Introduction

Biotic and abiotic factors influence species distributions (Grinnell 1917) limiting which organisms can coexist in a particular space and time, according to their ecological requirements (Chesson and Warner 1981). Therefore, changes in these conditions alter the species composition. However, the variation in both the environment and biological communities is unlikely to respond to only a single process and may be influenced by the interactions among ecological processes that act at different scales (Vasseur et al. 2014), as expected for complex systems (Angeler et al. 2011). As a corollary, several processes acting simultaneously at different temporal scales trigger environmental changes and drive the variation in attributes of communities, such as diversity and species composition (Cheng et al. 2013).

On floodplains, for example, the flood pulse occurs every year except in extreme hydrologic events or due to anthropogenic activities (e.g. dam construction), and it is the main source of variation for the structure of biological communities (Junk et al. 1989). Flood pulses cause significant variations in the water level, which influence the dynamics of energy and materials in the system. On a broader temporal scale (i.e., years), climate phenomena such as the El Niño-Southern Oscillation (ENSO) may cause rainfall anomalies that affect the hydrological regime (Neiff 1990). As a result, for instance, subtropical floodplains can have longer periods of high water (the ENSO warm phase, El Niño) and low water (the ENSO cold phase, La Niña) that alternate with periods not affected by ENSO (normal periods). These anomalous flood and dry periods strongly affect the distribution of aquatic communities (Agostinho et al. 2004; Bortolini et al. 2016; Bovo-Scomparin and Train 2008; Simões et al. 2012, 2013).

Shifts in communities across space can be synchronized (temporal concordance) with environmental variability over time (Kent et al. 2007; Zhang et al. 2018) resulting in annual patterns of succession (Anneville et al. 2002). Nevertheless, temporal shifts in communities can also occur decoupled from underlying environmental conditions, because of a delayed response to environment or as a result of processes operating at multiple temporal scales (Hastings et al. 2018). That decoupled variation can occur through community stability in species richness (De Boeck et al. 2018; Gotelli et al. 2017), or when the increase in the species replacement results mainly from the colonization-extinction dynamics (Angeler and Johnson 2012; Fukami et al. 2005), where changes in composition are related mainly to temporal or spatial factors (Leibold and Chase 2018b). Therefore, if processes operate at different scales, it is essential to identify at which temporal scales patterns of diversity emerge. This question becomes important in a future where climate events could vary in frequency and duration, and where it is unclear how diversity can be affected by climate extremes (De Boeck et al. 2018; Pecl et al. 2017).

Temporal patterns in species composition can be disentangled by investigating the components of diversity across temporal scales. The γ diversity (sample collection in space or time) can be decomposed into within samples diversity (α) and the variation in diversity among them (β), with the partitioning of gamma diversity (Lande 1996). As samples accumulate over space or time, it becomes possible to determine the contributions from α and β across multiple scales (e.g., at flood-pulse and ENSO scales) in a hierarchical design (β1, β2, etc.) (Lande 1996; MacArthur 1965; Crist et al. 2003). Although several studies have analyzed the variations in diversity at several spatial scales by using such a hierarchical design (Beck et al. 2012; Braghin et al. 2016; Chaparro et al. 2018; Dittrich et al. 2016), few studies have included the variation of diversity at temporal scales (Fukami et al. 2005; Simões et al. 2013; Thomas et al. 2018).

Certain environmental processes acting at different temporal scales, such as the annual flood pulse or ENSO climate events are relatively easy to delimit on floodplains (Bovo-Scomparin and Train 2008; Souza Filho and Stevaux 1997). Hence, floodplains are suitable models to analyze the effects of temporal processes on communities. A hierarchical analysis such as the additive partitioning of diversity (Lande 1996) could be a useful alternative to identify the importance of different temporal scales, such as the flood-pulse scale or the ENSO climate-event scale, on the variation of diversity. Temporal analysis is a growing area of investigation, fueled by the increasing availability of data series. Data series from different environments and from organisms belonging to several trophic levels are necessary to advance the understanding of temporal patterns and temporal dynamics (Franklin 1989).

