A sediment-speciﬁc family-level biomonitoring tool to identify the impacts of ﬁne sediment in temperate rivers and streams

Anthropogenic modiﬁcations of sediment load can cause ecological degradation in stream and river ecosystems. However, in practice, identifying when and where sediment is the primary cause of ecological degradation is a challenging task. Biological communities undergo natural cycles and variation over time


Introduction
Streambed deposited fine sediment (<2 mm) is an important, natural component of freshwater ecosystems and is critical for habitat heterogeneity and ecological functioning (Owens et al., 2005;Wood and Armitage, 1997;Yarnell et al., 2006). However, anthropogenic activities can alter sediment delivery and dynam-chemical (e.g. pH, heavy metals, pesticides, nutrients) and biological pressures (e.g. invasive species), (Clews and Ormerod, 2009;Liess and Schulz, 1999;Moore and Ramamoorthy, 2012;Sousa, 1984;Townsend, 1996). The use of conventional (non-biological) monitoring approaches to attribute cause and effect necessitates measurement of multiple parameters, at sufficient temporal resolution, and for a significant period of time (Grove et al., 2015;Harris and Heathwaite, 2012). Biomonitoring tools, which use changes in the presence, abundance or behaviour of biota to indicate, gauge and track changes in the environment (Friberg et al., 2011;Gerhardt, 2000;Wright et al., 1993), can provide a valuable lowercost alternative to conventional monitoring. Ideally, for reasons discussed in more detail in the following paragraphs, biomonitoring tools should, where possible: (i) have a biological basis, (ii) be easily integrated into standardised biological sampling and recording methods, (iii) be developed and tested over sites from the full range of river and stream ecosystems to which they are intended to be applied, and (iv) have a strong relationship with the pressure of concern.

Biological basis
There is an emerging consensus among those involved in developing biomonitoring tools, that more reliable tools have a biological basis, i.e. they use biological and ecological traits that influence the tolerance of organisms to a given pressure, and are linked to ecological niche theory (Bonada et al., 2006;Friberg et al., 2011). One of the reasons for this is that these types of tools are not vulnerable to statistical artefacts that may affect purely statistical models. The biological traits (e.g. respiration, locomotion, dispersal, feeding) of taxa in a community reflect the spatial and temporal variations in the environmental factors of a habitat, that act as "filters", with successful combinations of traits enabling survival and reproduction (Poff, 1997;Statzner et al., 2001b;Townsend et al., 1997). Certain traits or combinations of traits can result in sensitivities/tolerances to particular environmental pressures and therefore these traits have the potential to be used in biomonitoring to discriminate between types of human disturbance (Statzner et al., 2001a). If taxa are selected for inclusion and their sensitivities weighted based on the biological and ecological traits, which influence their sensitivity to a given pressure, the resultant biomonitoring tool will have a mechanistic linkage (rather than a purely correlative linkage) between the pressure of concern and the biotic response (Friberg, 2014).

Easily integrated into standardised biological sampling and recording methods
One of the many benefits of biomonitoring as opposed to conventional monitoring of multiple environmental parameters is that the biological data collected from the same biological sample can be interpreted by an array of different biomonitoring tools to identify potential pressures. In order for this efficiency and costeffectiveness to be realised, the biomonitoring tools must be able to make use of data collected using standardised biological sampling methods and recorded to standardised, minimum taxonomic levels. Monitoring agencies in different countries commonly carry out invertebrate identification and recording at different taxonomic levels. For example, within Europe the level of invertebrate taxonomic knowledge varies between countries, as do resources, resulting in different taxonomic resolutions and a lack of comparability between data (Hering et al., 2010;Schmidt-Kloiber and Hering, 2015). Within the UK, England and Wales recently moved from family level invertebrate recording to a mixed level, 1 consisting of family-, genus-and species-level identifications, with the majority of taxa being identified to species (Davy-Bowker et al., 2010). However, Scotland and Northern Ireland currently record at family-level. 2 A family-level tool is not only essential for those countries that record at this taxonomic level, but is also crucial for those countries which have recently switched to a higher taxonomic resolution, as it will allow for retrospective analysis of historic family-level data. This ability to assess the historic conditions at sites is particularly important when attempting to distinguish between natural temporal variations (Resh et al., 2005) and those caused by anthropogenic activities, as this can require numerous years of baseline data. Furthermore, a lowercost family-level biomonitoring tool is likely to be highly desirable for non-regulatory work (e.g. river restoration projects, aquatic research and citizen science programmes) where budgets and taxonomic expertise may be more limited.

