Introduction

Cultural eutrophication, resulting from excessive nutrient loading from human activities, is a worldwide problem for both ecological status and drinking-water quality. Eutrophication has many undesirable symptoms often including high phytoplankton biomass, dominance of harmful cyanobacteria, oxygen depletion, decrease in water transparency, unpleasant taste and odor, and drinking-water treatment problems (Smith & Schindler, 2009). Generally, phosphorus is considered the most important limiting nutrient in lakes and the one responsible for eutrophication (Schindler, 1977, 2012), although other investigators have advocated for consideration of both nitrogen and phosphorus (Conley et al., 2009; Paerl et al., 2011; Moss et al., 2013). Although Lewis (2000) hypothesized that nitrogen limitation of phytoplankton is common in tropical lakes, studies based on large datasets have reported no relationships with latitude (Huszar et al., 2006; Elser et al., 2007; Kosten et al., 2009). The concept of nutrient limitation for algal growth is the keystone of eutrophication research because it implies that (i) a single nutrient should be the primary limiting factor for algal growth in a given lake; (ii) algal growth in a given lake should be positively related to the supply of this nutrient; and (iii) practical control of algal growth and of eutrophication in the lake should involve restricting the input of this nutrient to the system (Smith, 1998).

Recent studies have suggested that climate change may exacerbate the symptoms of eutrophication (Moss et al., 2011; Jeppesen et al., 2014; Rigosi et al., 2014). According to the report of the Intergovernmental Panel on Climate Change (IPCC, 2013), many arid and semiarid regions are likely to become warmer and drier by the end of this century due to human-induced climate change. For instance, climate models for northeastern Brazil predict a temperature increase of about 4°C and rainfall reduction of about 40%, and increases in the intensity or duration of drought are likely by the end of the 21st century (Marengo et al., 2009, 2010). Lower precipitation and more frequent droughts are expected to accentuate fluctuations in water level, an important element of hydrology driving man-made lakes in arid and semiarid regions. A recent review indicates that water-level reduction often results in higher nutrient concentrations, higher phytoplankton biomass, and lower water transparency in both shallow and deep lakes and reservoirs (Jeppesen et al., 2015). In addition, reduction in water level due to drought and water removal for multiple purposes might increase salinity and conductivity (Jeppesen et al., 2015). Lower precipitation will reduce runoff and potentially decrease nutrient loading (Jeppesen et al., 2009, 2011). This reduction, however, is not expected to compensate for the negative consequences of water deficits that will lead to nutrient concentration from point sources and reinforcement of eutrophication in lakes (Beklioglu et al., 2007; Özen et al., 2010). Moreover, drought may increase the water residence time in lakes, and some studies have shown that longer water residence times during dry years increase cyanobacteria biomass and dominance (Elliott, 2012; Romo et al., 2012).

In comparison to temperate regions, very little is known about how water-level fluctuation affects the ecology of reservoirs in the tropics. Man-made lakes in warm semiarid northeastern Brazil show seasonal variations in water volume and increasing concentrations of nutrients and phytoplankton chlorophyll, and decreasing water level in dry season. However, below a critical threshold of water volume, the high levels of abiotic turbidity, caused by sediment resuspension by wind and fish, may inhibit phytoplankton growth, leading to a reduction in nutrient concentrations, phytoplankton biomass, and cyanobacteria dominance (Jeppesen et al., 2015; Medeiros et al., 2015). In the same semiarid region, Bouvy et al. (2000) showed that the lack of water renewal linked to a drought event was the major determinant of both eutrophic conditions and cyanobacteria dominance. A more recent study showed that the effects of drought on water quality and ecology in Ethiopian reservoirs depend on their propensity to dry out (Teferi et al., 2014). Reservoirs that refilled after drying had significantly lower nutrient concentrations; lower biomass of phytoplankton, cyanobacteria, and fish; greater macrophyte cover; and clearer water than reservoirs that did not dry (Teferi et al., 2014).

In tropical regions, cyanobacteria may show persistent annual dominance with relatively small changes during the year (Figueredo & Giani, 2009; Soares et al., 2009) and often are toxic (Bouvy et al., 1999; Molica et al., 2005; Duong et al., 2013). Although cyanobacteria blooms are typically associated with nutrient enrichment, their appearance has been related to several factors, such as adaptation to low light conditions (Scheffer et al., 1997), high temperature, and water column stability (Paerl & Huisman, 2009), low CO2 concentrations and high pH levels (Shapiro, 1990; Caraco & Miller, 1998), nitrogen limitation (Smith, 1983), ability to store phosphorus (Isvánovics et al., 2000), production of allopathic substances (Leão et al., 2009), and resistance to herbivory (Wilson et al., 2006). Nevertheless, cyanobacteria are sensitive to flushing (Elliott, 2012; Romo et al., 2012) and suspended clay particles (Cuker et al., 1990; Allende et al., 2009). Cyanobacteria form a diverse group of organisms that differ widely in functional traits, and as a result, a wide variety of environmental variables may determine their population dynamics (Soares et al., 2013; Rigosi et al., 2014).

