Regional flow–ecology relationships in small, temperate rivers

Flow–ecology relationships within river systems are an important area of ongoing investigation, because of potential applications such as understanding the ecological impact of flow alteration at modified sites. This study analyses relationships between flow characteristics and benthic macroinvertebrates from 18 streams of similar size and typology within Northern England, to develop quantitative flow–ecology relationships applicable at regional scale. High and low flow event frequencies displayed statistically significant relationships with the ecological metrics of LIFE Score, Shannon's Diversity and a velocity flow affinity trait score. Results suggest that flow event frequencies have a significant role in influencing ecology within the river network system. Hence, this indicates that future flow regime design in the region may be enhanced if this variable is considered.


| INTRODUCTION
A global increase in water demand and energy requirements has led to the widespread proliferation of flow impoundments. The resulting flow modification, even by small impoundments and hydropower schemes, can adversely impact riverine ecology (Anderson et al., 2017;Poff et al., 1997), and despite recent efforts, there remains a lack of consensus as to how ecological impacts arising from flow regime change should be mitigated (Gillespie, Desmet, et al., 2015). A better understanding of the relationship between ecology and flow regime is therefore a critical area of investigation. Such understanding is imperative for the design of mitigation measures such as environmental flows (e.g., Hough et al., 2019), defined by the Brisbane Declaration, 2007 as '... the quantity, timing, and quality of water flows required to sustain freshwater and estuarine ecosystems and the human livelihoods and well-being that depend on these ecosystems.' (Overton et al., 2014, p. 861).
Several theoretical frameworks describe the relationship between riverine ecology and the flow regime (e.g., Junk et al., 1989;Poff et al., 1997;Vannote et al., 1980), and there is substantive and growing evidence to show how components of the flow regime, such as the timing, magnitude, frequency, duration and variability of flow peaks, can influence a range of ecological metrics (e.g., Praskievicz & Luo, 2020).
Magnitude is seen as a significant influence in the river system because of its effects upon river morphology, river habitat, sediment and nutrient transport, and physical forcing upon biota (Power et al., 1995). Timing is also because of morphological and behavioural adaptations of biota (Lytle & Poff, 2004). Frequency, duration and variability are likewise influential, because of their impact on nutrient cycling (Junk et al., 1989) or role as biological filters (Rolls et al., 2012).
Previous studies have discussed the challenges presented by rivers as open systems and the degree of uncertainty often associated with studies investigating specific variables (Konrad et al., 2011), when attempting to better understand flow-ecology relationships and possible mitigation of ecological impacts (e.g., arising from flow modification as a result of impoundments). The challenge is further enhanced because of the conflicting interests of multiple stakeholders present in most systems (Summers et al., 2015), such as water utility companies, industry and the general public. Developing mitigation measures that satisfy multiple stakeholders, while also making sufficient provision for environmental requirements such as ideal flow times and volumes for the system's biota, is a difficult task.
The building block approach (King & Louw, 1998) has been widely used to determine environmental flows as a means of mitigating the impact of modified flows, for example, for the design of flows downstream of impoundments. However, this site-specific, intensive approach relies upon expert judgement, which is impractical for mitigating the impacts of the majority of smaller-scale systems (i.e., flows >5 m 3 /s), which are widespread and frequently failing to meet legislated ecological targets (Voulvoulis et al., 2017). Thus, there is a need for general and transferable information about flow-ecology relationships to support flow design in such systems.
This study focuses upon the relationship between flow and ecology at a regional level, with the aim of informing future mitigation recommendations. Specifically, we consider flow-ecology implications within smaller-scale river systems as an area in need of further research (Voulvoulis et al., 2017). Studies continue to affirm the use of regionalscale efforts (O'Brien et al., 2018) rather than site-specific evaluation, as such work can offer significant scientific value and act as a first step towards designing mitigation measures within impacted systems without detailed and expensive site investigation (Hough, 2020;Poff et al., 2010). Flow-ecology trends identified at a regional level, based on considering combined datasets from different sites, may allow for the establishment of transferable environmental flow principles between sites of similar character (Arthington et al., 2006), which may increase the number of sites meeting legislated targets.
This study thus aims to make first steps towards addressing the needs of smaller-scale riverine systems by developing a flow-ecology model applicable at a regional scale. We utilize and agglomerate historic long-term flow and ecological datasets across sites in the north of England to identify ecologically-influential flow characteristics at a regional level. Such data are freely available and thus allow for analyses that are not too resource-or time-intensive, maximizing transferability. Analyses were performed on river systems of similar characteristics in order to reduce the likelihood of noise from uncontrolled sources of variation obscuring observable relationships (Konrad et al., 2011) and allow clearer examination of a range of hydrological drivers; magnitude of flow in particular may overwhelm other hydrological drivers when assessed across too broad a scale, because of its dominant influence upon hydraulics and morphology (Monk et al., 2006). This investigation therefore focuses on rivers of a similar magnitude of mean daily flow and physical character, located across the region of Northern England.
The study also focuses on functional, as well as taxonomic, measures of ecological community structure. Focusing upon taxonomic composition alone may not detect some influences that flow exerts upon ecosystems, such as in cases where composition is altered but overall richness is not (Chinnayakanahalli et al., 2011). A broader suite of metrics is therefore required to fully assess ecological impact (Arthington et al., 2018). In this study, we combine diversity and trait characteristics with ecologically important flow metrics to identify the strongest flow-ecology relationships within the region studied.

