Introduction

Natural floodplains in river basins are among the most biologically diverse and productive ecosystems on the planet (Ward et al. 1999; Tockner and Stanford 2002). The natural river flow regime is considered the key factor driving ecological functioning and biodiversity patterns in floodplain ecosystems (Junk et al. 1989; Bunn and Arthington 2002). Flood pulses actively shape the floodplain surface, creating a spatio-temporal heterogeneous environment supporting a great variety of habitats (Ward et al. 1999; Tockner et al. 2010; Davidson et al. 2012).

Like other freshwater wetlands, river floodplains are among the most threatened ecosystems and over the past 30 years species diversity has declined faster here than in terrestrial or marine ecosystems (Loh and Wackernagel 2004; MEA 2005; Revenga et al. 2005). Alteration of the flow regime is generally considered one of the most serious and continuing human threats to the ecological intactness of these ecosystems (Poff et al. 1997; Postel and Richter 2003; Poff et al. 2007).

Everywhere, rivers have been regulated for various purposes, including public water supply, irrigation, navigation, flood mitigation and electricity generation, thereby contributing considerably to human development and economic prosperity (Dynesius and Nilsson 1994). As a consequence, over 60 % of global river systems have been affected by altered stream flows (Revenga et al. 2000), and this figure is projected to increase (World Wildlife Fund 2004). Damming is one of the most common types of water flow management, and has profound effects on river hydrology (Middleton 2002), affecting both the total discharge and the variability (i.e. frequency, duration, timing, magnitude and rate of change) of flow (Poff et al. 1997).

Several studies have described the potential consequences of river regulation for riverine ecology in a ‘qualitative’ manner (e.g. Kingsford 2000; Nilsson and Berggren 2000; Bunn and Arthington 2002; Poff et al. 2009; Tockner et al. 2010). Attempts to quantify relationships between river regulation and biodiversity have, however, been so far incomplete or inconclusive (Poff and Zimmerman 2010). Therefore it is difficult to incorporate impacts of river regulation on biodiversity in strategic environmental assessments, which frustrates sound decision making (Arthington et al. 2006; Jackson et al. 2001; Stone 2008).

At a global level, human impacts on river systems are qualitatively assessed, using expert models or direct correlations between species distribution data and abiotic data (Sala et al. 2000; Xenopoulos et al. 2005). Quantitative assessments using empirically based relationships between flow alteration and ecological impact and state of the art models are necessary for more reliable impact assessments to support policy makers at both a regional and global level. Global models of environmental change and biodiversity, such as IMAGE (MNP 2006) and GLOBIO (Alkemade et al. 2009, 2011; Netherlands Environmental Assessment Agency 2010), are important tools for exploring policy options and scenarios for biodiversity conservation (e.g. within the framework of the Convention on Biological Diversity or the Ramsar Convention on Wetlands). The GLOBIO model comprises a module for terrestrial (Alkemade et al. 2009), and (recently) for inland freshwater ecosystems, referred to as GLOBIO-aquatic (Alkemade et al. 2011; Janse et al. unpublished manuscript). Considering the relatively high biodiversity value and associated conservation importance of floodplains, the effects of alteration of their natural flow regime need to be implemented in GLOBIO-aquatic.

The GLOBIO model is based on empirically based cause–effect relationships between environmental drivers and biodiversity impacts, formulated through meta-analyses of published data (Alkemade et al. 2009). The impact on biodiversity is expressed in terms of the mean species abundance (MSA) of original species, relative to their abundance in an undisturbed control situation (Alkemade et al. 2009; Secretariat of the Convention on Biological Diversity 2006), which is a measure of intactness of biodiversity similar to the Biodiversity Integrity Index (Majer and Beeston 1996) and the Biodiversity Intactness Index (Scholes and Biggs 2005).

This study contributes to the GLOBIO model by exploring the consequences of river regulation on the biodiversity intactness of flood-dependent wetlands. To uncover and quantify any general patterns at the regional and global level, we systematically searched the scientific literature for existing datasets relating community composition to hydrological change. We calculated the MSA to quantify effect size. As the MSA is not frequently used in meta-analyses and does not completely cover the complex biodiversity concept, we compared the MSA to a second indicator: the relative change of species richness (Faith et al. 2008). Meta-analyses were used to combine effect sizes across all studies and test the overall significance of flow regime alteration. Linear mixed effect models were used to explore potential effect modifiers.