There is an urgent need for empirical studies to improve our understanding of the impacts of current droughts and future climate change on tropical freshwater ecosystems. We hypothesized that a reduction in water level caused by droughts will aggravate eutrophication symptoms, leading to higher cyanobacteria biomass and dominance. To test this hypothesis, we analyzed physical and chemical variables and plankton communities of 40 man-made lakes in semiarid northeastern Brazil at the end of the wet and dry seasons. We also constructed a predictive model of cyanobacteria biovolume in these lakes.

Materials and methods

Study sites

The 40 man-made lakes are located in a warm semiarid region (05°34′S to 06°44′S and 35°49′W to 38°25′W), encompassing six hydrographic basins in the state of Rio Grande do Norte in northeastern Brazil (Fig. 1). These lakes are impoundments of temporary streams constructed to store water for the long dry season typical of that region, and they are used for multiple purposes such as drinking water, irrigation, ranching, recreation, fishing, and aquaculture. The historical annual mean precipitation is 726 mm (1960 to 2006 period), with a well-defined rainy season between January and July, and an almost complete lack of precipitation during the remaining months. Precipitation levels exceeding the evapotranspiration rate usually occur only from February through April. Monthly mean air temperatures range from 24.2 to 26.5°C and are lowest from June through August (Data source: Agricultural Research Company of Rio Grande do Norte State—EMPARN). The local soils are shallow and highly susceptible to erosion; the predominant vegetation is a mosaic of thorn scrub and seasonally dry forest called Caatinga (Leal et al., 2005). In our study area, about 75% of the Caatinga has been devastated due to human activities (deforestation, fire, agriculture, pasture, and urbanization) (Santos, 2014). In addition, on average only about 12% of municipal sewage is treated (Data source: Water and Sewerage Company of Rio Grande do Norte State). Most of the lakes were constructed recently (mean age = 51 years) and are shallow (mean depth = 5 m), with surface areas ranging from 1 to 11 km2, and the maximum storage volume varies widely (Table 1).

Fig. 1
figure 1

Map and location of Rio Grande do Norte, Brazil, showing the 40 man-made lakes studied (dots)

Full size image
Table 1 General features of 40 man-made lakes in northeastern Brazil (Data source: State Department of the Environment and Water Resources, SEMARH)
Full size table

Sampling

The lakes were sampled twice, in December 2007 (end of the dry season) and July 2008 (end of the wet season). Sampling was conducted in the deepest part of each lake, near the dam. Water samples for nutrients, suspended solids, chlorophyll a, and phytoplankton analysis were collected with a Van Dorn bottle at 1-m vertical depth intervals until 1 m above the bottom and integrated to obtain a representative sample from the water column of each lake. Filtration for dissolved nutrients and chlorophyll a analyses was conducted directly after collection, and the water and filters were frozen until analysis. Bacteria samples were fixed with formalin (2% final concentration). Phytoplankton samples were fixed in Lugol’s solution. Zooplankton samples were collected with a 50-µm-mesh plankton net towed vertically. The volume of filtered water was calculated as the product of the net mouth area and the tow depth, assuming a 100% filtering efficiency of the net. The zooplankton samples were preserved in a final concentration of 4% formaldehyde solution. Water temperature, dissolved oxygen, pH, electrical conductivity, and turbidity were measured in situ at 1-m intervals (Horiba model U-22 multiparameter probe, Kyoto, Japan). Water transparency was measured with a Secchi disk.

Climate variables (mean air temperature, precipitation and evapotranspiration) were estimated from isoline maps available from the web site of the Agricultural Research Company of Rio Grande do Norte State (www.emparn.rn.gov.br, July 30, 2010).