To understand the influence of different temporal scales on diversity, we analyzed a 16-year time-series dataset of the composition of zooplankton, phytoplankton, and ciliates in shallow floodplain lakes. The variation in community structure (species composition) was partitioned according to a series of additive hierarchical temporal scales: the sample (α), intra-season variation (β1, between months within a season), variation between flood seasons (β2, high-water and low-water seasons), and variation between ENSO-related events and periods not affected by ENSO (β3—La Niña, El Niño and normal periods). Because the seasonal processes (e.g. flood pulse) are suggested to be the main temporal factors structuring communities in floodplain environments (Junk et al. 1989; Zhang et al. 2018), we compared the influence of the flood seasons (β2) with processes acting at finer (α and β1) and broader (β3) temporal scales. We tested the hypothesis that β2 is the most important source of variation for the gamma diversity of all the plankton communities. Hence, we expected the proportion explained by β2 to be higher than the other gamma components. We also investigated the influence of environmental and temporal factors on composition and temporal turnover of plankton communities.

Methods

Study area

We conducted this study on the Upper Paraná River floodplain in southern Brazil. The region has a subtropical climate with mean monthly temperatures above 15° C and precipitation above 1500 mm yr−1. The hydrological regime is characterized by a high-water season (October–February) and a low-water season (June–September). However, the frequency, duration, and intensity of the floods have changed due to the construction of several dams upstream in the main channel (Souza Filho et al. 2004). The samples were collected in two lakes (Fig. 1). Patos Lake (22°49′33.66″ S; 53°33′09.9″ W) is permanently connected to the Ivinhema River (an important tributary of the Paraná River) and has an area of ca. 114 ha and a mean depth of 3.5 m. The Ventura Lake (22°51′36.41″ S; 53°36′4.89″ W) is located 200 m from the Ivinhema River. It has an area of ca. 89.8 ha and a mean depth of 2.5 m. Typically, water from the Paraná River reaches these lakes when the water level rises above 4.5 m, while the Ivinhema River reaches the Ventura Lake when the water level rises above 2.75 m (Souza Filho 2009). The selection of these two lakes to test our hypotheses allowed us to determine whether the effect of the different temporal scales on plankton communities was consistent for two lakes with different characteristics, likely indicating the synchronous response of communities in different habitats to extrinsic environmental factors (Kent et al. 2007).

Fig. 1
figure 1

Location of the sampling sites on the Paraná River floodplain

Full size image

Sampling and data analyses

Plankton community

Samples were collected at the subsurface (20 cm deep) of the limnetic zone, for all communities and limnological variables. Zooplankton (n = 60) and phytoplankton (n = 60) samples were collected quarterly from 2000 through 2015 (except in 2001 and 2003, when only two samplings were conducted). Ciliates (n = 12) were collected twice a year from 2010 through 2015.

Phytoplankton samples were taken directly with bottles and preserved with 1% acetic Lugol solution. For each sample, at least 400 individuals (cells, colonies, and filaments) were counted in random fields under an inverted microscope (Lund et al. 1958; Utermöhl 1958) and expressed as ind.mL−1. Phytoplankton biomass was estimated from the biovolume (mm3 L−), by multiplying the density of each species by its respective volume. The volume of each cell type was calculated from the approximate geometric shape (Sun and Liu 2003). Zooplankton samples were taken from the subsurface, using a motorized pump to filter 600 L of water through a plankton net (68 µm), and fixed with 4% formaldehyde buffered with calcium carbonate. We identified testate amoebae, rotifers, cladocerans, and copepods at the species level (Deflandre 1929; Elmoor-Loureiro 1997; Gauthier-Lièvre and Thomas 1958; Koste 1978; Matsumura-Tundisi 1986; Reid 1985; Segers 1995), mounting some on slides and coverslips under an optical microscope at magnifications of 40 to 400 × , depending on the taxonomic group. Zooplankton was quantified (individuals.m−3) through subsampling with a Hensen-Stempel pipette and counting at least 10% of the concentrated sample (at least 80 individuals) in Sedgwick-Rafter chambers (Bottrell et al. 1976). For ciliates, water samples were collected using 4-L polyethylene flasks. In the laboratory, samples were concentrated to 100 ml using a plankton net (10 µm) and immediately counted and identified in vivo within 5 h after sampling, following the live-counting technique proposed by Madoni (1984). Ciliate abundance was expressed as individuals.L−1. For all three communities, the number of taxa present in each quantitative sample was considered as the species richness (alpha diversity). Gamma diversity was considered as the total taxa recorded in each lake during the entire study period (i.e. the total of samples). Note that for each plankton community included in this study, the same team (laboratories of Phytoplankton, Zooplankton and Ciliates of the Núcleo de Pesquisas em Limnologia, Ictiologia e Aquicultura—Nupélia) collected and identified the organisms. Thus, the possible temporal variation of the plankton communities is not attributed to differences in the expertise of researchers who identified the samples.