Methods

Systematic search and data extraction

In February 2011, relevant published peer-reviewed articles on the impact of hydrological alterations on the biota in floodplain wetlands were searched using the electronic databases ISI Web of Knowledge and Scopus (Table 1). The setup of the systematic search is based on the guidelines provided by the Collaboration for Environmental Evidence ( 2013).

Table 1 Components that constitute the aim of the system search
Full size table

Studying the relationship between hydrology and ecology can typically be approached from either a hydrological or an ecological perspective. We used concepts from both disciplines for the search terms, and compiled search strings for extracting relevant articles from the databases as effectively as possible (Table 2). No grey literature was used in this systematic review. Possible bias resulting from this exclusion was verified.

Table 2 Search terms used to find relevant literature
Full size table

Titles and abstracts of all returned hits were checked and judged against the aim of the review. From the papers that were not discarded we selected those that met the following criteria:

  • Relevant subject: Community composition in any floodplain wetland. Studies were included irrespective of habitat or spatial scale, except for experimental studies with a focus on micro scale processes.

  • Types of intervention: Human induced alteration of the natural flow regime, described in quantitative units.

  • Types of comparator: Undisturbed reference conditions in space or time (Before-After or Control-Impact studies).

  • Types of outcome: A quantitative comparison in terms of community composition (species richness, abundance or density).

The search query resulted in 686 unique articles, of which 28 fulfilled the selection criteria and contained usable data. 20 papers reported on community composition in terms of species abundance and 19 papers reported on local species richness (11 studies reported on both). All 28 articles reported on at least some differences in species composition between regulated (control) and unregulated (impact) sites. In several cases the differences were reported to be non-significant and were attributed to natural variation between sites. However, most studies showed differences that could be interpreted as effects of hydrological alterations, although not all changes had similar direction. The majority of the papers presenting data on species richness reported a reduction in species richness (16 out of 19). Studies on species abundance revealed a more complex picture and reported on shifts (positive and negative) rather than general decreases in the number of individuals. Table 3 shows which taxonomic groups were represented by the 28 papers, along with the types of hydrological regulation and the (terrestrial) biomes indigenous to where the studies took place (See Online Resource 1 for a short description of each study).

Table 3 Descriptive representation of the selected articles resulting from the search query
Full size table

The available data on species richness and abundance, along with sample variance (if available) was extracted from the selected publications. Java Plot Digitizer 2.5.1 software was used to extract data from figures if values were not presented in tables. Abundance may have been recorded as density (number of individuals) or relative cover, depending on the taxonomic group studied. We preferred averaged data over cumulative numbers and, if presented, used transformed or rarified data to correct for differences in sampling size. To assess the hydrological alteration, we included the type and degree of hydrological alteration in the database by means of a short qualitative description and extraction of quantitative data of any hydrological variable provided. Data from disparate locations or different degrees of hydrological alteration presented in one article were treated as different datasets and considered independent. This resulted in 29 datasets for which we could calculate the MSA, and 32 datasets for which we could calculate the relative species richness. Additionally, we recorded system characteristics that could be regarded as cofactors, such as location (coordinates) and taxonomic details of the species studied.

Data synthesis and presentation

Ecological change

We used the MSA and the relative species richness as metrics of effect size for a quantitative comparison of datasets. To estimate the MSA of a dataset, we calculated the relative abundance of each species represented in the dataset (Eq. 1a). Effect sizes of all species were then averaged to calculate the Mean Species Abundance (Eq. 1b):

$$R_{is} = \frac{{A_{isd} }}{{A_{isc} }}$$
(1a)
$$MSA_{s} = \frac{{\mathop \sum \nolimits_{i} R_{is} }}{{N_{s} }}$$
(1b)

where R is is the ratio between the density of species i in the disturbed site d and its abundance in the undisturbed reference situation c in study s, for A isc  >0. MSA s is the mean species abundance estimated in study s. Ns is the number of species in study s. If the abundance of a species in the disturbed site was higher than in the undisturbed floodplain, the ratios were truncated to 1. The variance of the MSA value for each study was estimated by calculating the variance of the external or the internal error (cf. Benítez-López et al. 2010; see Online Resource 2 for details). Taking a conservative approach, the larger of the two variances was used in the meta-analysis (DerSimonian and Laird 1986).