Sample analysis

Filtered samples were used for dissolved nutrient analysis. Nitrate was reduced to nitrite in alkaline-buffered solution by passing the sample through a column of copperized cadmium filings (Wetzel & Likens, 2000). Soluble reactive phosphorus (SRP) was determined through the reaction with a reagent composed of molybdate, ascorbic acid, and trivalent antimony (Wetzel & Likens, 2000). Ammonium nitrogen was analyzed by the phenol-hypochlorite method, using nitroprusside as a catalyst (Wetzel & Likens, 2000). Reactions with acidic ammonium molybdate and the reduction by sodium sulfite were used to analyze dissolved silica (Wetzel & Likens, 2000). Non-filtered samples were used to analyze total phosphorus, determined by oxidation using persulfate (Wetzel & Likens, 2000), and total organic nitrogen (Kjeldahl nitrogen) (Mackereth et al., 1978). Chlorophyll a (Chl-a) was extracted from filters (Whatman 934-AH) with 96% ethanol at room temperature for 20 h, and absorbance was measured at 665 and 750 nm (Jespersen & Christoffersen, 1988). To determine the concentrations of total suspended solids (TSS), the water samples were filtered with Whatman 934-AH filters and the residue retained on the filter was dried to constant weight at 105°C. To determine the concentration of inorganic suspended solids (ISS), the filters were incinerated in a muffle furnace at 550°C for 15 min and the ashes were weighed. Organic suspended solids (OSS) were determined by the difference between the TSS and ISS, according to APHA (1998).

Bacterioplankton samples were prepared according to Hobbie et al. (1977). Each sample was stained with acridine orange (final concentration: 0.01%) and then filtered through a polycarbonate membrane (Millipore, pore size 0.2 µm, diameter 25 mm, black). Bacteria abundance was estimated by direct counting at 1000x magnification under an epifluorescence microscope (Olympus BX-60), according to Daley & Hobbie (1975). In each sample, 10 random fields and at least 200 cells were counted.

Phytoplankton populations were enumerated in random fields (Uehlinger, 1964) using the settling technique (Utermöhl, 1958) under an inverted microscope (Zeiss Axiovert 10, Jena, Germany). The units (cells, colonies, and filaments) were enumerated for at least 100 specimens for the most frequent species (Lund et al., 1958) at ×400 magnification.

Zooplankton was counted under a microscope (Olympus, CX 41) in 1-ml Sedgwick-Rafter chambers. Between three and five subsamples were counted for each sample, until a minimum of 100 individuals of each taxonomic group (rotifers, copepods, cladocerans) were counted.

Calculations and data analysis

Relative water column stability (RWCS) was used as an index of water column stability at the time of sampling. RWCS was calculated according to the following equation (Padisák et al., 2003):

$${\text{RWCS}} = \frac{{D_{b} - D_{s} }}{{D_{4} - D_{5} }},$$

where D b is the density of the bottom water, D s is the density of the surface water, and D 4 and D 5 are the water densities at 4 and 5°C, respectively, according to Hutchinson (1957).

Dissolved inorganic nitrogen (DIN) was calculated as the sum of ammonium, nitrate, and nitrite. Total nitrogen was calculated as Kjeldahl nitrogen plus nitrate. CO2 concentrations were calculated indirectly, from measurements of alkalinity, pH, and water temperature, according to Stumm & Morgan (1996) and Weiss (1974). The trophic state of the lakes was defined based on Chl-a and nutrient concentrations, according to Nürnberg (1996). For that, Chl-a and nutrient concentrations were averaged over two seasons.

To evaluate the potential nutrient limitation on phytoplankton growth, we used the following indicators (Kosten et al., 2009): (1) if lake water TN/TP ratios were below 20 (molar based), a lake was considered N limited, and if they were above 38, a lake was P limited (Sakamoto, 1966); (2) if DIN/SRP ratios were below 13 (molar based), a lake was considered N limited, and if they were above 50, a lake was P limited (Morris & Lewis, 1988). If the ratios were in between those reference values, a lake was considered no N or P limited; and (3) SRP and DIN were compared to concentrations that have been considered to roughly limit phytoplankton growth; phosphorus was considered limiting below 10 µg P l−1 (Sas, 1989) and N below 100 µg P l−1 (Reynolds, 1999). Clearly, these are only approximations, as nutrient limitation depends on the attributes and storage capacities of the individual species (Reynolds, 1999).

Bacteria biomass was estimated by converting mean biovolume to biomass. To estimate the bacteria biovolume, we used an equation proposed by Massana et al. (1997): (4/π) × (W 2) × [L − (W/3)], where L is the length and W is the width. To convert biovolume to biomass, we used the equation of Norland (1993): m = CV a, where m is biomass, V is the estimated biovolume, C is the conversion constant (120 fg C cel−1), and a is a correction factor (0.76). The biovolume was calculated using the software Image-Pro Plus. At least 250 cells were measured to calculate the biovolume in each sample.