Developed and tested over the full range of different rivers and streams
There are a number of reasons why biomonitoring tools should be developed and tested using data from sites across the full spectrum of rivers and streams to which they are designed to be applied. Firstly, biological communities and species distributions vary naturally, partly as a result of environmental gradients, biotic interactions (McGill et al., 2006) and their ecological requirements (Schmidt-Kloiber and Hering, 2015). As such, a biomonitoring tool must incorporate a range of taxa to ensure that any sampled site will have the potential to include a sufficient number of taxa with sensitivity weightings. Secondly, it cannot be assumed that biological communities will respond uniformly to the same pressure in different rivers and streams. Not only may the pressure of concern occur alongside other, different pressures (potentially having synergistic or antagonistic effects), (Folt et al., 1999), but differences in environmental characteristics (e.g. habitat complexity) may also affect a biological community's resistance and resilience to a specific pressure (Dunbar et al., 2010a,b;Lake, 2000).

Strong relationship with pressure of interest
Given the implications of incorrect assignment of ecological status of streams for both water and land managers (from unjustified burdens being placed on the users of water resources, to environmental damage going undetected), it is important that biomonitoring tools have a strong relationship with their pressure, in addition to the previous three criteria. Nevertheless, a strong correlation does not rule out the possibility of the tool indicating other aquatic pressures or variables that may occur in parallel with the pressure of concern. A statistical approach may yield strong correlations to the pressure of concern, but these can be the result of statistical artefacts e.g. an inadvertent relationship with an associated pressure (Table 1). Although a biological basis provides a mechanistic linkage for a correlation, biological traits are not always unique to a particular pressure, and as such the influence of confounding pressures also cannot be ruled out (Schuwirth et al., 2015).  Matthaei et al. (2010), Petticrew et al. (2007), Slattery and Burt (1997), Tockner et al. (1999), Wood and Armitage (1999) Decrease/increase in abundance, dependent on the ecological requirements of macroinvertebrates (Extence et al., 1999).

The Proportion of Sediment-sensitive Invertebrates (PSI) index
The PSI index is a biomonitoring tool that is designed to identify the degree of sedimentation in rivers and streams (Extence et al., 2011). Invertebrate sensitivity ratings (A-Highly sensitive, B-Moderately sensitive, C-Moderately insensitive, D-Highly insensitive) were assigned following an extensive review of the literature, and using expert knowledge of ecological and biological traits, thus providing the tool with a sound biological basis. Traits that may result in sensitivity to fine sediment include feeding, locomotion and respiratory attributes. For example, scrapers/grazers and passive filter feeders may experience decreased food availability or damage to feeding appendages as a result of deposited fine sediment (Larsen and Ormerod, 2010;Nerbonne and Vondracek, 2001). The PSI index has both family-level and species-level versions, allowing it to be easily integrated into standardised biological sampling and recording methods. However, an evaluation of these versions of the index across a wide range of temperate river and stream ecosystems (Turley et al., 2014), showed it to have moderate correlations with fine sediment (r s = −0.61, p < 0.01 and r s = −0.64, p < 0.01 respectively), that although comparable with the average invertebrate-based, pressure-specific tool, used throughout the EU (Birk et al., 2012), limits confidence in its application. Recent work has demonstrated enhancements to the species-level PSI index through the use of empirical data to supplement the expert knowledge and literature, which were used to determine the original four fine sediment sensitivity ratings (Turley et al., 2015).
The aim of the present study was to investigate if similar empirical-weighting techniques could be used to enhance the per-formance of family-and mixed-level versions of the PSI index, whilst retaining their biological basis, and ensuring that the tools are compatible with historic and contemporary datasets recorded to family-level and/or mixed-level, and have a strong correlation with fine sediment.