Environmental variables

Water temperature (WT,  °C), pH, dissolved oxygen (DO, mg L−1), conductivity at 25 °C (Cond., µS cm−1), phosphate (PO4, µg L−1), nitrite (N–NO2−, µg L−1), nitrate (N–NO3−, µg L−1), ammonium (N–NH4+, µg L−1), suspended organic matter (SOM, mg L−1), and alkalinity (mEq L−1) were determined following the methods described in APHA (2005). The sum of the nitrate, nitrite, and ammonia was considered as the dissolved inorganic nitrogen (DIN, µg L−1). The maximum depth (Zmax, m) was measured at each sampling site. The depth of the euphotic zone (Zeu, m) was calculated as 2.7 times the Secchi depth (Cole 1994). The ratio between the euphotic zone and water depth (Zeu:Zmax) was used as a measure of light availability in the water column. Precipitation and water levels in the Paraná and Ivinhema rivers were provided by the Agência Nacional de Águas (ANA) and Itaipu Binacional. The cumulative precipitation and the mean water level 5 days before sampling were used as precipitation and water-level measurements, respectively.

Data analysis

A constrained analysis of principal coordinates (CAP, Anderson and Willis 2003) was performed to characterize the temporal variation in environmental conditions, and to test for environmental differences related to seasons and ENSO events. CAP was performed using Euclidean distance for scaled variables within each lake.

To compare species richness within communities in the two lakes we constructed species accumulation curves. The curve was estimated by the method of data random permutation with subsamples without substitution (Gotelli and Colwell 2001). In addition, to evaluate the representativeness of our sampling for the richness of each community in the lakes we used nonparametric estimators of diversity (Chao and Jakknife 1). Samplings that took place between October and March were considered as high-water season samples, and samplings between May and September as samples of the low-water season. The samplings were also categorized as periods influenced by La Niña (n = 16), El Niño (n = 14) and normal periods (n = 30) (see Table S1) using the Southern Oscillation Index (SOI; https://www.ncdc.noaa.gov/teleconnections/enso/indicators/soi/data.csv) to determine the years with more influence of each event. Then, we performed additive partitioning of diversity (species richness) to verify the contribution of alpha (α) and beta (β) diversity to gamma diversity (Crist et al. 2003). For all communities we considered the within sample diversity (α), the intra-season variation (β1), variation among low- and high-water seasons (β2), and the variation among climate events (β3) (Fig. 2). In all cases, the total diversity (γ) was obtained by the sum of the average number of species within samples (α) and among samples (β). As all three components are measured at the same dimension (number of species), the derived beta diversity measured the effective number of species in a pool of samples not contained in an average community (Chao et al. 2012).

Fig. 2
figure 2

Hierarchical design of diversity partitioning of total diversity (γ). α = sample diversity. β1 = intra-season variation. β2 = variation between flood seasons. β3 = diversity between ENSO climate events and normal climate events. α + β1 + β2 + β3 = γ

Full size image

The statistical significance of each component of diversity in explaining variation was tested through 999 randomizations according to a null model, in which samples were randomly permuted among hierarchical levels (across temporal scales). Thus, the null hypothesis states that the components of gamma diversity can result from the random allocation of lower-level samples into higher-level samples (Crist et al. 2003). When p values were less than 5%, the observed component of the diversity was considered significantly different from the null expectation.

For each lake, partial redundancy analysis (pRDA) was used to evaluate the relative importance of temporal and environmental predictors for the variation in species composition and directional change in species composition (temporal turnover) of phytoplankton, zooplankton, and ciliates (Baselga 2010). pRDA is a useful tool to determine the role of niche-associated environmental processes and stochastic processes associated with time, that are not functionally related to changes in the environmental conditions (Legendre and Gauthier 2014), although it is limited in distinguishing among stochastic processes (e.g. ecological drift versus stochastic extinction) (Smith and Lundholm 2010). We interpreted the significance of the environmental factors as a signal of niche processes affecting community structure, and the significance of temporal predictors as a signal of stochastic processes.

In the pRDA, for each community, we used a presence-absence species matrix as response for variation partition of species composition. Prior to the analyses, the composition matrices were Hellinger-transformed as recommended for matrices with many zeros and analyzed by linear methods such as pRDA (Legendre and Gallagher 2001). For partitioning of the temporal turnover, we used a matrix including the significant PCoA vectors calculated using Sorensen dissimilarity and Lingoes correction for negative eigenvalues (Legendre 2014).