| METHODS
This study utilized Indicators of Hydrologic Alteration (Richter et al., 1996) derived from historical flow data, in order to identify hydrological characteristics at each site. Sites were characterized ecologically based on macroinvertebrate diversity and flow preference.
Relationships between flow and ecology metrics across all the selected sites were analysed using multiple linear regression.

| Site selection and data
Sites were selected from a range of sites across Northern England Selected study sites ranged from 0.31 to 2.83 m 3 /s annual mean daily flow, with a minimum of 5 years continuous flow and ecological sampling data, with samples in both seasons each year. When identifying appropriate sites, some were also excluded because of external factors that could influence invertebrate composition, such as poor water quality. The sites selected for study were of 'good' chemical quality according to the most recent EA assessment. Eighteen sites were selected for analysis. They were all low gradient, straight or low sinuous, alluvial reaches on a sandstone and/or mudstone bedrock. Most were unmodified reaches in agricultural areas, although some reaches were in urban or suburban settings with some channel modification (see Appendix 1). Site characteristics were obtained from EDINA Digimaps Ordnance Survey Service (2020) and Google Earth Pro.
Publicly available time series datasets were obtained from the EA and the Centre of Ecology and Hydrology National River Flow Archive (CEH, 2021). Flow data were in the form of mean daily flows. The time series of flow data varied from 12 to 56 years of continuous data between sites; with 10 sites having over 30 years of data. Appendix 3 addresses potential concerns relating to the use of time series of varying lengths. Ecological data, collected as part of EA routine monitoring, included taxon abundance at a species or family level and Loticinvertebrate Index for Flow Evaluation (LIFE) scores (Extence et al., 1999), typically with samples taken in spring and autumn each year and spanning 5-10 years. The coordinates of the data were checked to ensure that the sites for the flow and ecology data had no significant intervening flow inputs such as tributaries between them.

| Data analysis
A number of ecologically relevant flow variables were obtained from the flow data, based on principles outlined by Richter et al. (1996) and using indicators advocated for within the hydrological community (Dunbar et al., 2010;Monk et al., 2006): Q10, Q25, Q50, Q95, standard deviation, range, annual maxima and minima, mean daily flows, and frequency and duration of high and low flow events. Statistical analysis of IHA variables was conducted to check that the length of time series data at each site was sufficient to generate stable and reliable flow statistics. Ecological data from each site were processed to provide velocity affinity and Shannon's diversity metrics for spring and autumn seasons; LIFE score was already available in EA data. LIFE is a widely used metric for the ecological monitoring of freshwater benthic macroinvertebrates based upon the flow affinities of macroinvertebrate species and families (Dunbar et al., 2010). Taxonomic diversity was used as a measure of ecological response between sites using the Shannon diversity index (H 0 ) for macroinvertebrate family data in spring and autumn: where s is the number of families present in the sample and p i is the proportional abundance of each family.
Because of variation in the taxonomic resolution of the invertebrate data (data varied between species and family level depending upon site and time of measurement), all data were converted to family level, and the mean annual family abundances were calculated for each site separately in spring and autumn samples.