Minimally-impacted sites
It was important that empirical data from minimally-impacted sites were used in this study, in order to reduce confounding pressures in the process of assigning sensitivity weightings. In this instance, a freely available dataset from the UK was utilised. However, this study can be seen as an exemplar to be used in other countries where such datasets exist.  Wright (2000) and Clarke et al. (2003), but is summarised here. The database contains invertebrate, water quality and catchment characteristics data, recorded at each site over at least one year, between 1978 and 2004. The 835 reference condition sites, on streams and rivers across the United Kingdom ( Fig. 1), encompass a wide range of environments, varying in their (i) climate -mean annual precipitation totals between 1961 and 1990 of 430-2930 mm and mean annual temperatures between 1961 and 1990 ranging from 7.93 to 11.45 • C, (ii) geology -varying from catchments dominated by hard igneous rocks to catchments dominated by soft sedimentary rocks and (iii) topography -altitudes at river source varying from 5 to 1216 m above sea level. The stream and river sites also vary in their morphometry with widths ranging from 0.4 to 117 m and average depths ranging from 0.02 to 3.00 m (widths and depths are a mean of three seasonal measurements). All of the sites are considered to be as close to reference condition as it is possible in the United Kingdom, and they have no, or only very minor, anthropogenic alterations to the values of the chemistry and hydromorphology, supporting biota usually associated with such minimally-disturbed conditions.
The 835 sites were sampled for macroinvertebrates using the UK standard method; a standardised three-minute kick sample technique using a 900 m mesh hand net, followed by a one-minute hand search. All in-stream habitats identified at the site were sampled in proportion to their occurrence (Environment Agency, 2009). The database has abundance records of different macroinvertebrates identified to (i) family level, (ii) mixed level, and (iii) species level (Wright, 2000). Each site has season-specific records of community composition: spring (March-May), summer (June-August) and autumn (September-November). As the technique used for macroinvertebrate sampling was semi-quantitative, model development and testing utilised log abundance data rather than raw abundance data or presence/absence data. This recognises the potential importance of changes in abundance when attempting to identify pressure gradients, without placing spurious confidence in the semi-quantitative sampling technique (Furse et al., 1981).
The fine sediment data within the RIVPACS IV database consisted of visual estimates of the percentage of the substrate cover composed of silt and clay (<0.06 mm) and sand (≥0.06 and <2.00 mm), as an annual average based on three seasonal measurements. The visual estimate method used to collect these data is described in the River Habitat Survey Field Survey Guidance Manual (Environment Agency, 2003). Briefly, it involves the operator carrying out a visual inspection over a given reach, estimating the substrate composition and recording the percentage of bedrock, boulders and cobbles, pebbles and gravel, sand, and silt and clay.
Family and mixed level macroinvertebrate data, and site substrate cover were extracted from the RIVPACS database and compiled in Microsoft Excel. The percentage of the substrate cover consisting of sand, silt and clay, were combined to provide an overall estimate of fine sediment (<2 mm) for each of the 835 minimally-impacted sites. The dataset was then split into a training dataset (751 sites, three seasons, n = 2252) and test dataset (84 sites, autumn season only, n = 84) using random allocation, to ensure testing of the indices could be considered independent of the development stages. Similar to Kelly et al. (2012), this 90:10 split was chosen to ensure that the indices were developed using a significant number of sites, whilst leaving a suitable number for independent testing.

Agriculturally impacted sites
The family level indices were also tested across a continuum of agriculturally impacted stream sites (upstream catchment land-use ranging from 0 to 90% arable fields, n = 754) in England, using part of a dataset described by Pearson et al. (2016). In brief, this dataset included seasonal macroinvertebrate data, seasonal fine sediment data, and water chemistry data (mean of the preceding 12 months) that were collected during routine monitoring by the Environment Agency in 2006. For the present study, more detailed macroinvertebrate abundance data were obtained from the Environment Agency, for samples collected in the autumn, along with fine sediment data based on the mean of spring and autumn visual estimates, and Environmental Quality Ratios (based on the observed and expected scores-calculated using RICT) for an index designed to indicate the effects of organic pollution (WHPT index, Paisley et al., 2014). All 754 sites used in the present study were located on independent water bodies, as detailed by the water body names and grid references provided by the Environment Agency.