Phytoplankton biovolume (in cubic millimeters per liter) was estimated by multiplying the abundance of each species by the mean cell volume (Hillebrand et al., 1999), based on measurements of at least 30 individuals. To estimate zooplankton biomass, at least 30 randomly selected individuals of the most abundant species were measured. For rotifers, geometric formulas were used to estimate biovolume (Ruttner-Kolisko, 1977). Fresh weight was estimated from the biovolume of each individual, assuming that 106 µm3 corresponds to 1 µg of fresh weight (Ruttner-Kolisko, 1977). Dry weight was estimated as 10% of fresh weight (Pace & Orcutt, 1981). Microcrustacean biomass (copepods and cladocerans) was estimated using regression equations relating dry weight and body length, according to Bottrell et al. (1976). Zooplankton were classified according to their food preference: small filter-feeders (rotifers, nauplii, and copepodites), medium-sized filter-feeders (calanoid copepods and cladocerans), and omnivorous–carnivorous (cyclopoid copepods) (Loverde-Oliveira et al., 2009).

The ratio of zooplankton biomass (μg dry weight l−1)/algal biomass (Chl-a multiplied by 66, to convert μg l−1 to μg dry weight l−1) was used as a proxy for grazing pressure (Jeppensen et al., 2005), assuming that zooplankton use phytoplankton as their sole food source. The ratio gives a rough indication of the proportion of phytoplankton standing stock that is grazed per day, assuming that zooplankton consumes its biomass per day (Jeppesen et al., 1994).

Statistical analysis

Differences between the two study periods (dry and wet season) in relation to limnological variables were assessed by t test (α = 0.05). To explore the relationships among cyanobacteria biovolume and limnological variables, standard linear regressions were used. The potential multicollinearity among predictive variables was tested using the variance inflation factor (VIF > 10) (Legendre & Legendre, 2012). The 10 original variables were maximum depth, RWCS, Secchi depth, water temperature, pH, inorganic suspended solids, total nitrogen, total phosphorus, biomass of medium-sized zooplankton filter-feeders, and potential grazing pressure. The Akaike Information Criterion unbiased for small sample size (Akaike, 1974) was used to identify a single best model (Anderson, 2008). The Akaike weights were interpreted as the probability that i is the best model, given the data and set of candidate models. For t test and regression analyses, the biotic and abiotic data, except pH, were Log (x + 1) transformed to meet the assumptions of the analyses. The assumption of homogeneity of variance in the t test was assessed through Levene’s test. When this assumption was not satisfied, we used the t test with unequal variance to compare the differences between the seasons (Zar, 2010). The assumptions of linearity of relationships between the independent and dependent variables, and that the residuals are normally distributed in regression analyses were assessed through a scatterplot of predicted values versus residuals and a normal probability plot of residuals. All statistical analyses were performed using Statistica 7.0 (StatSoft, Inc., Tulsa, OK, U.S.A.).

Results

Abiotic environment

In 2007, the mean annual precipitation was lower (507 mm) and in 2008 was higher (980 mm) than the historical mean (726 mm) (Fig. 2).The total precipitation from January through December in 2007 was 36% lower than the cumulative precipitation from January through July in 2008. The mean air temperature in December 2007 (30.6°C) was four degrees higher than in July 2008 (26.5°C), when the water sampling was carried out.

Fig. 2
figure 2

Mean values of historical (1960–2006 period) monthly mean precipitation, air temperature, evapotranspiration, and precipitation in the two study years. Open circle indicates sampling month in 2007 and closed circle in 2008 (Data source: Empresa de Pesquisa Agropecuária do Rio Grande do Norte—EMPARN)

Full size image

Large environmental differences were observed between the two contrasting climate conditions. As a consequence of the higher mean air temperature and evapotranspiration, and the lack of precipitation during the dry season (Fig. 2), significant reductions in water storage and maximum depth (z max) occurred in dry compared to wet conditions (Table 2; Fig. 3). Significantly higher water temperature and water column stability (RWCS) were also observed in dry than in wet conditions. The lakes showed low water transparency (Secchi depth < 1 m) in both climate conditions but significantly lower Secchi depth and higher turbidity, concentrations of total suspended solids (TSS), and organic suspended solids (OSS) were observed under the dry conditions. There was no significant difference, however, in the concentrations of inorganic suspended solids (ISS) between the seasons (Table 2; Fig. 3). Significantly higher values of electrical conductivity and pH, and lower CO2 concentrations occurred in the dry season (Table 2; Fig. 3). Although there was no significant difference in the mean oxygen concentrations from entire water column, oxygen concentrations near the bottom were significantly lower in dry than wet season (Table 2; Fig. 4). The water near the bottom seldom became anoxic (ODb, dissolved oxygen 1 m above the bottom <1 mg l−1), but this occurred somewhat more frequently in dry than in wet conditions (Fig. 3).

