Water quality dynamics of floodplain lakes in relation to river flooding and cattle grazing

Vast areas of tropical and subtropical floodplains adjacent to large rivers are occupied by shallow lakes that provide essential ecosystem services such as water purification and supply (Lewin and Ashworth 2014). The functioning of floodplain lakes is typically regulated by fluctuations in the river level (Amoros and Bornette 2002). In tropical and subtropical systems, lakes can be flooded by the river for several months (Junk et al. 1989). These long floods greatly increase the hydrological connectivity of lakes and the exchange of materials (i.e. organic matter, nutrients, sediments, organisms) with the river and diverse aquatic habitats of the floodplain (Neiff 1990; Junk and Wantzen 2004). When the river stage is below the bank-full level, lakes experience a lower hydrological connectivity or even a temporary isolation, depending mainly on the distance to the river and on topography (Drago 2007).

The river–floodplain exchange of matter during floods has a major impact on water quality of floodplain lakes in general, but particularly in tropical and subtropical systems (Hamilton and Lewis 1987; Junk and Wantzen 2004; Mayora et al. 2013). Concentrations of water constituents such as nutrients and organic matter respond to high overbank flow periods in variable ways, depending on the balance between mass inputs from the river and rewetted floodplain areas, dilution, and washing-out effects (Hamilton and Lewis 1990; Maine et al. 2004; Castillo 2020). In addition, water-quality differences among lakes decrease because of their higher hydrological connectivity (Mayora et al. 2020). By contrast, during isolation periods, water quality is subject to local processes such as biological uptake of nutrients, local inputs of materials, and sediment resuspension (Thomaz et al. 2007). The spatial variability of these processes leads to the divergence of water quality of lakes (Lewis et al. 2000).

The natural dynamics of water quality in floodplain lakes have been greatly altered worldwide by land-use pressures. One of the major threats for these water bodies in large tropical and subtropical systems is the use of floodplains for cattle grazing (Lynch 2012; Junk 2013). Such land use has been promoted over the past few decades because of the conversion of traditional grazing areas into croplands (Bena et al. 2016). Cattle resuspend sediments when entering lakes, increasing turbidity and translocating nutrient to the water column through resuspension (Pettit et al. 2012). Moreover, animal waste is deposited directly in lakes or is carried into them through runoff or an increased water level (Mesa et al. 2015). In addition to its well-known contribution of ions, organic matter, N and P (Mesa et al. 2016; Park et al. 2019), cattle dung can be rich in silicon, an important nutrient for siliceous phytoplankton (Reynolds 2006; Komiyama et al. 2013). In turn, chlorophyll-a is expected to increase with nutrient inputs, whereas pH and dissolved oxygen are expected to decline owing to the increased availability of decomposing organic matter (Affonso et al. 2011; Masese et al. 2020).

Water quality of floodplain lakes deserves more attention in the current scenario of increasing anthropogenic pressure. This work aims to evaluate water-quality dynamics of floodplain lakes in relation to river flooding and rotational cattle grazing. The research area was located in the subtropical floodplain of the Middle Paraná River, one of the few in the world that still exhibits characteristics typical of a large free-flowing river. Surveys were conducted before and after a river flood that inundated the entire floodplain and prevented livestock from grazing in the area for several months. After the flood, cattle pressure differed among the studied lakes. We hypothetised that (1) the concentration of most water constituents decreases after floods through dilution and wash-out; (2) before floods, cattle presence produces an increase in nutrients, organic matter, chlorophyll-a and conductivity, but a decrease in dissolved oxygen saturation and pH levels in comparison with periods without cattle; whereas after floods, cattle effects on water quality are diluted by the increased water volume; and (3) differences in cattle pressure promote differences in water quality among lakes.

The Paraná River is located in central South America. It flows mainly from north to south along 3800 km, covering a basin area of 3.1 × 10⁶ km² (Fig. 1). The middle section of the river is placed in a subtropical region and extends from its confluence with the Paraguay River (27°29′S, 58°50′W) to the Argentinian city of Diamante (32°4′S, 60°38′W). The high sediment load together with the decrease in the general slope determines the development of an ~10–50-km-wide floodplain (13 000 km²) along the right bank of the Middle Paraná River. The floodplain contains a heterogeneous mosaic of different types of aquatic environments, including lakes with variable morphometry and connectivity to the main channel (Paira and Drago 2007). The hydrology of floodplain water bodies is closely linked to the hydrological regime of the river. River floods increase the depth and hydrological connectivity and largely decrease the water residence time of floodplain lakes. During isolation periods, water-column depth of lakes can be affected by fluctuations in the river level through underground seepage. Hydrological dynamics are highly irregular in timing and intensity owing to the variable intensity, duration and frequency of river floods (Neiff 1990; Drago 2007).