Developing the Empirically-weighted PSI (E-PSI) indices
For both the family level and mixed level E-PSI indices, the fmincon function (active-set algorithm), a nonlinear optimisation method of MATLAB (Mathworks, version R2014a), was used to test up to 100,000 iterations of taxon sensitivity weightings, to find an optimum set of weightings that resulted in the strongest Spearman's rank correlation coefficient between index scores and fine sediment in the training dataset. The Spearman's rank correlation was used, as the data were found to be non-normally distributed and show heteroscedasticity, and could not be successfully transformed. In order to maintain the biological basis of the original PSI index, those taxa with PSI ratings of "highly sensitive" and "moderately sensitive" were constrained to sensitivity weights between 0.50 and 1.0, and those rated as "moderately insensitive" and "highly insensitive" between 0 and 0.49. Using a range of weights acknowledges the breadth of ecological niches that macroinvertebrates occupy, as well as their differing potential as indicators. A modified version of the original PSI index's equation was used to calculate E-PSI scores: Formula used to calculate E-PSI scores. The sum of each sensitive taxon's log abundance (log A sens ), multiplied by the corresponding sensitivity weightings (W), is divided by the sum of all taxon log abundances (log A all ) multiplied by the corresponding sensitivity weightings (W). This value is then multiplied by 100 to provide the E-PSI score. Note: Log abundance categories in E-PSI were simplified to: 1-9 individuals present = 1; 10-99 = 2; 100-999 = 3; 1000+ = 4. E-PSI scores range from 0 to 100; 0 representing a site dominated by fine sediment, 100 representing a site with minimal amounts of fine sediment.

Developing the family level E-PSI index
In total, 79 invertebrate families with PSI sensitivity weightings were present in the family level training dataset. Following the nonlinear optimisation procedure, an optimum set of family sensitivity weights formed the E-PSI fam index.

Adjustments for family-level sensitivity variations
In the process of developing the original, family-level PSI index, inevitably some generalisations were made regarding the sensitivities of invertebrate families to fine sediment. Families can comprise numerous genera and species, each with a range of different environmental requirements and sensitivities (Lenat and Resh, 2001), making assignment of sensitivities at family-level a difficult process. To determine whether all families identified as sensitive by the original PSI index were correctly identified as useful indicators at this taxonomic level, the empirical data was investigated. Using data from all 835 sites within the RIVPACS dataset, the percentage of fine sediment that corresponded with the 75th percentile of the family's total abundance was calculated for each taxon.
This involved using raw abundance invertebrate data, sorting all 2504 samples in order of increasing fine sediment, and calculating the cumulative abundance for each taxon. The 75th percentile of abundance was then calculated for each family, and this value or the next highest value was cross-referenced to obtain the corresponding sediment value for that abundance. This family-specific sediment value, represented the maximum fine sediment percentage at which at least 75% of the families abundance occurred. Similarly to Relyea et al. (2012), this 75% value was used as it recognises the fact that taxa can occur in sub-optimal conditions, often at reduced densities. These family-specific fine sediment values were used to inform the creation of a modified version of the family-level E-PSI index, by excluding taxa whose sediment value was ≥33%. The families that exceeded this threshold were removed from the index altogether (instead of being reassigned to the insensitive category) to maintain the biological basis. The removal of these families is not an admission of their tolerance to fine sediment, rather it acknowledges that when considering fine sediment at the reach scale, they are not useful as indicator taxa. The 33% threshold was used as pre-vious research showed that at the patch scale, sediment cover of approximately one third resulted in significant declines in invertebrate richness (Larsen et al., 2009). Once these taxa had been removed from the index, the nonlinear optimisation procedure was performed to return a new set of optimum sensitivity weights based on the new altered community composition/taxon list, forming the E-PSI fam69 index 2.2.3. Developing the mixed level E-PSI index In addition to the family level index, a mixed level version was developed using the same procedures. In total, 355 invertebrate taxa with PSI sensitivity weightings were present in the mixed level training dataset. Following the nonlinear optimisation procedure outlined in Section 2.2, an optimum set of sensitivity weights formed the E-PSI mixed index.