As temporal predictors, we used a matrix constructed based on asymmetric eigenvector maps (AEM) (Blanchet et al. 2008a). AEM considers time linearity; it is suitable to account for temporal variation because the time-associated processes are directional (Legendre and Gauthier 2014), that is, the variation in a specific time period is influenced by a previous state. Eigenvector analyses such as AEM allow modeling temporal patterns at different scales, for example, from days (fine temporal scale) to years (broad temporal scale) (Borcard et al. 2011). The AEM calculation uses a matrix with the distances between sampling units, a matrix giving direction to the connectivity among them, and a matrix of coordinates giving the location in space or time of sampling units (Blanchet 2009). In our case, we used the number of months separating the samplings as the distance between sample units. The matrix of connectivity was based on the time directionality. Finally, we created a coordinates matrix including the temporal sequence of the samplings and simulating a straight line to represent the time linearity (see Fig. S1 for a representation of the matrices used in the AEM calculation).

For ciliates the number of variables exceeded the number of observations, hence the environmental factors were reduced in dimensionality through principal component analysis (PCA). The number of PCA axes explaining significant variance for each lake was determined through a randomization-based procedure (Ter Braak 1988). Environmental variables were randomized (1000 times) and each time a significance test for the eigenvalues was done. Thus, for phytoplankton and zooplankton five PCA axes (65.8% of variance) were retained in the open lake and four PCA axes in the isolated lake (59.9% of variance). For ciliates one axis was retained in each lake (37.5% and 31.2% of explained variance for the connected and isolated lake, respectively). The function ‘rndLambdaF’ from the ‘PCDimension’ (Coombes and Wang 2018) was used to determine the number of significant PCA axes.

In addition, we considered a biotic component consisting of PCoA vectors generated using Bray–Curtis dissimilarity for each community. Broken stick criterion was used to select PCoA vectors. That is, the PCoA vectors with eigenvalues larger than expected for the random partition of a stick of length 1 were retained. Thus, zooplankton PCoA vectors (26 vectors, 83–85% of variance) were used to account for the effect of grazing on ciliates and phytoplankton. Likewise, phytoplankton PCoA vectors (22 vectors, 73–80% of variance) accounted for the effect of resource availability for zooplankton and ciliates in the pRDA. As ciliates had a smaller number of samples, only samplings of phytoplankton and zooplankton corresponding to ciliates samplings were used to performed PCoA ordinations (2–3 PCoA vectors, 62–64% of variance), and the biotic component for ciliates.

We performed a forward-selection (Blanchet et al. 2008b) procedure of abiotic environmental variables, biotic PCoA vectors and temporal variables based on AEM to select the set of variables significantly related to each community to be used in pRDA analysis. The forward selection method considers significance (p < 0.05; 999 permutations) and adjusted R2 as stopping criteria: first, significance is tested in a global model with the entire set of variables, also calculating a global adjusted R2. If there is no significant relationship, the selection is stopped. Then, variables are included successively. When a variable leads to an adjusted R2 greater than the global adjusted R2, the analysis is stopped and the variable excluded. We used adjusted R2 values as a result for variation partitioning, since these values are not affected by the number of predictive variables and make the results comparable (Peres-Neto et al. 2006). The significance of the components was tested at p < 0.05 with 999 permutations. To evaluate the effects of the environment and time on the communities, we retained the explanation of each pure fraction for interpretation. To verify if the relationship among explanatory variables and the composition and turnover of phytoplankton and zooplankton would be maintained with a temporal window similar to the ciliates (samplings twice a year for 6 years, n = 12), we ran additional pRDAs using 10 samples (two by year) from 2010 to 2015 for each community.

Finally, Procrustes tests were performed to verify if the relationship among composition and environmental and temporal predictors was concordant among communities and between lakes. The analysis was carried out using the pRDA ordinations for each community and 9999 permutations to evaluate the significance of the correlation between ordinations. All analyses were performed using R software (R Development Core Team 2018). To calculate the asymmetric eigenvector maps we used the package “AEM” (Blanchet 2009), to obtain PCoA vectors we used the package “ape” (Paradis et al. 2004) and for the remaining analyses we used the package “vegan” (Oksanen et al. 2012).

Results

Environmental characterization

The two lakes showed high temporal variation (variation coefficient > 50%) in nutrient concentrations (DIN, PO4), suspended organic matter (SOM), light availability (Zeu, Zeu:Zmax), and euphotic zone (Table 1). For both lakes, CAP ordination showed significant differences among flood seasons (connected lake F = 2.33, P = 0.02; isolated lake F = 2.20, P < 0.01) and ENSO events (connected lake F = 1.75, P = 0.03; isolated lake F = 2.80, P < 0.01). Water level (Iv.WL) and variables related to light availability were the main factors contributing to variability among climate events (Fig. 3 a, b).

