Responses of fish and invertebrates to floods and droughts in Europe

Floods and droughts, two opposite natural components of streamflow regimes, are known to regulate population size and species diversity. Quantifiable measures of these disturbances and their subsequent ecological responses are needed to synthesize the knowledge on flow–ecosystem relationships. This study for the first time combines the systematic review approach used to collect evidence on the ecological responses to floods and droughts in Europe with the statistical methods used to quantify the extreme events severity. Out of 854 publications identified in literature search, 54 papers were retained after screening and eligibility checks, providing in total 82 case studies with unique extreme event—ecological response associations for which data were extracted. In this way, a database with metadata of case studies that can be explored with respect to various factors was constructed. This study pinpointed the research gaps where little evidence could be synthesized, for example, drought event studies and fish studies. It was demonstrated that in many cases the studied metrics (abundance, density, richness, and diversity) showed statistically significant decreases after or during the event occurrence. The responses in invertebrate density and richness were in general more negative than the corresponding responses in fish. Biota resistance to floods was found to be lower than the resistance to droughts. The severity of extreme events was not found to be an important factor influencing ecological metrics, although this analysis was often hampered by insufficient number of case studies. Conceivably, other factors could mask any existing relationships between disturbance severity and biotic response.


INTRODUCTION
The natural flow of a river varies on a range of time scales, from hours to years and longer (Poff et al., 1997). Flow regimes vary regionally, and their properties are typically controlled by environmental factors such as climate, topography, land cover, soils and geology, and anthropogenic factors such as morphologic alteration, water abstraction, dams, or diversions. Extreme high and low flows are two opposite natural components of flow regimes of rivers worldwide. These excesses and deficits in water movement are often perceived by stream ecologists as disturbances (Lake, 2000) that regulate population size and species diversity across a range of spatial and temporal scales (Lytle & Poff, 2004) and that are "the dominant organizing factor in stream ecology" (Resh et al., 1988). For example, some consequences of developing droughts are (a) reduction and fragmentation of habitat space, (b) breaking longitudinal connectivity, (c) deterioration in water quality, and ultimately (d) loss of biota (Lake, 2000). Sequential drying of different habitats that act as refuges when connectivity is lost triggers a stepped response of the biota (Environment Agency, 2013).  (Lytle & Poff, 2004). The effects of single hydrological extreme events are highly context dependent, ranging from deleterious to beneficial, and reliant upon event magnitude, extent, and timing relative to life cycles of constituent species (Ledger & Milner, 2015). Much insight into the nature of extreme flow-biota relationships is offered by long-term hydroecological datasets comprising community metrics and streamflow time series, such as the one available for the Little Stour River in the UK (Wood & Petts, 1999;Wood & Armitage, 2004;Stubbington, Wood, & Boulton, 2009a;Stubbington, Boulton, Little, & Wood, 2015).
Hydrologists have developed a wide range of indices that quantify the severity of hydrological extreme events. These include, for example, flow duration curves, low-flow frequency curves, continuous low-flow events analyses, baseflow separation techniques and recession analysis for droughts, (Smakhtin, 2001;Keyantash & Dracup, 2002;Lake, 2011); and flood frequency or flood peak magnitude, duration above a threshold (high-flow pulses) for floods. Unfortunately, these indices are rarely used in ecological studies to characterize hydrological extremes under investigation, which hampers any comparisons between events across different studies (e.g., (Lake, 2011) for droughts). While there have been studies relating ecological scores to hydrological metrics, they are rarely targeted to extreme events. For example, Monk, Wood, Hannah, and Wilson (2008) used the Lotic Invertebrate Index for Flow Evaluation (LIFE) scores to study the interannual dynamics in instream macroinvertebrate community response in 83 sites across England and Wales. The results allowed to distinguish the responses between dry (1990)(1991)(1992) and wet (1996)(1997) years, but not between individual events. Quantifiable measures of the disturbances, of their effects on abiotic and biotic components, and of the subsequent responses by the biota would help to progress and usefully compare ecological studies in a systematic manner (Lake, 2000). There are flow thresholds where invertebrates and fish show a behavioral response to drought conditions (Environment Agency, 2013). Among 20 research priorities aimed at addressing knowledge gaps in the context of geomorphological and ecological role of floods, Death, Fuller, and Macklin (2015) specified a few directly related to the largely unknown role of extreme events severity, in particular, hydrological indices thresholds. In our view, lack of reported extreme event indicators in ecological studies can only be overcome by completing the hydrological analysis associated with the published material.
Against this background, the objective of this study is to identify evidence in quantitative response of freshwater biota to hydrological extremes in Europe. More specifically, three research questions were formulated: (1) Are freshwater biota significantly impacted by extreme hydrological events? (2) Do ecological responses to extreme events differ between different groups, such as fish and invertebrates or between flood and drought events? (3) Are ecological responses influenced by the severity of flood or drought events? In order to answer these questions, we gathered published evidence through a systematic review, enhanced by a consistent quantification of hydrological extreme events, and employed a robust statistical framework to quantify hydrological extremes-biota relationship in Europe.
We investigated the responses of fish and invertebrates only, as the published evidence is largest within these species (e.g. Garcia De Jalón et al., 2014;Edwards, Baker, Dunbar, & Laizé, 2012;Lake, 2011), selecting studies in Europe that reported biological sampling results (pre-and during-or post-event values) for at least one of the ecological metrics: abundance, density, taxon richness, or diversity (sensu Shannon diversity index or similar indices). Because we were seeking relationships between hydrological events and subsequent ecological responses, we excluded studies for which establishing such connections was impossible.
Even though there have been some previous explorative studies to develop flow-ecosystem relationships, their primary focus was either on the effects of flow alterations (Lloyd et al., 2003;Poff & Zimmerman, 2010;Webb et al. 2013) or of a whole array of natural and anthropogenic changes in different flow regime components, notably including droughts, floods, and high flows (McManamay, Orth, Kauffman, & Davis, 2013). Only Jones and Petreman (2013) focused clearly on the effects of extreme flows on fish populations, but in contrast to other studies (including ours), their methodology did not contain systematic evidence collection and had much more narrow geographical scope (Lake Ontario region). McManamay et al. (2013) reported predominantly negative responses of fish and invertebrates due to droughts and more variable, although predominantly positive, responses due to floods in the South Atlantic Region of the United States. Because their study was aimed to help local managers in developing environmental flow standards in the South Atlantic Region, it focused to a large extent on region-specific anthropogenic flow alterations, which makes a clear difference from our study aiming to better understand the ecological responses to floods and droughts in Europe.
In this paper, we follow the terminology introduced by Lake (2011), that is, whenever we refer to "disturbances," "responses," and "perturbations," we mean, respectively, the following: (a) "disturbances" -hydrological extreme events, that is, either floods or droughts, understood here as (natural) events, having a particular, defined time of occurrence; (b) "responses" (to the disturbance)-impacts of a certain event on biotic components of the ecosystem, here measured by the change in aforementioned ecological metrics; (c) "perturbations"-disturbances and responses considered together. In order to clearly distinguish between biota resistance (capacity of the biota to withstand the stresses of a disturbance) and resilience (capacity to recover from the disturbance; Lake, 2011), in this study, we focus only on the first property, trying to capture evidence of the direct, usually immediate and maximum response in selected metrics.

Evidence collection
We used systematic review methods in order to collect evidence required to address the aforementioned research questions, as they provide a methodological framework to reduce bias present in narra- tification, Screening, and Eligibility, as summarized in Figure 1 and described below.

Identification
Literature search of scientific peer-reviewed studies (journal and conference proceedings articles) was performed using the Thomson Reuters Web of Science Core Collection in June 2014, assumed the main environmental publication electronic database source (see, e.g., Newman et al., 2015). The search terms design was focused on retrieving publications addressing the research questions (

Screening and eligibility
Study inclusion (or exclusion) criteria were applied to consecutively narrow search results and derive only relevant articles ( for full text filtering. The two most frequent exclusion reasons at this stage were (a) lack of hydrological extreme events (b) studies outside Europe. Full text retrieval was successful in 154 cases. All these cases underwent eligibility checks (cf. Figure 1). The full texts screening resulted in a further 74 publications to be excluded. Papers were then analyzed for their quantitative data on ecological responses in terms of abundance, density, richness, and diversity (cf. last two rows of