Fig. 1.  Location of the study area and sites. (a) The drainage basin of the Paraná River system. The figures on the right show the location of the sampling sites (L1, Lake 1; L2, Lake 2; and L3, Lake 3) and the paddocks where each lake was included (white lines) during (b) low-water conditions and (c) flood conditions, with the purpose of illustrating the surface water expansion and contraction on the floodplain.

Extensive cattle grazing to fatten up animals constitutes the preponderant anthropogenic activity in the floodplain of the Middle Paraná River (Sabattini and Lallana 2007). The floodplain is suitable for this activity during low water conditions because of the availability of valuable natural pastures from the foraging point of view and good-quality water for livestock supplying. The studied floodplain area is subject to a rotational grazing regime, with the purpose of maintaining healthier and more robust pastures (Mesa et al. 2015). This strategy involves a frequent movement of cattle through different fenced paddocks, which are alternatively grazed and rested. The animals have free access to the lakes located in the paddock being grazed. It is worth noting that cattle must be removed from the floodplain during river floods, when the area is unsuitable for grazing.

We studied the following three shallow lakes located on the floodplain of the Middle Paraná River: Lake 1 (31°41′02″S, 60°31′15″W), Lake 2 (31°40′46″S, 60°30′34″W) and Lake 3 (31°40′59″S, 60°30′44″W; Fig. 1). Each lake was located in a different paddock used for rotational grazing of cattle (stocking rate: 7.58, 10, and 4.01 head ha^–1 for the Lake 1, Lake 2 and Lake 3 paddocks respectively). Lakes became connected to each other and to the river during the river flood. However, lakes were hydrologically isolated during all the samplings, which were conducted before and after the river flood. Aquatic vegetation formed dense patches in all the lakes before the flood. It was scarce during the first months after the flood, but gradually re-established in the littoral zones during the rest of the study period. Emergent and floating macrophytes were the most important in terms of surface cover. Emergent plants were mainly represented by Ludwigia peploides and Saggitaria montevidensis in Lake 1, and by Myriophyllum aquaticum and Ludwigia peploides in both Lake 2 and Lake 3. Floating plants were mainly represented by Salvinia auriculata, Pontederia rotundifolia, Eichhornia crassipes, and Victoria cruziana (Lake 1), Azolla sp. and Victoria cruziana (Lake 2), and Azolla sp., Salvinia sp., Eichhornia crassipes and Victoria cruziana (Lake 3).

Samplings were conducted during morning hours between September 2015 and March 2017. Sampling times were selected in accordance with grazing rotation among paddocks to include situations with cattle presence and absence in each lake (Fig. 2a, b). Surveys were discontinued during the river flood that occurred between December 2015 and July 2016. The sampling design included pre-flood surveys (September–December 2015) and post-flood surveys (August 2016–March 2017) in the three studied lakes. We also discriminated an early post-flood period (August–November 2016) and a late post-flood period (December 2016–March 2017) on the basis of the start of the isolation phase (Fig. 2a, b).

Fig. 2.  (a) Water temperature and river hydrometric level, and (b) sampling dates indicated with black arrows. Floodplain lakes were isolated during all samplings, but were connected to each other and to the river during the river flood (which is highlighted). Grey and white segments indicate periods with cattle presence and absence respectively. The shorter-term sampling conducted in Lake 2 immediately after the reintroduction of cattle in the floodplain following flood recession is indicated with a dotted square.

Livestock were reintroduced in the study area (Lake 2 paddock) in October 2016, two and a half months after flood waters had receded. More frequent samplings were performed during October and November 2016 in this lake (Fig. 2a, b).

