as the main tributary. The lake is used for bathing, fishing and nautical activities. Two types of monitoring data were used in this study: (i) data acquired monthly over a 2018-2019 seasonal cycle at three points in the lake and (ii) data acquired almost weekly over 13 summers, from 2007 to 2019 in a bathing zone. Additional environmental data were collected from governmental and meteorological data bases.

Sampling Sites and Sampling
To examine if nutrients, phyto-and zooplankton were evenly distributed in the reservoir we realized a seasonal cycle. Samples were collected from July 2018 to August 2019 at the entrance, in the middle and in the lower basin of the reservoir (Figure 7) from a boat. Duplicates of five liters of sub-surface water were collected with a 1 m vertical tube sampler at each point for nutrient, phytoplankton and zooplankton analyses. Duplicate samples for dissolved nutrient analyses were filtered on board on sterile Minisart CA 0.45 µm. Nutrient samples were kept at 4 • C until return to the lab, where they were stocked at −20 • C until analysis. Temperature and oxygen were measured along depth profiles with an Idronaut Ocean Seven 316 Plus CTD.

Nutrients
Particulate phosphorus, total dissolved phosphorus (TDP), particulate nitrogen and total dissolved nitrogen (TDN) were measured by colorimetry after digestion with persulfate according to [70], with a limit of detection of 6 µg P L −1 and 50 µg N L −1 . Orthophos-phate (PO 4 3− ) was analysed by the ammonium molybdate method [71] with a limit of detection of 3 µg P L −1 . After nitrate (NO 3 − ) reduction into nitrite (NO 2 − ) with vanadium chloride, NO 2 − (originally present and reduced nitrates) was measured by colorimetry using sulphanilamide and N-1-naphthylethylenediamine dihydrochloride [72], with a limit of detection of 50 µg N L −1 . Colorimetric measurements were realised using a Gallery Photometric Analyser Gallery Plus (Thermo Fisher, Saint Herblain, France).

Phytoplankton
For the seasonal cycle, microalgae and cyanobacteria composition was determined directly in the fresh sample when possible or were preserved in Lugols solution and stored at 4 • C. Identification were realized to the genus level according to Komárek. Phytoplankton cell counts were carried out with a Nageotte cell according to [73]. For large colonies, the number of cells was estimated using photos. Microscopic photos were used to assess the number of cells in colonies of picocyanobacterial, using Pegasus software. At least, 400 individuals were counted per sample. For both, a light microscope (Olympus BX 50, Rungis, France) was used. Prior to cell counts and identification, if necessary, microalgae and cyanobacteria were concentrated on a 1 µm Poretics polycarbonate membrane filter, thanks to a filtration pump but with low vacuum pressure. Transfer and resuspension of cells in 1 ml of water was done while the membrane was still wet. Intact colonial chlorophytes and cyanobacteria were checked on the microscope to ensure that damage due to the concentration of cells was low.

Zooplankton
Zooplankton was concentrated from two liters of water by filtering through a 50-µm net, then narcotized with soda water and preserved in 70% ethanol at 4 • C until counting. For counting, 1 mL of the concentrated sample was distributed on a Sedgewick-Rafter counting chamber. For each sample, a minimum of 400 rotifers were counted and identified at the genus or species level, according to USEPA protocol [74] with a Zeiss microscope, based on the identification manuals for rotifers [75] and cladocerans [76]. In low-abundance samples, the totality of the 1 mL was observed. In addition, crustaceans and cladocerans were counted.  1). Saxitoxin was determined with an LOQ of 0.6 µg L −1 , anatoxin LOQ < 0.3 µg L −1 and cylindrospermopsin LOQ < 0.6 µg L −1 . All cyanobacterial cell densities data were converted to biovolume with the use of relevant geometric formulas and data literature [80,81], to account for differences in contribution from larger species and smaller species.

Abiotic Parameters
Daily meteorological data (air temperature, global radiation intensity and wind speed) were collected from Météo-France database from 2007 to 2019. Measurements were realised at the Météo-France station of Ploërmel, located 0.5 km from the reservoir (Figure 7). The light energy sensor broke in July 2017, from this date measurements from the Rennes-St-Jacques Météo-France station, located 60 km from the reservoir, were used. The number of days with an air temperature above 20 • C, or with solar radiations greater than 2800 J cm −2 were calculated. The water flow as well as nutrient concentrations were measured in the Yvel river, 3 km upstream the entrance of the reservoir (Figure 7). The water flow data were provided for the 2007-2019 period from the national database Hydro France [82] and corresponded to the daily mean values calculated from continuous stage records. From these data, we calculated the residence time of the water in the reservoir, knowing that its volume is approximately 3 million m 3 . To take into account the time needed between climate variation and phytoplankton response, hydrological and meteorological data were averaged over 6 days preceding the sampling date of cyanobacteria. Wind variation completed the average data. Windy summers were identified by the number of days of wind above a threshold of 4 m s −1 during a 24 h average following [82]. Monthly data for nitrate and total phosphorus concentrations were provided from the database of the Yvel-Yvet watershed manager (the Syndicat Mixte du Grand Bassin de l'Oust, SMGBO) from 2007 to 2019. By combining these nutrient data with flow data, we calculated the quantities of NO 3 − and Ptot entering the reservoir over the summer period or the complete hydrological year (November to November). The city of Ploërmel also provided reservoir water level data from the drinking water purification station.

Statistical Analyses
Despite being available, we discarded data from 2002 to 2005 as some copper sulphate treatments had been applied in the whole reservoir up to summer 2005. Treatments with calcium carbonate and hydrogen peroxide also took place, but locally in the bathing area, between 2013 and 2015 and in July 2018, respectively. During these years, except 2018, no significant difference was observed between the bathing area and the rest of the lake. We then chose to keep the data from the long-term monitoring bathing area for 2013-2015, while we used data from outside of the enclosed area only in 2018. Monitoring of nutrient concentrations started in 2007, thus data from 2006 were not used in analyses, leaving an extra year after the copper sulphate treatments.
All statistical analyses were performed in R Studio version 1.2.5019 [83]. In order to link cyanobacterial blooming genera with environmental parameters and characterize the sampling dates, a canonical correspondence analysis (CCA) was performed on Hellinger transformed genera biovolume (response matrix) and centred-reduced environmental parameters (explanatory variables). To reduce the number of explanatory variables, the ordistep function of the vegan package has been used to find the most parsimonious model based on Akaike Information Criterion (AIC). The less significant explanatory variables were eliminated from the CCA, in order to identify only the significant ones. The significance of the final CCA has been tested through a permutation test with the function envfit. A reference distribution under H0 from the data themselves is generated by permutations of rows and columns and by calculating the new percentage of constrained variance (sum of all canonical eigenvalues). The collinearity between explanatory parameters has also been checked.
We used generalized additive models (GAM) for time series of abiotic parameters, to analyze their temporal dynamics according to the season and the year. GAM takes into account the nonlinear response between the dependent variable (the abiotic parameter) and the explanatory variables (time, month of the year and year), using smooth functions called splines. Finally, the final shape of the relationship is determined by the data themselves. The independant variables in the initial model were the month of the year and the year, except for nitrates, for which it was only time (date) due to little data ( Table 3). The best model was selected based on the gain in deviance explained (DE) relative to the initial model, while minimizing the Akaike's information criterion (AIC). Models were adjusted using the R package 'mgcv' and the function gam() with the selection of the Restricted Maximum Likelihood (REML) method. The family was Gaussian and the link function was identity. A Log 10 transformation was done on the abiotic parameter when needed.