| Velocity affinity
Velocity affinity has been utilized in a number of ecological analyses (Schneider et al., 2016, Conallin et al., 2010. It was used in this study because of its strong relationship with the flow rate, and it represents the expected response of biota to various flow conditions. Species preferences were taken from Bis and Usseglio-Polatera (2004). Preferences were assigned to families by taking the mean trait affinity value of all species present within that family, an approach justified by the general similarity of traits within families, as seen in other studies such as White et al. (2017). Each family was also sorted into particular categories of flow preference, F I G U R E 1 Locations (solid circles) of all study sites across the North of England described in Table 1 categories (e.g., very fast flow) were given higher weightings (see Table 1) because of the fact that taxa possessing extreme traits tend to be less common in typical conditions, yet the presence of even small numbers of such taxa is suggestive of a system's character (Petchey & Gaston, 2006). Generally across sites, species preferring medium flows were prolific, and thus, weightings were used to better demonstrate fluctuations in functional distributions. Flow velocity categories were each given a score between 1 and 8. The abundances of families present in each category, relative to the total population, were multiplied by the weighted score. The sum of these values constituted the overall trait score, that is, a trait score of '1' indicates a site dominated by lentic flow affinity species, whereas '8' indicates that fast flow affinity species dominate.
Many families, while having some affinities for either high or low flows, also exhibited moderate affinities for a range of flows and therefore may be considered rather generalist with regard to flow preference. These were put into two categories; generalists with lowmedium preferences, and generalists with medium-fast preferences, demonstrated in Table 1. At low-medium flows, most families in the sampled regions appear to be generalists, with those of specific lowmedium affinity being very rare. As such, the weighting for the lowmedium affinity was weighted the same as the low-flow affinity, which was also rare at most sites. Trait scores varied between spring and autumn seasons because of differing family populations between the two periods, and thus, ecological metrics were assigned to both seasons separately.

| Flow variables and relationships
Using the data across all selected sites, a principal components analysis (PCA) was undertaken to reduce redundancy among the hydrological variables. PCA is a method commonly used in redundancy analysis and the approach followed Monk et al., 2006, and Chinnayakanahalli et al., 2011 PCA was based on a Pearson product moment correlation using the metrics listed in Table 2 and performed using R version 3.2.4 (R Core Team, 2016). Variables were sorted into distinct groups based upon the strength and direction of vectors within in the PCA biplot.
The biplot distinguished two groups within the variables which were labelled 'magnitude' and 'temporal' (Figure 2 in Section 3). The groups were used to identify redundant variables, as variables within the same group were correlated and were considered to have a high degree of mutual explanatory power in relation to the dependent variable. Thus, multiple variables from the same group were not used in subsequent regression modelling.
T A B L E 1 Trait score categories and associated weightings All metrics utilized are described in Table 2 below: For each data matrix, multiple linear regression was used to fit a regionally applicable model for each ecological trait within each season. Regression models were created for all combinations of nonredundant variables (i.e., all combinations of variables that would contain one 'magnitude' and one 'temporal' variable), along with each variable individually (as univariate models).
Model fitting was performed for each ecological dependent variable with combinations of flow variables as the independent variables.
The best fitting models for each dependent variable, in spring and autumn, respectively, were determined. These were judged from p values, R 2 values, and as the primary deciding factor, the Akaike information criterion (AIC); a measure of the relative quality of a statistical model, taking into account both the variation explained and the model complexity (Aho et al., 2014). Variables above a p value threshold of 0.2 were not analysed further to find their R 2 and AIC values, because of their obvious lack of statistical significance.