Table 2 T tests for differences in limnological features of 40 man-made lakes in Rio Grande do Norte, Brazil, between dry and wet seasons (n = 80)
Full size table
Fig. 3
figure 3

Box plots of a percentage of maximum storage volume; b z max maximum depth; c water temperature; d RWCS, relative water column stability; e Secchi disk depth; f turbidity; g TSS total suspended solids; h OSS organic suspended solids; i ISS, inorganic suspended solids, j Cond electrical conductivity; k pH; l CO 2 carbon dioxide concentrations; m DOb dissolved oxygen concentrations 1 m above the bottom found in 40 man-made lakes in Rio Grande do Norte, Brazil, in the dry and wet seasons (n = 80). The horizontal lines inside the box plots indicate the median, and the boundaries of the box plots indicate the 25th and 75th percentiles. Whiskers above and below indicate the 90th and 10th percentiles. Dots and stars are outliers and extreme points, respectively

Full size image
Fig. 4
figure 4

Box plots of a SRSi, soluble reactive silica; b N NO − 3 nitrate; c N NH4 +, ammonium; d DIN dissolved inorganic nitrogen; e SRP soluble reactive phosphorus; f DIN:SRP molar ratio; g TN total nitrogen; h TP total phosphorus; i TN:TP molar ratio found in 40 man-made lakes in Rio Grande do Norte, Brazil, in the dry and wet seasons (n = 80). The horizontal lines inside the box plots indicate the median, and the boundaries of the box plots indicate the 25th and 75th percentiles. Whiskers above and below indicate the 90th and 10th percentiles. Dots and stars are outliers and extreme points, respectively. Light-gray areas represent potentially N-limited conditions; dark-gray areas represent potentially P-limited conditions; and white, no limitation

Full size image

Higher N NH4 + and TN concentrations were observed in dry than in wet conditions, but no significant differences were found in SRSi, NO3 −, DIN, SRP, TP, DIN/SRP, and TN/TP ratios between the two seasons (Table 2; Fig. 4). Only 4% of the samples had TN/TP ratios >38 and 24% had SRP <10 µg l−1, suggesting no potential P limitation in most systems in both climate conditions (Fig. 4e, i). On the other hand, DIN/SRP ratios <13 and DIN concentrations <100 µg l−1 were observed in 85 and 84% of the samples, respectively (Fig. 4d, f). Therefore, analysis of the nutrient indicators suggested a potential N limitation to phytoplankton growth in the lakes in both seasons.

Plankton communities

Based on Chl-a and nutrient concentrations, most of the systems were hypereutrophic and eutrophic (Table 3). Bacteria and phytoplankton biomass (Chl-a concentrations and total phytoplankton biovolume) and cyanobacteria biovolume were significantly higher in dry than in wet conditions; the biovolume of diatoms, chlorophyceans, cryptomonads, zygnemaphyceans, and other minor phytoplankton groups (dinoflagellates, euglenoids, and xanthophyceans) did not change (Table 4; Fig. 5). The contribution of cyanobacteria to the total phytoplankton biovolume (cyanobacteria percentage) was significantly higher, while the percentages of diatoms and other minor phytoplankton groups were significantly lower under dry conditions; the percentages of chlorophyceans, cryptomonads, and zygnemaphyceans did not change with the season (Table 4; Fig. 6).

Table 3 Proportion of lakes mesotrophic, eutrophic, and hypertrophic in a set of 40 man-made lakes in Rio Grande do Norte, Brazil
Full size table
Table 4 T tests for differences in bacteria, phytoplankton, and zooplankton communities in 40 man-made lakes in Rio Grande do Norte, Brazil, between dry and wet seasons (n = 80)
Full size table
Fig. 5
figure 5

Box plots of a bacteria biomass; b Chl-a chlorophyll a; c Phyto total phytoplankton biovolume; and biovolume of d Cya cyanobacteria; e Dia diatoms; f Chl chlorophyceans; g Cry cryptomonads; h Zyg zygnemaphyceans; i Others other phytoplankton groups (dinoflagellates, euglenoids, and xanthophyceans), j Nost Nostocales cyanobacteria; k Osci Oscillatoriales cyanobacteria; l Chroo Chroococcales cyanobacteria found in 40 man-made lakes in Rio Grande do Norte, Brazil, in the dry and wet seasons (n = 80). The horizontal lines inside the box plots indicate the median, and the boundaries of the box plots indicate the 25th and 75th percentiles. Whiskers above and below indicate the 90th and 10th percentiles. Dots and stars are outliers and extreme points, respectively. Note the different scaling of the y-axis

Full size image
Fig. 6
figure 6

Box plots of percentages of total phytoplankton biovolume of a cyanobacteria; b diatoms; c chlorophyceans; d cryptomonads; e zygnemaphyceans; f other phytoplankton groups (dinoflagellates, euglenoids, and xanthophyceans) found in 40 man-made lakes in Rio Grande do Norte, Brazil, in the dry and wet seasons (n = 80). The horizontal lines inside the box plots indicate the median, and the boundaries of the box plots indicate the 25th and 75th percentiles. Whiskers above and below indicate the 90th and 10th percentiles. Dots and stars are outliers and extreme points, respectively