Data extraction
The full text of each of the retained articles was read for case study (i.e., perturbation) identification and data extraction. If a paper contained relevant data on two or more perturbations, each of them was treated as a separate case study; if it contained data on responses of different biota to the same event, each of them was treated as a separate case study. However, one case study could provide ecological response data for up to four analyzed ecological metrics. In total, 54 papers provided data on 82 case studies. Description of all fields included in the data extraction database can be found in Table S2. They can be grouped into three categories: (a) hydrological event, (b) ecological response, (c) geographical location (saved also in Geographic Information System software). For each location, an approximate upstream catchment area was calculated and classified according to the order of magnitude (e.g., 10-100 km 2 ). Several important assumptions had to be made upon data extraction in order to ensure the coherence of the undertaken approach: 1. If for a given case study sampling results were provided for different habitats or different sites on the same river or on nearby rivers, all results were averaged.
2. If a given case study contained data for different taxons, the results were averaged across taxons (except when the authors clearly  Table S1 for the full search expression used)  Lack of values for specified metrics Sampling design Including at least two samplings, one before and one after (or during) an even Only one sampling or many sampling but without a separation by the event (e.g., only post-event values) Note: a The following metrics were initially considered: abundance, density, richness, diversity (e.g., Shannon index and similar indices), biomass, mortality, reproduction/recruitment, and growth. GIS, Geographic Information System.
distinguished results between analyzed taxons), to achieve consistency with studies with averaged results.
3. If for a given case study sampling results were provided for multiple dates preceding the event, the last one was attributed as a reference sampling date for this event, except when there were premises in the paper to select another date (e.g., in the same season or month the preceding year).
4. If for a given case study sampling results were provided for multiple dates following the event, the date that produced the largest relative change with respect to the reference value was attributed to the event except when the authors specified a date.
As most investigated papers did not report sufficient quantitative information on the severity of the hydrological extreme events, additional analysis was conducted to quantify consistently the severity of the disturbances to answer the research questions.

Floods and droughts severity metrics
Drought and flood episodes have different generation processes, spatial and temporal scales, with floods persisting over days to months and across local (0.5 km 2 ) to regional (10,000 km 2 ) scales while droughts last for months to decades over areas of 50-1.5M km 2 (Garner, Van Loon, Prudhomme, & Hannah, 2015). As a result, methods for characterization and quantification of floods and droughts are also different. In particular, flood events are usually quantified at their peak and frequency of nonexceedance calculated using the extreme value theory (Madsen, Rasmussen, & Rosbjerg, 1997). In contrast, due to their slow onset, droughts are generally defined as periods when flow is lower than a threshold considered as representative of low-flow conditions, and duration and deficit volume are common metrics to quantify drought (Van Loon, 2015). In addition, because the cumulative impact of droughts on the terrestrial ecosystem increases with affected area, drought spatial extent has also been used as a measure of severity.

Flood indices
The flood index metric applied here is the nonexceedance probability of the maximum daily streamflow recorded for each event of interest, expressed as return period T or average number of years between two events of the same magnitude or larger. The Peak-Over-Threshold (or partial-duration-series) method was selected because it selects all independent extreme flood events independently of their periodicity (Madsen et al., 1997). Following Bayliss and Jones (1993) a total of 3·N (N = number of complete years of record) independent flood peaks were sampled from daily mean river flow, with a 7-day minimum duration between two selected peaks. A Generalized Pareto Distribution was fitted on each Peak-Over-Threshold sample based on the probability weighted moments technique (Madsen et al., 1997), giving a uniform relationship between a flood peak magnitude and its return period T. The six case studies for which we were not able to identify a representative gauging station were located in Spain. For them we extracted precipitation time series from the high-resolution gridded precipitation dataset Spain02 (Herrera, Fernández, & Gutiérrez, 2016) and associated precipitation events severity with case study flood events.