Testing of the E-PSI indices
The developed E-PSI indices were tested using both the minimally-impacted, independent dataset (n = 84), as well as the agriculturally impacted dataset (n = 754), both described in Section 2.1. The performance of these indices were evaluated and compared using their Spearman's rank correlations with fine sediment, due to the non-normally distributed and heteroscedastic data, and the inability to successfully transform. All correlations were interpreted using the Dancey and Reidy (2007) classifications of correlations; 0.1-0.39 = weak, 0.4-0.69 = moderate, 0.7-0.99 = strong.
In order to examine the relationship between E-PSI and fine sediment further, a separate test was conducted on the minimallyimpacted (RIVPACS) sites with discharge data (n = 443). The aim of this exercise was to restrict the potential collinearities between fine sediment and covariates relating to longitudinal gradients, particularly those relating to flow (e.g. discharge, velocity and slope). To achieve this, the sites were grouped by stream power (total stream power per unit stream length, W m −1 ) into five approximately equal groups (see Supplementary material B). The Spearman's rank correlations within each group between the E-PSI indices and fine sediment were then calculated.
Stream power ( ) was calculated using the following formula: = yQs where y is the specific weight of water (9810 N m −3 ), Q is the average water discharge (m 3 s −1 ), and s is energy slope (m m −1 , which can be approximated by the river slope), (Knighton, 1999).

Comparisons with other biomonitoring tools
The Lotic-invertebrate Index for Flow Evaluation (LIFE) (Extence et al., 1999) and the Combined Fine Sediment Index (CoFSI) (Murphy et al., 2015) were calculated and their relationship to both fine sediment, PSI and E-PSI indices were assessed in terms of their Spearman's rank correlation coefficients. LIFE was chosen to determine the influence of known interactions between flow and fine sediment deposition and re-suspension (Dewson et al., 2007). Correlations between LIFE, E-PSI indices and fine sediment were also compared within stream power groups.
The recently developed, predominantly species-and genuslevel CoFSI was included in this analysis, as it is a purely statistically based index designed to indicate fine sediment pressures (Murphy et al., 2015). All mixed-level indices were calculated using data of the same taxonomic detail, at the level used for Water Framework Directive reporting. The subscripted text identifies the taxonomic level (family or mixed) of the data and the number of taxa used in the calculations of the index. All correlations are significant at the 0.01 level (2-tailed). Mixed level indices could not be tested over the agriculturally impacted sites, as data at this taxonomic level were not recorded.

Biological indices and fine sediment
The best performing family-level and mixed-level E-PSI indices were strongly, negatively correlated to fine sediment metrics in the minimally-impacted dataset and similarly, but moderately correlated in the agriculturally impacted test dataset ( Table 2). The agriculturally impacted test dataset had WHPT scores of 2.5-8.0 and Environmental Quality Ratios of between 0.4 and 1.3 (see Supplementary material C) indicating that the sites were, at the very least, impacted to varying degrees by organic pollution.
The E-PSI indices with the strongest correlation with fine sediment in the minimally-impacted test dataset were the E-PSI fam69 and E-PSI mixed (r s = −0.72 and r s = −0.70, p < 0.01, respectively). The optimisation process was responsible for improving the correlations between fine sediment and the E-PSI fam index compared to the PSI fam index, in the minimally-impacted test dataset (r s = −0.66, p < 0.01 compared to r s = −0.59, p < 0.01). The largest improvement following the optimisation process was for the E-PSI mixed index in the minimally-impacted test dataset, which represented an improvement of 10 percentage points (r s = −0.70, p < 0.01 compared to r s = −0.60, p < 0.01). LIFE fam , and LIFE mixed had moderate correlations with fine sediment (r s = −0.57 and r s = −0.51, p < 0.01, respectively) and CoFSI had a strong correlation with fine sediment (r s = −0.72, p < 0.01).
Within the stream power groups (Table 3), the strongest correlation between E-PSI fam69 and fine sediment was r s = 0.71, p < 0.01 in the group of sites with low stream power, with moderate correlations in all other groups (ranging from r s = 0.59 to r s = 0.68, p < 0.01).

Adjustments for family-level sensitivity variations
By including only those taxa whose corresponding sediment value was <33%, the strength of the correlation between the E-PSI fam index and fine sediment was improved in the unimpacted test dataset, but had essentially the same correlation over the impacted sites. The exclusion process resulted in 10 families being removed entirely from the calculation of index scores. These sensitive families had corresponding sediment values of between 36% and 83% fine sediment. Three of these families had taxa rated as insensitive in the mixed level tool: Limnephilidae, Scirtidae and Gammaridae (for further information see Appendix A). Table 3 Spearman's rank correlation coefficients between the best performing E-PSI indices, LIFE and fine sediment, in the minimally-impacted dataset, using autumn data, and all sites with discharge data (n = 443). Sites are grouped by total stream power per unit length (W m −1 ). a Stream power groups 1 and 5 (n = 88), stream power groups 2,3 and 4 (n = 89). All correlations are significant at the 0.01 level (2-tailed), unless otherwise indicated, (ns) not statistically significant.