Full size image

Cyanobacteria contributed the most to total phytoplankton biovolume, in both dry (72.4%) and wet (53.8%) conditions (Table 4). The t test showed a significant increase in biovolume of Nostocales (filamentous potentially N2-fixing cyanobacteria) and Oscillatoriales (filamentous non-N2-fixing cyanobacteria) in dry conditions (Table 4; Fig. 5). Nostocales species were the dominant forms in both seasons, contributing 62 and 42% to cyanobacteria biovolume in dry and wet conditions, respectively. Oscillatoriales contributed 27 and 32% to cyanobacteria biovolume in dry and wet conditions, respectively. The most important phytoplankton species in the systems was the nostocalean Cylindrospermopsis raciborskii (Woloszynska) Seenaya & Subba Raju. On average, C. raciborskii contributed 22% to total phytoplankton biovolume; it was present in 70% of the samples and reached >30% of total phytoplankton biovolume in 30% of the lakes (Table 5). During the survey, heterocyted filaments, on average, comprised 31% of C. raciborskii abundance. No significant differences were observed in Chroococcales (unicellular or mucilaginous colonial cyanobacteria) between the two seasons (Table 4; Fig. 5).

Table 5 Dominant phytoplankton species in 40 man-made lakes in Rio Grande do Norte, Brazil (n = 80)
Full size table

Zooplankton biomass and the proxy for grazing pressure did not differ significantly between dry and wet conditions (Table 4). On average, medium-sized zooplankton filter-feeders dominated in both seasons, contributing 57 and 48% to the total zooplankton biomass under dry and wet conditions, respectively (Fig. 7).

Fig. 7
figure 7

Box plots of biomass of a SFZ small zooplankton filter-feeders; b MFZ medium-sized zooplankton filter-feeders; c OCZ omnivorous–carnivorous zooplankton; d Zoo total zooplankton; e Grazing, grazing pressure found in 40 man-made lakes in Rio Grande do Norte, Brazil, in the dry and wet seasons (n = 80). The horizontal lines inside the box plots indicate the median, and the boundaries of the box plots indicate the 25th and 75th percentiles. Whiskers above and below indicate the 90th and 10th percentiles. Dots and stars are outliers and extreme points, respectively. Note the different scaling of the y-axis

Full size image

Regression analyses

The model probabilities (w i) indicated that a model including Secchi depth, RWCS, pH, inorganic suspended solids (ISS), and TN was likely to best explain the biovolume of cyanobacteria, given the data and set of candidate models (model 1, Table 6). Cyanobacteria biovolume was positively related to RWCS, pH, and TN and negatively related to Secchi depth and ISS. This model accounted for 40% of the variation in cyanobacteria biovolume and was (0.17/0.09) = 1.9 times more likely to be the best explanation than the extended model including grazing pressure (model 3, Table 6). All variance inflation factors were lower than the critical heuristic value of 10, suggesting that collinearity among the explanatory variables did not strongly affect the results.

Table 6 Rank of the five best linear regression models explaining the cyanobacteria biovolume (Cya) in 40 man-made lakes in Rio Grande do Norte, Brazil
Full size table

Discussion

Our hypothesis that drought conditions may aggravate the symptoms of eutrophication was supported by the data and model in this study. Our results suggest that drought impact the hydrological, chemical, and physical characteristics of man-made lakes in tropical semiarid regions, favoring cyanobacteria blooms.

Studies conducted in other man-made lakes in semiarid northeastern Brazil have also shown that the annual rainfall deficit and lack of water renewal linked to an El Niño event seem to be the major factors responsible for both eutrophic conditions and cyanobacteria blooms (Bouvy et al., 1999, 2000). Studies elsewhere have also suggested that droughts will decrease the flushing rate and increase the water residence time, contributing to cyanobacteria blooms (Elliott, 2012; Romo et al., 2012). Indeed, at the end of the wet season, when the large majority of studied lakes reached their maximum capacity or overflowed, there were reductions in the total phytoplankton biovolume and the absolute and relative cyanobacteria biovolume, indicating that flushing was an important density-independent loss factor to control large cyanobacteria, because of their relatively slow growth rates (Scheffer, 1998; Reynolds, 2006). Thus, our study is in agreement, at least in part, with the general prediction of de Senerpont Domis et al. (2013a, b); that in tropical systems, temporal variability in precipitation can be an important driver of the seasonal development of plankton, and that intense precipitation events may disrupt cyanobacteria blooms. However, these authors assumed that precipitation intensity will increase in the tropics, even in semiarid regions, which is not the case (Sarmento et al., 2013). For semiarid northeastern Brazil, an increase in dry periods and reduction in intense precipitation events are projected (Marengo et al., 2009). Consequently, with the projected reduction in mean precipitation and intense precipitation events, and longer dry periods, we can expect more-prolonged cyanobacteria blooms in tropical semiarid regions.