Drought indices
Following Parry, Hannaford, Lloyd-Hughes, and Prudhomme (2012); Stahl and Demuth (1999), the drought metric used is the maximum Regional Deficit Index (RDI), which gives the maximum proportion FIGURE 2 Location of flood (a) and drought (b) case studies, representative flow gauges used for flood frequency analysis, Regional Deficit Index (RDI) regions, and European Environment Agency (EEA) biogeographical regions. A blue circle around some of the flood case studies denotes the fact that no gauging station was assigned, either because the return periods were extracted from the paper, or because no suitable gauge was identified. Numbers next to symbols denote the number of case studies associated with a given symbol on a map of a region under low flow conditions during a drought event. The higher the index, the more generalized flow deficits in rivers across the region, and the more extensive and severe the drought. The concept follows the well-established "threshold-level" concept (Zelenhasic & Salvai, 1987)  gauges. Additional regions were created to cover ecological sites outside the original drought catalogue using data from relevant measuring authorities across Europe. Each ecological site identified from the systematic review was assigned to a region either containing the site or whose boundary was closest to it (three cases).

Flood and drought metrics categorization
To enable a rigorous comparative assessment of all levels of analysis to answer the research question, flood and drought indices were summarized in three classes of severity each: low, medium, and high. For flood events, categories were assigned based on the value of return period T with threshold values of 2 and 20 years (  Parry et al. (2012). In order to distinguish between single season, multi-season and multi-annual droughts, thresholds for drought duration were set at 3 and 12 months. The full classification scheme is included in Table 3.

Response ratios for ecological metrics
For each case study (perturbation), the values of at least one of four ecological metrics (Abundance, Density, Richness, and Diversity) were retrieved for two dates t 0 and t 1 (before and after/during an event). For each metric, log response ratios were calculated as where RR EM is a response ratio for ecological metric EM (where EM can be Abundance Ab, Density De, Richness Ri, and Diversity Di), and x 0 , x 1 are the ecological metric values before and during or after an event, respectively. Positive values of RR indicate that values of a given metric increased between t 0 and t 1 , and negative values indicate that they decreased. The use of response ratios facilitates comparisons and output data presentation and has been also used in other systematic reviews in ecology (Newman et al., 2015;McManamay et al., 2013).

Statistical tests
Three types of statistical tests were distinguished to address three specified research questions.
To test whether biota are significantly impacted by extreme hydrological events, we applied the one-sample t-test. This test is applied for (1) To test whether ecological responses to extreme events are influenced by their severity, we applied the one-way analysis of variance (one-way ANOVA), which is a generalization of the two-sample t test for more than two samples. The null hypothesis states that samples in specified groups are drawn from populations with the same mean values. Here, one-way ANOVA is applied for comparing response ratios between three classes of flood/drought severity metrics: low, medium, and high.
All statistical analyses were performed only if subgroup counts were higher or equal than three, following the recommendation from the systematic review of Newman et al. (2015).

Synthesis of case studies
The systematic review and hydrological analyses resulted in a database of hydrological extreme event-ecological response associations in Europe of 82 case studies (CS) originating from 54 papers satisfying the systematic review criteria (cf . Table 4 for list of all selected CS with their attributes and Table S3 for the whole database). Figure 2 shows all flood    Kaendler and Seidler (   Note: a Code: AuthorNamePublicationYear_EventType_BiotaTypeNumber. EventType can be "D" for drought or "F" for flood. BiotaType can be "F" for fish" or "I" for invertebrates. Number is added only if it is required, that is, there is more than one case studies related to one paper.