It is worth noting that cattle were rotated only between the Lake 2 and Lake 3 paddocks after the reintroduction of cattle in the floodplain following flood recession, unlike the pre-flood period when cattle were rotated among the three paddocks. Moreover, grazing periods were generally longer and rest periods shorter in the Lake 3 than in the Lake 2 paddock after the river flood. Therefore, Lake 1 (without grazing), Lake 2 (short grazing periods) and Lake 3 (long grazing periods) were subject to different cattle pressures during this post-flood period (Fig. 2a, b).

The variation of water temperature along the study period was estimated from air temperature following Drago (1984; Fig. 2a). The hydrometric level of the Paraná River was measured at the Paraná Harbour Gauge by Prefectura Naval Argentina and processed by Centro de Informaciones Meteorológicas (CIM-Universidad Nacional del Litoral; Fig. 2a). Isolation days of the lakes were calculated for each sampling as the number of consecutive days with a river hydrometric level lower than 3.9 m (connection or disconnection level estimated from our field observations). In addition, the depth of the sampling sites was measured during surveys with a folding rule.

Twelve ecologically relevant variables that have often been used to evaluate water quality were measured during each survey (Heinonen et al. 2000). These variables were selected because of their possible sensitivity to cattle grazing and hydrological fluctuations. Subsurface dissolved oxygen saturation, conductivity, and pH (HANNA portable checkers, Hanna Instruments, Buenos Aires, Argentina) were measured in situ. Subsurface water samples were collected in triplicate in the pelagic zone of the lakes and kept refrigerated during their transportation to the laboratory. Turbidity (formazin turbidity units, FTU) was spectrophotometrically estimated with a spectrophotometer HACH DR2000 (Hach Company, Loveland, CO, USA). A variable volume of water (100–200 mL) was filtered with a vacuum pump through Whatman GF/C glass-fibre filters, which were stored at –20°C for analysis of chlorophyll-a. Chlorophyll-a was spectrophotometrically estimated after extraction with acetone 90% within 3 weeks after sampling (American Public Health Association 2017). Filtered water samples were refrigerated until analysis of dissolved components within 24 h after sampling, whereas unfiltered subsamples were kept frozen until their analysis for total P and N determinations.

Soluble reactive P (SRP) was determined by the ascorbic acid method, nitrate + nitrite (NO₃^– + NO₂^–) by reduction of NO₃^– with hydrazine sulfate followed by diazotisation of NO₂^– with sulfanilamide, ammonium (NH₄⁺) by the indophenol blue method, and dissolved silica (SiO₂) by the molybdosilicate method. Water colour (platinum–cobalt (Pt–Co), mg L^–1) was measured at 455 nm and used as an estimator of chromophoric dissolved organic matter (CDOM). Total P was estimated by digestion with nitric and sulfuric acids, followed by SRP determination, and total N by alkaline digestion with potassium persulfate followed by NO₃^– + NO₂^– determination (American Public Health Association 2017).

Statistical analyses were conducted using the software CANOCO (ver. 5, Microcomputer Power, Ithaca, NY, USA; ter Braak and Šmilauer 2012) and SPSS (ver. 22.0, IBM Corporation, New York, NY, USA). Principal component analysis (PCA) with water temperature as covariate was used to explore the main variation patterns of the measured water-quality variables (dissolved oxygen saturation, electrical conductivity, pH, turbidity, SRP, NO₃^– + NO₂^–, NH₄⁺, SiO₂, total P, total N, CDOM and chlorophyll-a; Olsen et al. 2012). All variables were included in the analysis after being transformed (log(x+1); except pH), centred and standardised.

To test our first hypothesis, differences in water-quality variables among hydrological periods (Factor 1; pre-flood, early post-flood, and late post-flood periods) and among lakes (Factor 2; Lake 1, Lake 2 and Lake 3) as well as the interaction between both factors were analysed using two-way repeated-measures ANOVA (two-way RM-ANOVA). This allowed us to evaluate changes in water quality after the river flood, considering also differences among lakes (H₀: the mean values of water-quality variables are equal for all the hydrological periods and all the lakes, without an interaction effect between these factors). When the analysis resulted in a rejected null hypothesis, pairwise comparisons were performed with Tukey’s post hoc test. The relation between the river level and depth of the sampling sites was analysed through the Spearman test. This test was also used to analyse the relations of water-quality variables with depth of the sampling sites and isolation days.