For each dataset on species richness, the relative species richness (LR) was calculated by dividing the number of species Jx found in the disturbed (impacted) sites by the number of species Jo found in undisturbed reference (control) sites. The logarithm of the ratio was used as effect size for its statistical properties (Hedges et al. 1999):

$$LR = log\frac{{J_{x} }}{{J_{o} }}.$$
(2)

The variance of each individual effect size was calculated as:

$$\sigma_{LR}^{2} = \frac{{\sigma_{{J_{x} }}^{2} }}{{\sigma_{x}^{2} }} + \frac{{\sigma_{{J_{o} }}^{2} }}{{\sigma_{o}^{2} }},$$
(3)

where Jx and Jo are the species number and \(\sigma_{{J_{x} }}^{2}\) and \(\sigma_{{J_{o} }}^{2}\) are sample variances of the species number found in impacted sites (x) and undisturbed reference sites (o) respectively. In case Jx = 0 the data point was omitted.

Hydrological change

Ideally, the relationship between flow alteration and ecological variables would be expressed as a simple quantitative ratio (i.e. % ecological change in terms of % flow alteration) (Poff et al. 2009). However, the natural flow regime comprises numerous interacting components that are hypothesized to drive ecological processes. Moreover, measures of flow modification are mostly inconsistently reported in the literature (Olden and Poff 2003; Poff and Zimmerman 2010). Poff et al. (2009) remarked that ecological changes may also be formalised and empirically tested when they are expressed as categorical responses. In that vein, we decided to group hydrological alterations into three subclasses on an ordinal scale (low, medium, high), referred to as disturbance classes, to formalize the degree of alteration. If presented, formal indices of hydrological alteration were used to subdivide the cases over the three categories. Otherwise, the proportional changes of the primary hydrological variables that were presented were used for classification, with reference to the qualitative descriptions given by the authors (see Online Resource 2 for details).

Meta-analysis

For both metrics a random effect meta-analysis was used to derive a weighted mean for all datasets and test the global mean effect of flow regulation on the biodiversity of floodplains. Sampling variances were used to calculate weights and included in the analysis to give studies with relatively good precision more weight (Osenberg et al. 1999). Publication bias was assessed by inspecting funnel plots of asymmetry along with formal regression tests (Egger et al. 1997; Viechtbauer 2010), assuming that studies with large sample sizes would be near the average, and small studies (with more variance) would be spread on either side of the average. Heterogeneity was assessed by inspection of forest-plots and formal tests of heterogeneity Q and I2 (Borenstein et al. 2009). A significant Q (P < 0.05) denotes that the variance among effect sizes is greater than expected from sampling error and indicates that different studies do not share a common effect size. The I2 index quantifies the degree of heterogeneity and is a measure of the signal to noise ratio across the observed effect estimates (Huedo-Medina et al. 2006). To elaborate on the degree of hydrological alteration, separate meta-analyses were performed for each of the disturbance classes containing non-duplicated independent datasets. The meta-analyses were performed with the ‘metafor’ package available in R 2.15.1 software (Viechtbauer 2010).

To assess whether the relationship between hydrological alteration and ecological impact is influenced by other variables, we explored sources of heterogeneity by building linear mixed-effect (LME) models with several potential effect modifiers as fixed effects and the source article as a random effect. As well as the degree of hydrological alteration (disturbance classes), the factors considered were the percentage of non-natural land use in the catchment area, biome-type and taxon. We used the coordinates of each study site to retrieve the percentage of non-natural land use (NNLU) in the catchment area upstream of the study site from the IMAGE database (MNP, 2006). We subdivided the studies over three categories: low (<10 %), medium (10–50 %) and high (>50 %) percentage of NNLU. The IMAGE framework was also used to assign biome types to the studies according to their location (cf. Benítez-López et al. 2010). Furthermore, species identity was used to divide the datasets over different taxonomic groups: mammals, birds, fish, invertebrates and plants, as these groups are expected to present distinct differences in their response to alterations. Models were compared by means of Akaike’s information criterion (AIC) and Akaike weights.