Statistical analyses
In response to the first research question, Figure 4 shows the results of one-sample t tests verifying whether response ratios related to flood and drought events for the whole population and different subgroups (biota types, event severity classes, EEA biogeographical regions, and catchment sizes) and different ecological metrics are statistically dif-ferent than 0. Note that all statistically significant tests show a decrease in given ecological metrics (i.e., negative response to hydrological extremes size. When analyzing biota types, the most robust findings (Figure 4) showed a decrease of invertebrate density and richness following flood events (p < 0.01 and N > 20) and a decrease of invertebrate abundance (richness) following flood (drought) events (p < 0.05 and N > 10).
Lower significance results for much smaller samples were associated with decrease in invertebrate abundance (five CS and p < 0.01) and fish density (six CS and p < 0.05) following drought events. Not enough CS between drought and fish biota metrics were available, but fish response to flood events (quantified from 6 to 8 CS) was found to be not significant, with examples of increase and decrease.
To address the second question, two types of analyses were made. The results of one sample t tests verifying whether response ratios of ecological metrics for different subgroups (biota type, event magnitude, European Environment Agency (EEA) region, and catchment size) are statistically different than 0 FIGURE 5 Mean plots of response ratios for selected ecological metrics and sub-groups. All between-group differences are statistically significant at the 0.05 level. T-test statistics values are given in the sub-title ical responses to one type of event, flood or drought, between fish and invertebrates was possible for all metrics in response to flood events and only for density in response to drought events. In this case, two out of five tests generated statistically significant results: abundance and richness differed between fish and invertebrates in response to flood events (Figure 5b,c). The results show that not only the magnitude but also the direction of response differed between biota types. While the invertebrate metrics were decreasing by 70% and 32% (mean values) for abundance and richness, respectively, the corresponding fish metrics were increasing by 31% and 11%, respectively.
In response to the third question, the one-way ANOVA was performed to test the effect of event severity class on ecological metrics for samples of at least three CS, but none gave statistical significance at 0.05 level. For illustrative purposes, Figure S1 in the Supporting Information shows the effect plot for RR De of invertebrates to floods, indicating that the difference between classes is not significant.
To further investigate the effect of event severity on ecological responses (as our statistical analysis was limited due to too small sample size), we also looked at selected individual papers that included at least two case studies (i.e., either two different floods or two different droughts) with different severity classes. The rationale was that comparing different case studies (extreme events) within one paper is potentially more homogenous and does not introduce so much noise as in the case of comparing studies from different publications. There were, in total, eight publications meeting this criterion, of which none included three different severity classes (low, medium, and high). There was no clear pattern in these eight publications concerning the responses to events of different severity. For example, in Argerich (2004);Řezničková, Pařil, and Zahrádková (2007); Gaudes, Artigas, and Muñoz (2010), the changes in ecological variables were positively correlated with severity classes (the more severe class the more negative response). On the other hand, in Majdi et al. (2012); Zorn, Van Gestel, and Eijsackers (2005); Wright, Clarke, Gunn, Kneebone, and Davy-Bowker (2004) the relationship was the opposite. In Effenberger, Sailer, Townsend, and Matthaei (2006) and Lobon-Cervia (2009), there was no correlation.