Reduced precipitation also led to changes in the chemical conditions in these lakes. Lower precipitation and higher evaporation resulted in a reduction in water storage and a 3-fold increase in mean electrical conductivity, increasing the risk of salinization and threatening the water supply. Similarly, in Lake Doirani, Greece (warm semiarid Mediterranean climate), electrical conductivity increased from 0.49 to 1.24 mS cm−1 and showed a significant correlation with the water-level reduction (Jeppesen et al., 2015). Excessive phytoplankton growth sharply increases the CO2 demand to support photosynthesis, decreasing free CO2 availability and consequently raising pH levels (Paerl & Huisman, 2009). This may explain the lower CO2 concentrations and higher pH levels during the dry season in our study. The higher pH resulting from CO2 consumption by high phytoplankton biomass (chlorophyll a and total biovolume) might also have given a competitive advantage to cyanobacteria, at least in the dry conditions, when the mean pH was 8.5. Experimental studies have shown that at alkaline pH (8–9) the relative contribution from cyanobacteria is high, independently of CO2 concentrations, while at pH around 7, the cyanobacteria contribution is negatively related to CO2 concentrations, suggesting that high pH per se may favor cyanobacteria dominance (Caraco & Miller, 1998). In addition, the alkaline pH found in the warmer and drier climate conditions may have favored cyanobacteria dominance, because these phytoplankton can use bicarbonate as a carbon source (Paerl & Huisman, 2009; Holland et al., 2012).

There is a consensus among the scientific community that eutrophication is a major cause of cyanobacteria blooms. Our results showed that most of our systems are eutrophic or hypereutrophic, thus contributing to excessive growth of cyanobacteria. A study to evaluate the effect of land use on water quality in the same semiarid region as our study found a positive relationship between land use (bare soil and urban area) and lake trophic status (Santos, 2014). The higher the proportion of bare soil surrounding the lakes, the higher was the concentration of nutrients (nitrogen and phosphorus), and the higher the proportion of bare soil and urban areas, the higher was the concentration of chlorophyll (Santos, 2014). Furthermore, man-made lakes in semiarid regions tend to have small surface areas relative to their basins (Thornton & Rast, 1993), which together with their long water residence time (months to years) makes these systems highly vulnerable to eutrophication and siltation, due to high input and retention of nutrients and sediments from the basin (Jeppesen et al., 2015). Despite the lack of precipitation and external nutrient supply in the dry season, ammonium and total nitrogen concentrations increased in the lakes studied here, potentially related to internal processes (evaporation, decomposition and internal loading). This was evidenced by the higher phytoplankton and bacteria biomass, and lower oxygen concentration in the water near the bottom during the dry season. Decomposition of organic material normally leads to the release of nitrogen as ammonium, a process called ammonification (Scheffer, 1998). Similarly, a detailed mass-balance study in two Mediterranean lakes found that nutrient concentrations (total phosphorus and dissolved inorganic nitrogen) increased during dry years despite reduced external nutrient loading (Özen et al., 2010).

Contrarily to recent studies (Huszar et al., 2006; Elser et al., 2007; Kosten et al., 2009; Rangel et al. 2012), our study supports the hypothesis that nitrogen limitation is common in tropical lakes (Lewis, 2000), as evidenced by our indicators of nutrient limitation. In general, those studies argue that local factors (land use, catchment characteristics, and hydrology) have a stronger influence on which nutrient is limiting than climate. There are two main potential explanations to the trend toward N limitation in these ecosystems. Firstly, the studied lakes have received urban wastewater effluents, which are proportionally richer in phosphorus than nitrogen (Moss et al., 2013), for decades and have been phosphorus overfertilized, as evidenced by high TP and SRP concentrations. Secondly, the high temperature in the sediment, intense bacterial activity, and sediment deoxygenation increase the release of phosphorus from sediments to water column, while nitrogen may decrease owing to denitrification (Lewis, 2000; Veraart et al., 2011; Moss et al., 2013).