Assigned sediment-sensitivity weightings
The sensitivity weightings that formed the best performing E-PSI indices (E-PSI fam69 and E-PSI mixed ) in the minimally-impacted test dataset are shown in Appendix B and C. An index calculator is also provided in Supplementary material A.

Restricting the collinearities between fine sediment and longitudinal gradients
The Spearman's rank correlations between E-PSI and fine sediment within each of the five stream power groups were moderate to strong (r s = −0.56 to r s = −0.73) ( Table 3).

Comparisons between biomonitoring tools
In addition to comparing the E-PSI indices to fine sediment, the relationships between indices were considered. The various indices had strong correlations with each other, ranging from r s = 0.79 to r s = 0.98, p < 0.01 (Table 4). All versions of the E-PSI index had weaker correlations with LIFE fam and LIFE mixed , compared to the original versions of the PSI index.
The correlations between E-PSI and LIFE decreased in successively increasing stream power groups (Table 3), from r s = 0.92 to r s = 0.47, p < 0.01 for E-PSI fam69 versus LIFE mixed . The correlation between LIFE mixed and fine sediment was strongest (r s = 0.72, p < 0.01) in the group of sites with low stream power, and was weakest at sites with high stream power.

Identifying fine sediment pressures
The results of this study show that the use of empirical data to assign sensitivity weightings within the PSI index's original biologically-based sensitivity ratings, has improved the performance of the tools, in terms of their correlation with fine sediment. As a result, both the family-level and mixed-level E-PSI indices have correlation coefficients with fine sediment, that are stronger than the average invertebrate-based, pressure-specific biomonitoring tool used throughout Europe in the implementation of the WFD (Birk et al., 2012). Furthermore, by including only those taxa whose 75th percentile of abundance corresponded with a fine sediment value of <33%, the E-PSI fam69 index has a strong correlation with fine sediment similar to that of the mixed-level E-PSI index (E-PSI fam69 versus fine sediment: r s = −0.72, p < 0.01). Applying this threshold to the family-level index was intended to acknowledge the likelihood of significant within family variation in terms of sensitivity to fine sediment, and identify any families that were not useful as indicators at the reach scale. This exercise appears to have had the desired effect, reducing some of the variation in the relationship that may have been caused by varying sensitiv-ities within families. Testing of the E-PSI fam69 index also showed similar improvements and correlation (r s = −0.68, p < 0.01) over the agriculturally impacted sites, suggesting that the index performs similarly over sites that are impacted by the multitude of pressures associated with agricultural land-use.
Nevertheless, correlations do not prove causality and they should be interpreted with some caution, partly because many in-stream influences are associated with each other in riverine environments (Allan, 2004). The benefit of biologically-based indices such as the PSI and E-PSI index is that the biological basis provides a mechanistic linkage between index scores and fine sediment conditions. The PSI index utilises existing knowledge on biological and ecological traits and sensitivities to fine sediment, to assign sensitivity ratings, which the E-PSI index combines with empirical data to assign more detailed sensitivity weightings (0-1).
Based on the available data on the ecological preferences of macroinvertebrates from Schmidt-Kloiber and Hering (2015), those families, which have been assigned the highest sensitivity weightings in E-PSI fam69 (Heptageniidae, Perlidae, Aphelocheiridae, Chloroperlidae, Lepidostomatide, Leuctridae, and Perlodidae), are dominated by species with a strong affinity to coarse substrate microhabitats, with the exception of Lepidostomatidae that have some affinity to coarse substrates, but are more often associated with woody debris. Habitat preferences are determined by an invertebrates biological traits or combination of traits. These preferences relate not only to the direct physical properties of the habitat (e.g. the substrate), but also the flow velocities, hydrological and thermal regimes, resource availability and biotic interactions associated with them. These taxa with the greatest E-PSI sensitivity weightings are mainly clingers, with tarsal claws allowing them to "grip" the surface of pebbles, cobbles and boulders in shearing flows (Pollard and Yuan, 2010). Nevertheless, they depend on the heterogeneity of course substrates to provide refugia from these shearing flows, as well as from predators. Previous research has shown reduced densities and richness of clingers, with increasing fine sediment cover (Pollard and Yuan, 2010;Rabení et al., 2005). Additionally, the turbulence caused by coarse substrate and the resulting oxygenation is likely to be important for many of these sensitive families. For example, Aphelocheirus aestivalis is a plastron-breather and so requires well-oxygenated waters (Seymour et al., 2015), and fine sediment may also disrupt its respiratory functioning.