DISCUSSION
The objective of this paper was to identify evidence in quantitative response of fish and invertebrates to floods and droughts as a first step toward a better understanding of the nature of flow-ecosystem relationships in Europe. Specified research questions dealt with significant changes in ecological metrics following floods/droughts events, significant differences in ecological response between different groups, and the role of extreme event severity. To address these questions and facilitate exploratory data analyses, a database of extreme events and corresponding ecological responses was developed using a systematic review framework. The majority of assembled responses were direct, immediate responses that quantitatively characterized biota resistance to extreme events.
The synthesis of European case studies suggests that there exists an evidence gap related to studying the responses of biota to drought events and to the fish responses to extreme events. Notably, the effect of droughts on fish richness, diversity, and abundance remains the biggest gap, with only one case study identified (for abundance) and no case studies for richness and diversity. A relatively low number of perturbations (maximum eight) were identified also for the effects of floods on fish. Metadata analysis as undertaken here is, however, dependent on hypothesis, search terms, and inclusion criteria (cf. 2. Some studies reporting mortalities of invertebrates (Sousa et al., 2012) or fish (Brooker & Morris, 1977) following very extreme events had to be excluded as mortality rates could not be translated into changes in population sizes and in consequence, the response ratios could not be calculated.
3. Several studies were excluded because of reporting values of ecological metrics that underwent a specific standardization and were thus not comparable to all other studies reporting the same metrics without standardization (Wood & Armitage, 2004;Stubbington et al., 2009a).
4. The most frequent exclusion reason for highly relevant studies was, however, the lack of pre-event sampling data that could be used for calculating the response ratios of ecological metrics. Examples include: studies on invertebrate responses following some major droughts in the UK: the 1989-1992 (Wood & Petts, 1999(Wright et al., 2002 drought; a study on invertebrate responses to a severe flood that occurred in a karst river in the UK in 2007 (Stubbington et al., 2009b). Some studies in the Mediterranean regions (Bravo, Soriguer, Hernando, 2001;Langton & Casas, 1998) tended to compare sampling results between drought and wet periods, which also made it impossible to extract the appropriate data for our purposes.
Despite these methodological problems, it was possible to (a) compare ecological responses before and after hydrological extreme events (cf. Figure 4); (b) compare ecological responses between different groups such as fish and invertebrates, or droughts and floods (cf. Figure 5); and (c) compare ecological responses between three event severity classes. When considering all samples together, CS showed statistically significant decreases in ecological metrics after the peak of the event, most frequently for invertebrates; for example, invertebrate abundance, density, and richness were significantly lower after the flood than before the flood. This is consistent with the findings of Greenwood and Booker (2015), who showed an overtime increase in invertebrates taxa richness after a flood from 22 years of data over 66 sites in New Zealand, inferring lowest values occurred immediately after the flood.
When comparing subgroups, sample size was often insufficient to show statistically significant responses. However results highlighted (a) higher magnitude of decrease in invertebrate density for floods than for droughts, and (b) a large decrease in abundance and richness of invertebrates compared to a small increase for fish following flood events.
Very few published studies explore the response of both floods and droughts for comparison with our conclusions, but Suren and Jowett (2006) found a decrease in invertebrate density was more common after floods than droughts based on five discrete flood and low-flow events in a New Zealand river, while Lake (2000Lake ( ,2003 suggested low biota resistance to floods, high resistance to seasonal droughts, and medium to low resistance for supraseasonal droughts. This is consistent with our conclusions of higher decreases of invertebrate density after floods than droughts, albeit from a much smaller drought sample  Oliveira, Cortes, and Albuquerque (2004) and Meffe and Minckley (1987) reported sharp post-flood decreases in taxa richness and density/abundance for invertebrates and very little effects on fish. The results of the study of Nislow, Magilligan, Folt, and Kennedy (2002) show a more complex pattern, with benthic invertebrate densities generally decreasing following the flood and salmonid responses strongly depending on the age: habitat change-triggered positive effects on overyearling fish compared to greatly diminishing numbers of age-0 salmonids.