Commonly, a major consequence of the N limitation is the excessive growth of N2-fixing cyanobacteria (Schindler et al., 2008; Moss et al., 2013), as observed in our study. We also found a positive relationship between cyanobacteria and TN concentrations. Interpretation of causality in such regression models is problematic, because fixation of atmospheric nitrogen by N2-fixing cyanobacteria may increase the input of nitrogen into the lake. The high incidence of filaments of C. raciborskii with heterocytes (31%) compared to other systems (<10%, Branco & Senna, 1994; Bouvy et al., 1999; Huszar et al., 2000; Soares et al., 2009) indicates that N2-fixation is a potentially important process in these systems. However, N2-fixation by cyanobacteria is not sufficient to compensate for N deficiency in many lakes (Lewis & Wurtsbaugh, 2008; Scott & McCarthy, 2010; Paerl et al., 2011).

In addition to nutrients, light is also a bottom-up control for phytoplankton growth (Reynolds 2006). Water transparency remained low in the lakes during the entire study period. Turbidity increased in the dry season, and these highly turbid conditions seem to be a decisive factor in favoring filamentous cyanobacteria. These cyanobacteria show a high product of the maximum linear dimension and surface/volume ratio, becoming good light antennae (Reynolds, 2006). The higher proportion of organic to inorganic suspended solids, particularly at the end of the dry season, also suggests that high algal biomass in the lakes is responsible for much of the underwater light attenuation. The patterns observed in the field as well as the physiologically based competition model indicate that dominance by shade-tolerant cyanobacteria, as originally proposed for Oscillatoriales, can be an alternative stable state of the algal community in shallow lakes, because these cyanobacteria are able to cause an increase in turbidity that augments their competitive advantage (Scheffer et al., 1997). The turbid conditions in these systems may have conferred a competitive advantage on C. raciborskii, and because of its superior shade tolerance, it is considered to be a ‘heterocystous Oscillatoria’ (Padisák & Reynolds, 1998).

In addition to the reduction in water storage, and its physical and chemical consequences, the higher temperatures in the dry season may also have favored the cyanobacteria. Recent studies suggest that warming will probably increase the magnitude, frequency and persistence until autumn of cyanobacteria blooms (Kosten et al., 2012; O’Neil et al., 2012; Jeppesen et al., 2014; Izaguirre et al. 2015). High temperatures may give a competitive advantage to cyanobacteria through direct and indirect processes (Paerl & Huisman, 2009). However, Lürling et al. (2013) recently found that cyanobacteria did not show higher growth rates than green algae, for example, with increased temperature. These authors suggested that the competitive advantage of cyanobacteria with global warming is more likely related to the indirect effect of stronger stratification, enhanced by their ability to migrate vertically and avoid settling. In our study, stratification was more stable in the dry season, and our regression model showed a positive relationship between the cyanobacteria biovolume and water column stability. Stratification should favor organisms with functional traits that allow them to regulate their position in the water column. When water column mixing is weak, cyanobacteria may form dense layers at the surface, shading their competitors (Huisman et al., 2004; Jöhnk et al., 2008). Nevertheless, although thermal stability may contribute to cyanobacteria dominance, destratification per se does not necessarily lead to cyanobacteria decline. Water column mixing may be sufficient to restrict the dominance of cyanobacteria such as Microcystis that are dependent on floating to compete for light (Huisman et al., 2004; Jöhnk et al., 2008) but may be less effective for shade-tolerant species such as C. raciborskii (Antenucci et al., 2005; Soares et al., 2013). This is a possible explanation for the high relative contribution of filamentous cyanobacteria at the end of the wet season, although with considerably lower biovolume.

Our best regression model also showed a negative relationship between cyanobacteria biovolume and inorganic suspended solids. Observational and experimental studies have shown that suspended clay may negatively affect cyanobacteria. For instance, a recent study conducted in a man-made lake in the same semiarid region as our study showed that sediment resuspension during a severe drought led to the collapse of the cyanobacteria biomass and dominance (Medeiros et al., 2015). Allende et al. (2009) showed that in inorganic-turbid lakes of the Pampa Plain (Argentina), phytoplankton communities were dominated by diatoms. An experimental study with mesocosms found that additions of suspended clay changed a phytoplankton community dominated by flagellated algae and Chroococcales cyanobacteria to a sparse, flagellate-dominated community (Cuker et al., 1990). Recently, modified clay has been used in combination with flocculents to control cyanobacteria blooms in lakes (Lürling & van Oosterhout, 2013).

In summary, the present study showed that drought increases the risk of salinization and deep anoxia, aggravates the symptoms of eutrophication, and increases the intensity of cyanobacteria blooms, most notably the potentially N2-fixing filamentous forms. Our regression model suggested that cyanobacteria biovolume was positively related to water column stability, pH, and TN, and negatively related to water transparency and concentrations of suspended inorganic solids. Our results imply that the future warmer and drier climate predicted for the semiarid region of northeastern Brazil will reduce the water quantity and quality of man-made lakes. Therefore, more effective reduction of external nutrient loading (both nitrogen and phosphorus) will be needed to manage the water quality of these ecosystems.