To test our second hypothesis, samplings with cattle presence and absence were compared considering separately the pre- and post-flood periods. We searched for differences in each water-quality variable between samplings with and those without cattle during the pre-flood and during the post-flood periods using RM-ANOVA (H₀: the mean values of water-quality variables are equal for samplings with and without cattle). Lake 1 was excluded when analysing the post-flood period because it was not subject to grazing after the flood. These comparisons allowed us to analyse the influence of cattle presence on water quality before and after the flood.

In addition, we evaluated changes in water quality in response to cattle reintroduction in the study area following flood recession. We searched for atypical values of each water-quality variable in Lake 2 during the shorter-term sampling performed immediately after the post-flood cattle reintroduction in its paddock. Atypical values were identified as values more than 3 interquartile ranges (IQRs) above the third quartile or below the first quartile of the entire dataset through a box plot analysis (Helsel and Hirsch 2002).

Finally, to test our third hypothesis, comparisons among lakes on each water-quality variable were performed using analysis of covariance (ANCOVA). Because we aimed to evaluate the influence of differences in cattle pressure among lakes, we considered only the samplings performed during the post-flood period after cattle reintroduction in the floodplain (H₀: the mean values of water-quality variables are equal for lakes subject to different cattle pressures). Because isolation time strongly influences the water quality of floodplain lakes (Zalocar de Domitrovic 2003), its effect on each water-quality variable was controlled by using the previously calculated isolation days as a covariate. In this way, before comparing the lakes, we removed the variance of each variable in each lake that was explained by isolation days. This allowed us to assess more accurately the differences among lakes and make the test more powerful by reducing within-lake error variance (Huitema 2011). When ANCOVA test resulted in a rejected null hypothesis, pairwise comparisons among lakes were performed with a Bonferroni adjustment.

The first two principal components (PC1 and PC2) of the PCA made to explore the major water-quality patterns explained 59.34% of the total variability (Table 1). The PC1 separated water samples in accordance with the hydrological periods (Fig. 3). The pre-flood and late post-flood samples were clustered together, being associated with higher concentration of nutrients and CDOM evaluated as water colour, whereas the early post-flood samples were associated with a higher dissolved oxygen saturation and pH. By contrast, the PC2 separated water samples in accordance with the sampling sites. Lake 3 was associated with lower total P, total N, turbidity, and chlorophyll-a than for the Lake 1 and Lake 2.

Table 1.  Loadings of water-quality variables, eigenvalues and variance explained by the first two principal components (PC1 and PC2) of the principal component analyses made to explore the major water-quality patterns
Absolute loadings greater than 0.45 are highlighted in bold to indicate the water-quality variables that had more influence on each component. DO, dissolved oxygen; SRP, soluble reactive P

Fig. 3.  Biplot of the first two principal components (PC1 and PC2) of the principal component analysis, showing the main variation patterns of water quality considering Lake 1 (squares), Lake 2 (circles), and Lake 3 (triangles) during the pre-flood (black symbols), early post-flood (white symbols) and late post-flood (grey symbols) periods. The PC1 and PC2 separated the water samples according to the hydrological periods and the sampling sites respectively. The measured water-quality variables are represented with arrows. The more parallel to a principal component axis is an arrow, the more it contributes to that component (see Table 1 for detailed information about variable loads on each component). Arrow length indicates how much variability of the corresponding variable is represented by the two components, whereas angles between arrows of different variables approximate their correlation. SRP, soluble reactive P.

Differences among the pre-flood, early post-flood, and late post-flood periods were significant for pH (F = 5.89), dissolved oxygen saturation (F = 15.56), total P (F = 5.50), SRP (F = 16.03), NH₄⁺ (F = 10.19), chlorophyll-a (F = 6.74) and CDOM evaluated as water colour (F = 3.55; two-way RM-ANOVA: P < 0.05, between-group d.f. = 2, within group d.f. = 26). By contrast, the differences among lakes and the interaction between lakes and hydrological periods were not significant (two-way RM-ANOVA: P > 0.05).