Table 1 Mean values, standard deviation (SD), and coefficient of variation (CV) of abiotic and biotic variables measured from 2000 to 2015 in the connected (n = 60) and isolated (n = 60) lakes
Full size table
Fig. 3
figure 3

Constrained analysis of principal coordinates (CAP) showing the variation related to flood seasons and Enso events in a connected (a) and an isolated lake (b) on the Upper Paraná River floodplain from 2000 to 2016. Only the variables significantly different among climate events are displayed

Full size image

Plankton communities

In total, 443 taxa of phytoplankton, 353 of zooplankton, and 78 of ciliates were recorded in these lakes. The connected lake showed higher gamma diversity (368 phytoplankton, 295 zooplankton, and 61 ciliate taxa) than the isolated lake (288 taxa phytoplankton, 268 zooplankton, and 51 ciliate taxa) for all three communities (Fig. 4). The accumulation curve did not reach an asymptote, indicating that the totality of species was not sampled in any of the communities despite the extensive temporal sampling design. The observed richness values represented between 60 and 76% of the expected species. The contribution of unique (i.e. species occurring in only one sample) and duplicates (i.e. species occurring in just two samples) to richness varied between 44 and 70% (Table S2).

Fig. 4
figure 4

Species accumulation curve for phytoplankton (a), zooplankton (b), and ciliates (c) samples, in the Patos (black) and Ventura (white) lakes

Full size image

All communities showed moderate to high temporal turnover (phytoplankton = 0.58; zooplankton = 0.55; ciliates = 0.68), with the connected lake (0.62 ± 0.14) showing significantly higher turnover (F = 28.48; P < 0.01 considering all communities) than the isolated lake (0.53 ± 0.14). Procrustes test showed that the variation in composition for taxonomic groups across lakes was concordant for all three communities. Within lakes, phytoplankton variation was correlated with the other two communities in both lakes (Table S3).

Importance of α and β components to the γ diversity of plankton communities

The additive partitioning revealed that larger temporal scales (flood seasons β2 and ENSO events β3) had the higher contribution to the gamma diversity for all three communities in both lakes (Fig. 5; Fig. S2). This result partially supports our first hypothesis (higher importance of component β2, variation between low- and high-water seasons), because differences in species richness among flood seasons were lower than expected by chance (except for the random pattern for ciliates in the isolated lake). Conversely, among climate events (β3), the observed differences in richness were higher than under the null expectation (see details in Table S4). The α component (sample level) was significantly lower than expected (p < 0.05) for all three communities.

Fig. 5
figure 5

Additive partitioning of total species richness of phytoplankton, zooplankton, and ciliates recorded in the Patos and Ventura lakes. The values of α and β components are expressed as percentage, and total diversity γ as species number. Asterisks represent components with a distribution different from the null model (p < 0.05). For detailed results, see Table S4

Full size image

Predictors of plankton communities

Regarding environmental variables, only abiotic variables (E component) were selected for ciliates, whereas for zooplankton and phytoplankton both abiotic and biotic variables were selected (E and B components) (Table 2). For temporal variables, the forward selection procedure selected 16 AEMs from the original 58 AEMs (Fig. S3). The selected AEMs were related to broader (e.g. AEM 1: complete sampled period; AEM 2: ENSO events) and to finer (e.g. AEM 25: flood seasons) temporal scales for the zooplankton and phytoplankton communities, and mostly broader scales for ciliates (Table 2). The pRDA showed that mainly the abiotic environmental (E) and temporal factors (T) considered in our analyses influenced both variation in composition and species turnover of phytoplankton and zooplankton communities, with a higher explanatory power for temporal turnover than for variation in species composition (Fig. 6). The biotic component (B) was significant for zooplankton composition at the isolated lake, and for temporal turnover of phytoplankton and zooplankton at the isolated lake (Fig. 6).