No statistically significant ecological response to extreme event severity class (low, medium, or high) was identified, but this might be caused by the very small sample size available for the analysis, except the effects of floods on invertebrates. Lack of robust quantitative relationships between flow and ecosystems had been reported in the past (Poff & Zimmerman, 2010;Jones & Petreman, 2013;McManamay et al., 2013;Nislow et al., 2002). Poff & Zimmerman (2010) found that the size of flow alteration was not correlated with subsequent ecological responses, which varied among the different taxonomic groups, and Jones and Petreman (2013)  It is an inherent challenge of most ecological reviews that individual studies composing the analysis were not designed specifically to address the research questions posed in a review (Poff & Zimmerman, 2010). This could be due to a lack of common experimental methodology for investigating the ecological effects of floods or droughts that is approved and used by the majority of researchers. In consequence, empirical design of individual CS from our database was heterogeneous. For example, the time lags between event occurrence dates and pre-event or post-event sampling dates were highly variable across studies, ranging from days to years for pre-event samplings and from days to months for post-event sampling. We were often faced with necessity to select sampling dates based on expert judgment, for example, when there were many pre-or post-event samplings and none of them were clearly indicated by the authors as "reference" or "impact." This heterogeneity can be partly explained by the stochastic nature of hydrological extreme events, which makes it difficult to plan field surveys in advance, with capturing of a series of extreme events within a long-term sampling data set sometimes fortuitous (Woodward, Bonada, Feeley, & Giller, 2015). In consequence, very few studies investigating ecological responses to floods and droughts follow the rigorous Before-After Control-Impact design (Edwards et al., 2012).
Another point is that insufficient methodological detail in ecological papers hampers systematic reviews (Haddaway & Verhoeven, 2015).
Failing to report sampling dates, extreme events occurrence dates, or quantitative indices of their severity are typical examples encountered in our review, increasing the uncertainty of our assessment.
Because floods and droughts are natural phenomena, part of the expected variation in the hydrological cycle (although they may be exacerbated by anthropogenic-driven climate change), one could question whether they are "harmful" to ecosystems. There is evidence that droughts eliminate weak individuals and prevent invasive species, and so can have a positive impact on the ecosystem (Everard, 1996).
Both droughts and floods may also be favorable for fish reproduction and recruitment (Keaton, Haney, & Andersen, 2005;Cattaneo, Carrel, Lamouroux, & Breil, 2001), and floodplain inundation may also lead to short-and long-term increases in ecological metrics of invertebrate assemblages (Ballinger, Nally, & Lake, 2005). Furthermore, even when the effects are "harmful", that is, biota and ecological processes have been greatly diminished after the disturbance, they often have sufficient capacity to recover (Lake, 2011). Many organisms, such as microbes, may return to a river within a few weeks of a drought terminating; the following year, higher plants (Wright et al., 2002) and macroinvertebrates (Wood & Petts, 1999) can recover, whereas reduction in fish numbers may persist for five or more years (Elliott, Hurley, & Elliott, 1997). So provided that another drought does not occur within this period, the ecosystem can normally recover, although Holmes (1999) found that some plant communities shifted permanently after drought, and never returned to predrought conditions. Death et al.
(2015) stated that the recovery of the biota from extreme flood events can be quick provided that instream habitat is not dramatically affected (then recovery would be much slower, if at all). Woodward et al. (2015) reported that most invertebrate populations returned to their predisturbance state within 3 years after a catastrophic flood that triggered a 10-fold decrease in abundance, although for some it took up to 10 years. It should be noted that because our study focused on direct, immediate effects and responses (resistance), investigating resilience and recovery was beyond its scope.
Further steps building on the outcomes of this work could include a more in-depth analysis of case studies for which collected evidence was the most abundant, that is, the effect of floods on invertebrate density. This could even include a more formal meta-analysis, provided that the effect sizes were additionally estimated for each perturbation. In the case of fish and/or drought CS, where evidence was more modest, it should be considered to extend the geographical coverage of review to the global scale. Another direction is a focus on recovery/resilience rather than pure resistance of biota. Further progress in synthesizing evidence on the ecological role of floods and droughts in Europe can also be achieved in a different way: by carrying out comprehensive flume studies across a range of physiographic conditions using a multi-factorial design allowing to control other factors than solely the hydrological stress, such as it has been on the ecological role of floods and droughts can also be achieved in a different way: by carrying out comprehensive flume studies across a range of physiographic conditions using multi-factorial experiments planned in the MARS project (Hering et al., 2015).