Dissolved oxygen saturation and chlorophyll-a increased whereas nutrients decreased after the flood. Dissolved oxygen saturation and total P returned to pre-flood values during the late post-flood period (Tukey’s post hoc test, P < 0.05, Fig. 4). By contrast, NH₄⁺ and SRP remained at low levels, whereas chlorophyll-a maintained an ascending trend until the end of the study period. Finally, CDOM and pH showed similar values during the pre-flood and early post-flood periods. However, CDOM increased whereas pH decreased significantly during the late post-flood period (Tukey’s post hoc test, P < 0.05, Fig. 4).

Fig. 4.  Values of water-quality variables in Lake 1 (squares), Lake 2 (circles) and Lake 3 (triangles) during the pre-flood (black symbols), early post-flood (white symbols), and late post-flood (grey symbols) periods. The horizontal and vertical lines show the median and interquartile ranges respectively. Different letters represent significant differences between periods (Factor 1) according to two-way RM-ANOVA (P < 0.05). Neither differences among lakes (Factor 2) nor the interaction between both factors was significant (P > 0.05). The width of the distribution of points is proportionate to the number of points at each y-value. SRP, soluble reactive P; FTU, formazin turbidity units.

In accordance, turbidity, total P, and CDOM evaluated as water colour were positively related to isolation days; whereas dissolved oxygen saturation and pH showed the opposite trend (Spearman test, P < 0.05, Table 2). The depth of the sampling sites (range: 0.4–1.0 m) showed the minimum values before the flood and varied positively with the river hydrometric level (Spearman test, rho = 0.45, P < 0.05). It was also related to water-quality variables. In this regard, turbidity, NH₄⁺ and total N increased significantly as the depth decreased (Spearman test, P < 0.05, Table 2).

Table 2.  Spearman correlation coefficients of depth of the samplings sites and isolation days with water-quality variables
The three studied floodplain lakes were considered together. Only significant (P < 0.05) correlations are shown. Numbers in bold indicate P < 0.01

There were no significant differences in water-quality variables between periods with cattle presence and absence before the flood (RM-ANOVA, P > 0.05). However, Lake 2 and Lake 3 experienced significant changes in water quality in relation to cattle rotation after the flood; periods with cattle presence showed significantly higher values of SRP (F = 14.37), total P (F = 29.04) and turbidity (F = 7.65) than did periods with cattle absence (RM-ANOVA, P < 0.05, between-group d.f. = 1, within-group d.f. = 10), but there were not significant differences for the other evaluated variables (P > 0.05) (Fig. 5).

Fig. 5.  Mean values (bars) and standard errors (vertical lines) of soluble reactive P (SRP), total P (TP) and turbidity (formazin turbidity units, FTU) during periods with cattle absence (dotted bars) and presence (striped bars) before and after the flood. Asterisks indicate significant differences between periods with and without cattle (RM-ANOVA, P < 0.05). Only variables that varied significantly in relation to cattle rotation are shown.

In relation to the shorter-term sampling conducted in Lake 2 immediately after the resumption of grazing following flood recession, atypically high values of NO₃^– + NO₂^– and NH₄⁺ (i.e. outliers more than 3 IQRs above the third quartile of the entire dataset) were observed. In this regard, concentrations of NH₄⁺ and NO₃^– + NO₂^– were always below 50 and 100 μg L^–1 respectively, in all the lakes during the rest of the study period. However, NH₄⁺ showed positive outliers in Lake 2 during two consecutive samplings conducted at the 3rd and 9th days after cattle reintroduction in the floodplain (306 and 441 µg N L^–1 respectively). Furthermore, a positive outlier of NO₃^– + NO₂^– (305 µg N L^–1) was detected on the 9th day. After that, concentrations of both N forms returned to values typical for these environments.

ANCOVA showed significant differences in conductivity (F = 6.82), total P (F = 5.60), dissolved oxygen saturation (F = 5.51) (P < 0.05) and turbidity (F = 7.50, P < 0.01, between-group d.f. = 2, within-group d.f. = 13) among lakes during the post-flood period after controlling the effect of isolation days. However, differences in water quality were contrary to our expectations according to differences in cattle pressure. The lake without grazing (Lake 1) showed the lowest dissolved oxygen saturation; whereas the lake subject to the longest grazing period (Lake 3) showed the lowest values of total P, turbidity and electrical conductivity (Fig. 6). In addition, as shown in Fig. 4, peaks of chlorophyll-a occurred in Lake 1 and Lake 2 (124 and 357 μg L^–1 respectively), whereas values were relatively low in Lake 3 through the entire post-flood period, although without significant differences among lakes.