Table 2 Environmental factors and AEMs selected by the forward selection method and used in the partition of variation in composition and temporal turnover for plankton communities in a connected and an isolated lake
Full size table
Fig. 6
figure 6

Relative importance (% of explanation) of the environment (E), biotic (B), temporal vectors (T), and shared components for the composition and turnover of plankton communities in the Patos (connected) and Ventura (isolated) lakes. Significant values are indicated with an asterisk, zeros indicate values lower than 0.5%, and values < 0 are not shown. The significance of the shared components is not testable. R represents the residuals of the analysis

Full size image

For phytoplankton, abiotic environmental and temporal factors were always important, but E showed higher explanation in connected and T higher explanation in isolated lakes, for both composition and turnover (Fig. 6). For zooplankton communities, T showed higher explanation in both lakes for both composition and turnover (Fig. 6). Temporal predictors related in different ways to changes in composition (Fig. 6). In connected lakes, phytoplankton and zooplankton showed higher correlation with small-scale AEMs, whereas in isolated lakes the relationship was stronger with large-scale temporal factors (Fig. S4). Ciliates seem to be related to broader temporal scales in both lakes (Fig. S4). Further, while phytoplankton and zooplankton tended to have a positive relationship between species turnover and time, ciliates display a negative correlation (Fig. S5). Note that the first AEM variable captured large temporal scale variation for all communities (r = 0.98 for correlation with year). However, the temporal extent for ciliates was smaller (5 years), and thus, the negative relationship with time may be similar to the pattern showed in the last 5 years of sampling for other communities (e.g. phytoplankton in the isolated lake r = − 0.43; P = 0.03).

The pRDAs for phytoplankton and zooplankton with a temporal window similar to the ciliates (n = 10) showed that in most cases, the residuals increased and the null effect of biotic factors on communities was maintained (Fig. S6). Phytoplankton (composition and turnover) went from being controlled by the environment and time to be controlled exclusively by the environment. For zooplankton, the null effect of the environment was maintained and, in some cases, temporal factors lost the effect on the community structure (composition at isolated lake and turnover at connected one).

Discussion

In this study, our goal was to evaluate the variation in the diversity of plankton communities at multiple temporal scales in two shallow lakes of a subtropical floodplain. Diversity changes at the samples (α), among samples (β) and in a collection of samples (γ) are linked (Chase et al. 2011; Magurran et al. 2018). By partitioning diversity across multiple temporal scales, we revealed that changes in species composition through time (β) have the higher contributions to γ diversity. Then, we found that β diversity (either as variation in composition or dissimilarity in composition) for the planktonic communities showed differences in the contribution of environmental and stochastic factors.

On the other hand, we found that planktonic communities displayed a concordant pattern of variation in community composition across lakes. However, the concordance among plankton communities was weak and the factors driving the community assemblage varied among them. This result is in line with previous studies and shows that different aquatic communities perceive the habitat in different ways (Heino 2010; De Bie et al. 2012; Padial et al. 2014).

The importance of temporal scales to plankton communities

The result of the partitioning of gamma diversity showed that the observed values of α diversity were significantly lower than the null expectation (for instance, in average only 11% ± 5% of species were present within samples). Thus, these communities tended to show high variability in species composition among samples. Moreover, our results suggest, as discussed below, that ecological processes influenced the observed non-random pattern of the communities. On the other hand, we expected that a temporal scale encompassing the flood seasonality (β2) would explain most of the variation in the plankton community since the flood pulse is thought to be the main dynamic force regulating the structure of communities of floodplains (Agostinho et al. 2004; Junk et al. 1989). Although β2 showed an important effect on the diversity (over 25–50% of the gamma diversity), the variation between flood seasons was lower than expected from a random allocation of samples among temporal scales. Patterns of richness remained stable over time perhaps because the number of species may not respond to the environment, as richness and abundance in communities can be regulated by low-level compensatory mechanisms (e.g. interactions and changes within functional groups), making the communities more stable over time (De Boeck et al. 2018; Gotelli et al. 2017; Hallett et al. 2014).

The stability of richness over time could also be related to an environmental variability among seasons that is not high enough to promote changes in the species composition. This low environmental variation could be related to the operation of dams upstream that have reduced the frequency and intensity of the flood pulse, and that consequently can affect the variation in species composition between the high- and low-water seasons (Agostinho et al. 2007; Agostinho et al. 2004; Bozelli et al. 2015). This decrease in the temporal variation of environmental conditions may hinder the establishment of a greater number of species (Chesson and Warner 1981; Stein et al. 2014), reducing the dissimilarity in species composition. In addition, dams negatively affect the plankton richness in downstream areas by preventing the arrival of new species from upstream. Because dams regulate the volume of water flowing downstream, they also reduce connectivity among habitats, which hinders the exchange of plankton inocula (Abrahams 2008; Agostinho et al. 2009).