| RESULTS
Calculation of all ecological metrics was possible for all sites except one, where missing data meant that metrics could not be derived.
Hydrological and ecological metrics for each site are listed in Table 3.  Table 4. The full list of models and associated statistics can be found in Appendix 2.
A number of statistically significant relationships were identified at regional scale, with all the best fitting models containing only one flow variable. These relationships are plotted in Figure 3. Mean annual

| DISCUSSION
In this study, we have examined the degree to which there are general relationships between hydrological characteristics and ecological metrics, across a set of similar rivers in Northern England..

| Velocity preference trait and LIFE scores
The results suggest that in this region, high flow event frequency has a significant influence upon the functional composition of a system in terms of velocity preference of families, explaining 20%-27% of the variation in preference when considering trait score, and 16%-26% of variation when considering LIFE score (based on R 2 values). This suggests that it may be possible to identify particular aspects of the flow regime which could be important to focus on when developing potential mitigation solutions for flow alteration. IHA variables including the duration and frequency of high and low flow events were found to strongly influence stream macroinvertebrates in a similar study based on the ELOHA method in the United States using biological metrics primarily based on functional group composition such as measuring the percentage of individuals adapted for filter feeding (Buchanan et al., 2013). The mechanisms underpinning the positive relationships between high flow and flow preference and LIFE scores seem likely to be straightforward; the more frequently high flows occur, the more resilient the community at a site becomes in terms of functional composition.
The influences of high flow event frequency as an ecological driver may have significant implications when considering environmental flow regime design in the region and also suggest significant limitations in current 'fixed' hands-off flow-based regulations (Arthington et al., 2006). A lack of high flow events within a modified system may lead to a lack of an important biological filter, resulting in systems being dominated by species that are highly competitive within a steady, moderate-to-low flow environment, as discussed by a number of studies examining river deviation from natural flows (Lytle & Poff, 2004;Summers et al., 2015). Incorporating a moderate frequency of high flows events into environmental flow regimes to mitigate the impacts of modification through impoundments may serve to balance a system's functional composition and be one facet in ensuring a stable and diverse ecosystem.
The only detected effect on family diversity was that low flow event frequency is negatively related to diversity, but in spring samples only. This may be because of differing conditions between the two seasons; functional composition is likely to differ significantly between the two seasons, either because of life history or external drivers. As such, response to the flow modification may vary because of these differences in composition between seasons. A negative correlation between low flow and diversity is consistent with other studies (e.g., Pardo & Garcia, 2016) If the influence of low flow event frequency is general, this could have significant implications for water managers wishing to increase biodiversity within managed systems. Low flows play a key role within natural river systems (Poff et al., 1997;Richter et al., 1996), and it would therefore be expected that such events would aid in regulating the ecosystem, preventing the dominance of certain species.  Nichols et al., 2006). This study also affirms the suggestion of Chinnayakanahalli et al. (2011) that taxon richness and functional composition respond differently to flow alternation, and using only one of these metrics may fail to recognize significant changes within the ecosystem and that a broader suite of ecological metrics are required in order to fully evaluate changes within the ecosystem (Arthington et al., 2018;Poff et al., 2017).
Results from this study are likely to have implications for water man-

This research has been supported by an Engineering and Physical
Sciences Research Council research studentship (EP/LO15412/1) and United Utilities as part of the STREAM industrial doctorate centre for the water sector. We thank two anonymous referees for comments which helped clarify and improve the manuscript.
To submit a comment on this article, please go to http://mc. Site data and geology: 6 years, this seems a good possibility-given that a shorter dataset will become increasingly less resilient to the influence of such events.
Colne ( To conclude on the results of this testing, we believe that there is good evidence that the length of the time series carries only a minor impact on calculated IHA metrics, justifying the approach used. We would also mention that overly shortening time series data would theoretically decrease the resilience of our metrics to extreme events, meaning that longer time series would be expected to better characterize the general hydrological character of each site, and hence we have used as much data as was available for each site.Graphs providing a visual illustration of differences between datasets for each metric follow.