Results

Meta-analysis

Our analysis showed that on average, alteration of the natural flow regime has a negative impact on biodiversity in floodplain wetlands. The meta-analysis of all MSA datasets results in a pooled effect size of 0.482 (Table 4)—a significant divergence from the no effect level (P < 0.0001). Publication bias was low (regression test, P > 0.5), and the funnel plots showed no asymmetry (forest plots and funnel plots are presented in Online Resource 3). The high values for Q and I2 indicated that the datasets did not share a common effect size (P(Q) < 0.0001; Table 4), and this was confirmed by inspection of the funnel plot, which showed large deviations in the effect size of datasets with small variances. The high fail-safe number indicated that a large number of studies reporting neutral effects would be needed to overturn the significance of the pooled effect size, and therefore the pooled effect size can be considered a reliable estimate of the mean effect.

Table 4 Results of the meta-analysis and subgroup analysis for the mean species abundance of original species (MSA)
Full size table

When considering the datasets on species richness, the all-encompassing meta-analysis also showed a significant effect (P < 0.01; Table 5), with a pooled LR of −0.140 (i.e. a 28 % reduction). There was no evidence for publication bias and, in contrast to the MSA, there was no evidence for heterogeneity (P(Q) > 0.1; Table 5); the low values for Q and I2 indicated that the variance in effect size values was not higher than expected from natural variation and the studies might share a common ‘true’ effect size.

Table 5 Results of the meta-analysis and subgroup analysis for the log of the ratio of species richness reported in disturbed and undisturbed (reference) sites (LR)
Full size table

Subdividing datasets over three classes of hydrological alteration did not lead to significant differences between the classes (Welch’s t tests; P > 0.1), and the pooled effect sizes per disturbance class did not all differ significantly from the no-effect level, as was shown by subgroup meta-analyses (Tables 4, 5). However, the disturbance class representing the strongest disturbance did show the largest pooled effect size, while lesser disturbance led to smaller (non-significant) effects, which was consistent for the MSA (Fig. 1) and the LR (Fig. 2). Increase in the intensity of hydrological alteration tends to lead to more drastic deviation from the reference state.

Fig. 1
figure 1

Outcome of the meta-analysis; MSA values for the three disturbance classes ±1 SEM (dist. class ‘low’, n = 4; dist. class ‘medium’, n = 10; dist. class ‘high’, n = 15). MSA = 1 indicates ‘no effect’. Asterisks denote denotes a significant difference (P < 0.001) compared to the no-effect level

Full size image
Fig. 2
figure 2

Outcome of the meta-analysis; average effect size of the log ratio of species richness for the three disturbance classes ±1 SEM (dist. class ‘low’, n = 4; dist. class ‘medium’, n = 10; dist. class ‘high’, n = 18). The single asterisk denotes a significant difference (P < 0.05) compared to the no-effect level

Full size image

Heterogeneity

For both metrics the degree of hydrological alteration (i.e. classes of disturbance) was revealed to be a relevant factor in explaining variance among the effect sizes (Table 6). Examination of the linear mixed-effect (LME) models revealed, however, that the degree of disturbance is not the only relevant factor. In fact, the most comprehensive LME models turned out to be the most parsimonious in explaining heterogeneity i.e. they presented the lowest AIC values (Table 6). For the MSA, the smallest contributive factor is the percentage of NNLU in the catchment area, as deduced from comparison with the second best explanatory model, while for the LR the type of biome is the least additive.

Table 6 Models expressing differences in effect sizes explained by different potential effect modifiers
Full size table

Discussion

The qualitative review of the selected articles that resulted from the literature search had already revealed that 84 % of the articles reported a decrease in species, corroborating previous summaries by Poff and Zimmerman (2010), who also reported 84 % of studies that presented negative ecological responses, and Lloyd et al. (2003), who found that 86 % of the studies presented an negative ecological effect (the latter also included in-channel responses). Although such an overview makes it clear that flow modification does affect biodiversity in floodplain wetlands in most cases, the responses of individual studies are variable to some extent. Hence, such a summary does not permit a quantitative estimate of the average effect and its significance. Here we provide evidence showing that river regulation has a significant overall negative effect on biodiversity intactness in floodplain wetlands, and present a reliable quantitative estimate of the mean response: on average, alteration of the natural flow regime reduces the mean abundance of the original species (MSA) by >50 %, and species richness by >25 %.