Seasonal variation associated with the flood pulse is important for plankton diversity (Simões et al. 2013). In fact, our results showed that the composition variation related to the flood pulse scale was important for gamma diversity of plankton communities. However, we found that processes acting on a broader temporal scale could be more important for maintaining diversity in floodplains. The variation in composition between normal periods and periods influenced by ENSO (β3) was higher than expected by chance (more than 40% of gamma diversity) for the plankton communities in these floodplain lakes. Previous studies have demonstrated the importance of environmental variability associated with El Niño and La Niña in maintaining plankton diversity (Bovo-Scomparin and Train 2008; Simões et al. 2013; Solari et al. 2014), but our goal in this study was to simultaneously evaluate the relative effect of ENSO and the flood pulse on plankton communities. Considering that the intensity of the flood pulse has decreased due to dam construction (Agostinho et al. 2005), the effect of ENSO events may be highly important for plankton diversity in the Upper Paraná River floodplain, leading to high environmental variability through time (Solari et al. 2014; They et al. 2015), which contributes to shifts in species richness and composition (Stein et al. 2014, Fig. S5).

We are aware that the underestimated richness values could affect the influence of alpha (underestimating) and beta diversity (overestimating) on gamma diversity (Crist and Veech 2006; Beck et al. 2012). In this study, the species accumulation curves for the three communities did not reach an asymptote, indicating that we did not sample the total of species. In fact, the observed number of species varied between 60 and 76% of the expected values (see supplementary material). Although this problem can be related to methodological artifact that underestimate species richness (e.g. limitation in individual counting for phytoplankton, Rodríguez-Ramos et al. 2014), the difficulty in reaching saturated accumulation curve could be common for hyperdiverse communities in studies that consider broad scales (Gering et al. 2003). In our case, reaching a saturated accumulation curve could be difficult, as we had a sampling design extensive in time with only two sites in a highly heterogeneous system such as a floodplain. In this sense, the regional pool could be an important effect on the local diversity, as plankton organisms are high dispersers and a higher number of species is expected to reach the lakes over time. Indeed, we showed that in many cases the composition and turnover of plankton communities were mainly influenced by the temporal factors suggesting a dispersal effect on community structure (see below). Although temporal extensive sampling designs could produce unsaturated accumulation curves, studies at broad temporal scales can provide a better assessment of the heterogeneity of species diversity (Gering et al. 2003). Hence, the results of gamma partitioning analyses must be interpreted with care (Beck et al. 2012). However, as observed patterns and null models are based on the same data set, the undersampling does not affect the assessment of alpha and beta components based on null models (Crist and Veech 2006; Beck et al. 2012).

Factors driving community composition over time

While no explanatory factors had a significant effect on ciliates, both environmental and temporal factors explained the variation in composition and temporal turnover of phytoplankton and zooplankton. Even after decreasing the length of the time series (fewer observations) for phytoplankton and zooplankton, the explanatory factors had a higher effect on them than on ciliates. In addition, even though different taxa responded in a similar way to environmental variability, the concordance was stronger for within taxa across lakes. Those results are in line with previous studies that show that different aquatic communities perceive the habitat in different ways (Heino 2010; De Bie et al. 2012; Padial et al. 2014) due to differences in traits (e.g. morphological, functional and ethological) and life history. For instance, differences in body size influence the response of plankton communities to the environment, with the smaller organisms (ciliates and phytoplankton) being more susceptible to environmental control than larger ones (zooplankton), as processes such as dispersal are more limiting for larger organisms (De Bie et al. 2012; Tonkin et al. 2018). In the case of ciliates, our results contrast with studies showing that environmental factors are important drivers of ciliate communities (Gimmler et al. 2016; Küppers and Claps 2012; Segovia et al. 2017). In this sense, by controlling the effect of the time on the environment influence, we decreased the probability of considering spurious relationships between community and environmental variation as significant.

The low explanatory power of the environment in our study can be related to multiple aspects. Differences in growth rates and a delayed response to environmental changes (Su et al. 2015; Hastings et al. 2018) among communities can influence the relationship between community variation and environmental factors. In addition, the predictability of phytoplankton and ciliates communities is higher at finer temporal scales (hours to months) and decreases at larger temporal scales (from months to years) (Dolan 2005; Xu et al. 2015; Thomas et al. 2018). Furthermore, the relationship among patterns of composition and environment tend to be stronger when high environmental variability is included; that is, the explanatory power of temporal or spatial factors decreases when more habitats or high environmental variability occur in the system (Heino et al. 2015a; Leibold and Chase 2018a).