Fig. 6.  Mean values (bars) and standard errors (vertical lines) of water-quality variables in lakes subject to different grazing intensities (Lake 1: without grazing; Lake 2: short grazing periods; Lake 3: long grazing periods) after cattle reintroduction in the floodplain. Different letters represent significant differences between lakes according to ANCOVA and pairwise comparisons with a Bonferroni adjustment after controlling the influence of isolation days. Only variables that differed significantly (P < 0.05) are shown. FTU, formazin turbidity units.

As expected, the dilution and washing-out determined by the river flood contributed to the decrease in nutrient concentrations. However, other water constituents such as chlorophyll-a increased after the flood. These effects were independent of the lakes, highlighting that the hydrological regime is the main regulating factor of water quality in floodplain lakes (Hamilton and Lewis 1990; Mayora et al. 2013, Castillo 2020), even in those subject to cattle grazing.

Particularly, we observed a decline in concentrations of total P, SRP and NH₄⁺ after the flood, suggesting a negative balance between nutrient inputs from the river and rewetted floodplain areas and washing-out effects (Villar and Bonetto 2000; Lizotte et al. 2012; Mayora et al. 2017). Dilution by the increased water volume (Weilhoefer et al. 2008) and interruption of cattle grazing (Mesa et al. 2015) could also have favoured the decrease in nutrient concentrations. However, chlorophyll-a increased during the post-flood periods probably because of the wash out of macrophytes, which usually occurs in response to large floods and is evidenced by a reduction in vegetation cover (Maine et al. 2004). Macrophytes negatively affect phytoplankton biomass by shading the water column, providing shelter for grazers, competing for nutrients, and producing allelopathic compounds (Sinistro et al. 2006; Cunha et al. 2012).

Moreover, a greater oxygenation was observed after the flood, probably because of the contribution of oxygen from phytoplankton photosynthesis and the input of well oxygenated water from the river (Maine et al. 2004). In addition, the wash out of macrophytes could have directly favoured water oxygenation because dense patches of emergent and floating vegetation can deplete oxygen by reducing the turbulence and the water surface available for gas exchange (Caraco et al. 2006). This could have important implications for the biogeochemical functioning of floodplain lakes because oxidation–reduction reactions mediate many transformations of organic matter and nutrients (Depetris and Pasquini 2007).

Ammonium concentration remained low over the entire post-flood period. It was not related to isolation days, but it had a negative relationship with depth, similarly to total N and turbidity. This would be related to the enhanced resuspension of sediments and associated nutrients through wind action, a process that is intensified by the shallowness (Lewis et al. 2000; Maine et al. 2004). Therefore, as claimed by Pettit et al. (2012), shallower conditions could intensify the effects of cattle grazing on water quality in floodplain lakes. In turn, we observed a positive relationship between the depth of the sampling sites in lakes and the river hydrometric level, suggesting that the former was affected by the river hydrology through underground seepage (Drago 2007). In this regard, the river level could affect the water quality of lakes during isolation periods through effects on lake depth and associated processes.

By contrast, total P returned to pre-connection levels during the late post-flood period. Unlike N, P does not have a permanent removal mechanism in isolated water bodies and is therefore more susceptible to accumulate with time (Cheng and Basu 2017). CDOM values also increased through the late post-flood period, probably owing to the production of organic debris by aquatic macrophytes (Maine et al. 2004). Bacterial mineralisation of CDOM consumes oxygen and releases CO₂, which can explain the observed decrease in both dissolved oxygen saturation and pH with time since isolation started (Zuijdgeest et al. 2016).

During samplings, cattle were observed drinking and grazing in different lakes and riparian zones, depending on rotation among paddocks. In addition, the grazed zones showed signs of cattle activity, i.e. hoof marks, trails and waste presence (G. Mayora, unpubl. data).