CONCLUSIONS
In this study, we synthesized knowledge on the direct responses of fish and invertebrates to flood and drought events in European rivers and streams. Systematic review methods were employed to collect evidence from existing ecological literature, and hydrological techniques used for extreme event estimation were used to classify the severity of floods and droughts from the identified papers. While the resulting database is a significant product in itself, this study pinpointed the research gaps where no or very little evidence can be synthesized at this stage (e.g., the effect of drought on fish), as well as the more widely researched areas that would benefit from more in-depth quantitative analyses (e.g., the effect of floods on invertebrates). It was demonstrated that the studied metrics (abundance, density, richness, and diversity) experienced statistically significant decreases following extreme events in a number of cases, particularly for invertebrate responses to flood (higher significance) and drought (lower significance) events. Lack of significance for the effect of floods on fish shows, on one hand, that the identified responses in studied metrics were both increasing and decreasing. On the other hand, this result should be treated with caution due to a relatively low number of case studies, compared to invertebrates. Furthermore, a comparison of ecological responses between different subgroups showed that (a) the responses in invertebrate abundance and richness were more negative than the corresponding responses in fish following flood events, and (b) invertebrate density decreased more after floods than after droughts. Finally, contrary to our expectations, the severity class of extreme events was either not found to be an important factor influencing ecological metrics, or the number of studies was too low to perform such analysis (in most cases for droughts and for fish). Conceivably, other factors such as hydromorphology, biogeographical region, river size, or inhomogeneity between studies could mask any existing relationships between severity and response. Thus, the call of Lake (2000) for quantification of disturbance-ecosystem relationships: "If we are to progress and usefully compare both disturbance impacts and the consequential biotic responses, we need quantifiable measures of the disturbances ( … ), of the effects on abiotic and biotic components ( … ), and of the subsequent responses by the biota." remains as valid and urgent as ever.
Hopefully, this paper also provides useful insights for future ecological studies regarding the type of information that should preferably be reported so that future evidence-based reviews could benefit from a more consistent material.