In studies testing both the effects of environmental and temporal factors, it is difficult to interpret the ecological meaning of temporal variation in composition. Legendre and Gauthier (2014) pointed out that neutral processes, such as random colonization and ecological drift (local extinction due to demographic stochasticity), could explain the community variation that is not related to environmental factors but related to a pure temporal fraction. In our partitioning analysis, the pure effect of environmental factors could be related to niche processes, while the pure effect of temporal matrices could be related to neutral processes (Leibold and Chase 2018b). Neither processes alone influence communities, but an interaction between processes resulting from differences in colonization histories and niche-based processes can be important in highly productive systems such as tropical and subtropical ecosystems (Steiner 2014; Purves and Pacala 2005). In fact, as mentioned above, we showed that both environmental and stochastic processes (time, residual variation) simultaneously influenced the community variation. In most cases, however, the stochastic processes showed the highest explanation of community variation. In this sense, there was a variation in composition and turnover explained by flood seasons and ENSO and their interaction with environmental variables. Thus, these events might influence floodplain environments through precipitation and water discharge regimes, which can influence the arrival of species from the regional species pool (Evans et al. 2017; Lansac-Tôha et al. 2016; Medley and Havel 2007; Simões et al. 2013).

The pure temporal component explained the highest proportion of the variation of zooplankton, suggesting that stochastic processes are driving their composition variation and turnover. The AEMs included in our analyses were associated with the variation occurring at shorter (months) and longer timescales (years). The significance of AEMs with low eigenvalues (e.g., AEM 34) represent a higher similarity among near (in time) samples, while AEMs with high eigenvalues (e.g., AEM 1) are related to higher dissimilarity among more distant samples (Heino et al. 2015b). This type of relationship is common in metacommunity studies that use AEMs to model the spatial relationships of communities (samples) (see Bortolini et al. 2017), relating the high dissimilarity between distant samples to dispersal limitation, and high similarity between neighboring communities to high dispersal rates. In our case, we did not study spatial but temporal variation and suggest that the selected AEMs could represent two ways in which changes in species composition occurred over time. On one hand, the AEMs related to broader temporal scales represented high dissimilarity among communities distant in time, because temporal extinctions (without replacement by migration from the regional pool) prevent some species from “reaching” optimal conditions for their establishment (Dornelas 2010; Lansac-Tôha et al. 2016). On the other hand, for finer temporal scales, AEMs represented high similarity among near communities (in time) because of the delay in the response of communities to environmental variation or due to a constant and high species dispersal to the lake.

Thus, a higher species turnover through time with a low explanation of environmental factors may be considered a result of dispersal limitation and ecological drift (Leibold and Chase 2018b) or an interaction between dispersal and environmental factors (Steiner 2014). In either case, the variation explained by AEMs could represent the effect of past conditions (Andersson et al. 2014) on the structure of microbial communities. It is likely that AEM significance, however, could also be associated with temporally structured environmental factors that we did not measure (Legendre and Gauthier 2014) and to multiple processes, that, although deterministic, might appear as stochastic over space and time (e.g. differences in the time of arrival, frequency-dependent processes, differences in dispersal rates) (Dornelas 2010; Hastings et al. 2018; Leibold and Chase 2018b).

We also noted a high residual variation for all communities. It has been suggested that the apparently random assemblage of the plankton communities is related to the high level of noise in the data (Attayde and Bozelli 1998) and to the lack of important explanatory factors included in the analysis (Beisner et al. 2006; Hessen et al. 2006), such as dispersal rates. Although we did not evaluate the dispersal effect, we are aware that the communities we studied are not closed systems and that dispersal could influence their structure (Evans et al. 2017; Heino et al. 2015b; Lansac-Tôha et al. 2016). In this case, the effect of dispersal could be related to the variation that the environmental and temporal factors evaluated did not explain. Hence, the joint investigation of temporal and spatial patterns of diversity is needed so we can have a deeper insight into the effects of stochastic and deterministic processes (Leibold and Chase 2018b).

Conclusions

We showed that the variation in composition at several temporal scales helped to maintain the diversity of plankton communities in shallow lakes of a tropical floodplain. The environmental variability related to ENSO events was reflected in the high temporal variation in species richness and composition, suggesting that the El Niño-La Niña cycle is highly important in maintaining plankton diversity. Thus, our findings highlight the importance of continuous monitoring in order to understand the processes underlying diversity variation at multiple temporal scales. Moreover, analyses of long time series are necessary to understand the effects of climate change on diversity. Besides, our results showed that plankton communities had similar temporal patterns of variation while the processes driving the species assemblages varied among them. While niche-associated (environmental) and stochastic (temporal) processes drove phytoplankton structure, mainly neutral processes were important for zooplankton, and neither for ciliates. Finally, regardless of the processes underlying the temporal predictors (neutral or niche processes, which we did not identify), our results showed that processes acting at different time scales influenced the variation of the plankton communities.