It has until now been difficult to detect general, transferable quantitative relationships between individual measures of flow alteration and ecological response (Poff and Zimmerman 2010). This is partly due to the fact that there is a large variety of natural flooding regimes, which is also reflected in our database. Another important factor here is inconsistent reporting on the myriad hydrological variables in the literature, which frustrates possible identification of any relationships. In the search for generalities and greater quantitative understanding at the regional and global level, we circumvented these difficulties by deliberately including different flooding regimes, and categorizing the responses of hydrological variables (both qualitative and quantitative) into three broad categories to formalize the degree of hydrological alteration. Although this method allows for some arbitrariness and the resulting comparisons are only semi-quantitative, we argue that the method does provide some valuable information on the degree and effect of alteration and an opportunity to test it.

Subgroup analysis revealed a pattern indicating an effect of the degree of flow modification for both MSA and LR; the studies that were classified as lightly disturbed presented on average the least ecological response, whereas the studies assigned to the highest disturbance category showed on average the strongest ecological response. The importance of degree of flow modification in explaining heterogeneity was confirmed by the LME models.

If one compares indices of biodiversity impact for the low degree of hydrological alteration, it appears that species richness is more robust than MSA, indicating that the abundance of species is decreasing while species richness is not affected per se, revealing a complementarity between the two indicators. On longer timescales, however, reduced abundances can have serious consequences when only juveniles are impacted (Braatne et al. 2007) and community regeneration is severed. In general, (severe) flow modification showed a net decrease in species richness (including the arrival of new species), implying that the MSA does not provide an exact picture of the average decrease in abundance of any species as it must also be influenced by the disappearance of original species.

The LME models indicated that the relationship between flow modification and biodiversity impact is not determined solely by the degree of alteration to the river; other factors also influence the impact on biodiversity. One factor we considered was the percentage of non-natural land use (NNLU) in the catchment area. NNLU itself may impact the biodiversity of streams (e.g. Weijters et al. 2009) as land use covaries with riparian degradation (Walsh et al. 2007), the inflow of nutrients (Crosbie and Chow-Fraser 1999) and other (toxic) pollutants. Although the studies used in our analyses control for environmental conditions other than the hydrological treatment, the reference conditions do not always reflect pristine reference conditions, and vulnerable species may have already disappeared. Indeed, inspection of our data suggests that areas with a low percentage of NNLU are more sensitive to river regulation than areas with high NNLU. Furthermore, the relationship seems to be determined by the biome in which the wetland is located, and the type of taxa studied. Preliminary analysis of our data indicates that in terms of the MSA, animals tend to have a stronger response to flow modification than plants, while in terms of species richness plants appear more sensitive. Poff and Zimmerman (2010) also mentioned variation in response for different taxonomic groups, showing that fish are especially sensitive to flow alteration. Although animals are more mobile than plants, the latter may show greater adaptive phenotypic plasticity and benefit from the presence of a seed bank (Jansson et al. 2000). Nonetheless, the effect on plant species may simply be delayed if plant communities respond slowly to new environmental conditions and sudden shifts in dominance may appear at a later point in time (Braatne et al. 2007).

These factors (and probably many others) should therefore be taken into account when further investigating and quantifying the relationship between flow modification and biodiversity impact, particularly for the MSA for which heterogeneity was found to be high in the dataset. For species richness, however, there was no evidence for heterogeneity, indicating that the underlying ecological response is more robust and may be easier to predict in a field situation.

Generally, a paucity of data limits a more detailed analysis (Lloyd et al. 2003). There were already only a few data points available to test the effects of a low degree of hydrological alteration, indicating that there is a bias in the scientific literature towards studies reporting on severely disturbed flow regimes (Poff and Zimmerman 2010). Also, there was a geographical bias in our dataset, with most studies conducted in North America, Europe and Australia. Moreover, most studies concentrated on impacts on vegetation, with a disproportionate interest in cottonwood forest ecosystems in the American West. By expanding the search query, checking more electronic databases and including ‘grey’ literature, the number of usable datasets could possibly be increased, allowing greater quantification of the relationship between river regulation and biodiversity impact.

Despite these limitations, our quantitative meta-analysis reveals that flow modification has an overall significant negative effect on biodiversity intactness in floodplain wetlands, and provides an important first indication of the effect size in terms of mean species abundance and species richness. Moreover, our analyses suggest that impact on biodiversity is positively related to the degree of hydrological alteration. These insights may now be incorporated into the IMAGE–GLOBIO framework (Online Resource 4; Janse et al. unpublished manuscript), and used for policy decision making (Poff et al. 2003), so facilitating more effective and wise natural resource management (Stanford et al. 1996).