Urine and dung can increase nutrient levels because they are rich in N, whereas dung is also rich in P (Food and Agriculture Organization of the United Nations 2017). In addition, cattle activity can affect water quality by resuspending sediments and associated nutrients (Pettit et al. 2012; Mesa et al. 2015). In this regard, SRP, total P and turbidity were significantly increased during periods with cattle, without there being significant changes in the other water-quality variables. Furthermore, although no persistent differences in N forms were observed between periods with and those without cattle, dissolved inorganic N (NH₄⁺ and NO₃^– + NO₂^–) increased sharply immediately after the resumption of grazing in the floodplain (Lake 2). This increase was very short-lasting, reinforcing the idea that efficient removal of dissolved inorganic N is a common feature of floodplain lakes (Tockner et al. 1999; Villar and Bonetto 2000; Mayora et al. 2017; Castillo 2020).

However, the above-mentioned significant variations in water quality in relation to rotational cattle grazing occurred after, but not before, the flood. This is contrary to our second hypothesis based on the dilution capacity of river floods, and can be explained by different factors. On one hand, the depth of the sampling sites before the flood had declined to such an extent (minimum value: 0.4 m) that bottom sediments and associated nutrients could have been resuspended by wind-generated waves, obscuring the influence of cattle presence on these variables (Lewis et al. 2000). On the other hand, the general lower concentrations of P after the flood may have decreased the P saturation level of bottom sediments, which play a key role in P retention in aquatic environments (Fisher and Acreman 2004). This could have enhanced P removal during periods without sediment disturbance by cattle action, increasing differences between periods with and those without cattle in comparison to the pre-flood period characterised by higher concentrations of P. In this regard, the rotational management showed to be a proper grazing practice to favour nutrient removal from the water column, as suggested by Mesa et al. (2015), whereas the hydrological connectivity of lakes to the river seemed to be essential for maintaining the removal capacity.

It is worth noting that all of the lakes had been subject to cattle grazing for several months at the beginning of the study period (L. Mesa, unpubl. data). Values of total P and N observed before the flood (respectively up to 835 and 6100 µg L^–1 in Lake 1) are among the highest observed for lakes of the Paraná floodplain according to our knowledge (e.g. Villar and Bonetto 2000; Maine et al. 2004; Mesa et al. 2015; Mayora et al. 2017, 2018). Moreover, isolated lakes of the Paraná floodplain are characterised by very low levels of dissolved inorganic N. Peaks of NH₄⁺ and NO₃^– + NO₂^– observed during the first days after reintroduction of cattle in the floodplain were double the highest concentrations registered previously in isolated lakes of this system (e.g. Villar and Bonetto 2000; Maine et al. 2004; Mesa et al. 2015; Mayora et al. 2017).

Contrary to our third hypothesis, factors other than differences in cattle pressure seemed to control the post-flood variability in water quality among lakes. The observed differences in water quality among lakes can be attributed to the fact that Lake 3 is larger than are the other ones. Larger lakes are less influenced by margins and experience lower inputs of allochthonous materials in relation to their volume (Rocha et al. 2009). Therefore, despite being subject to the longest grazing period, Lake 3 had a higher capacity for diluting materials entering from surrounding areas, which may have prevented eutrophication (Lewis et al. 2000; Liu et al. 2011). In this regard, cattle effects could be intensified in smaller lakes subject to long grazing periods. However, not only the size of the lakes but also the proximity of cattle to them could affect water quality. Future studies should consider monitoring livestock access to lakes and surrounding areas, which can provide more relevant information than grazing duration and stocking rate in the entire paddock and increase our insight into grazing effects on water quality.

Our results showed increased concentrations of nutrients and values of turbidity in floodplain lakes of the Middle Paraná River in relation to cattle presence. However, periodic river floods can revert the processes of nutrient enrichment and improve the intrinsic capacity of floodplain lakes for self-purification. This highlights the importance of maintaining the river–floodplain connectivity to attenuate the effects of livestock on water quality. Finally, dimensions of lakes should be considered in management strategies of floodplain grazing. Paddocks should be preferentially located in areas containing larger lakes with a greater dilution capacity to preserve water quality, mainly during shallower periods.

The authors declare that they have no conflicts of interest.

We are grateful to Consejo Nacional de Investigaciones Científicas y Técnicas (CONICET) and Agencia Nacional de Promoción Científica y Tecnológica (ANPCyT) (PICT 2012-0906 and 2018-2203 PI: Mesa L., PICT 2016-2890 PI: Mayora G., PICT 2017-2982 PI: Quintana R.) for the financial support.