Confounding pressures
A considerable challenge for research carried out over large spatial extents is the occurrence of collinearity between the variable of interest and other environmental variables (Pearson et al., 2016). Relatively few studies have considered the interactions and cooccurrence of in-stream pressures, which are likely to affect the performance of biomonitoring tools. A recent study of 9330 sites throughout 14 European countries, found 47% of rivers (90% of lowland rivers) throughout Europe to be subject to multiple pressures Table 4 Spearman's rank correlation coefficients (2 d.p.) between biological indices, in the minimally-impacted, autumn test dataset (n = 84). relating to hydrology, morphology, water quality and connectivity (Schinegger et al., 2012). This highlights the importance of a mechanistic linkage between indices and sediment conditions, given the potential for these multiple pressures to confound biomonitoring approaches.
In the present study, sites were grouped using site-specific stream power to restrict the collinearities between fine sediment and longitudinal gradients, particularly those relating to flow. Our results show that flow (indicated by LIFE) was related to fine sediment cover, and as such LIFE and E-PSI are strongly correlated. However, this relationship between fine sediment cover and flow weakens as stream power increases, likely because after the critical thresholds for particle entrainment and transport have been exceeded, further increases in stream power cease to result in much further entrainment, transport and removal of fine sediment from the river bed (though the increases in stream power may trigger entrainment and transport of coarser particles as bed load). This is likely to be the reason why the relationship between LIFE scores and E-PSI, and LIFE and fine sediment weaken in successively increasing stream power groups. These results add confidence to our interpretation that although fine sediment cover is related to stream power (for physical reasons), and there will therefore be a moderate to strong relationship between scores derived from sediment-specific and flow-specific biomonitoring tools; these relationships are not fixed, and thus where one pressure (e.g. flow) is stable, whilst the other is variable (e.g. sediment cover), the scores derived from the biomonitoring tools will diverge and provide valuable information on the likely cause of deviation from reference-condition community composition. In the case of LIFE and E-PSI, a similar approach to model development for LIFE could result in even greater decoupling of the two tools.

Statistically based sediment-specific tools
As noted, other studies have attempted to develop fine sediment-specific indices, using empirical data and statistical methods to select taxa for inclusion and to assign sensitivity weights to these taxa, to achieve a correlation between a tool's score and the pressure of concern. These approaches have achieved moderate-strong correlations with fine sediment but lack a biological basis and mechanistic linkage, which are important for disentangling the multiple pressures in rivers and streams (Friberg, 2014). Murphy et al. (2015) developed CoFSI, an index that endeavours to take account of both the amount and organic content of fine sediment. The authors used partial canonical correspondence analysis to rank taxa in terms of their sensitivity to fine sediment, based on empirical data from 179 stream sites. The re-suspension method of quantifying fine sediment was used, averaging samples from two erosional and two depositional patches to achieve a contemporaneous reach-scale average. Despite this technique necessitating some subjective assessment of "erosional" and "depositional" patches, moderate to strong correlations (r s = −0.54 to −0.70, p < 0.05) were observed between the resultant sediment data and CoFSI in their test dataset (Murphy et al., 2015). Due to the difficulties of selecting sites with minimal confounding pressures that at the same time represent a gradient of sediment pressures, their resulting test dataset (n = 83) is somewhat geographically restricted and is focused on agricultural streams, which may be impacted by an array of different pressures typically associated with agricultural practices (Allan, 2004;Matthaei et al., 2010;Weston et al., 2004). In the present study, CoFSI is shown to have a similar correlation with fine sediment compared to the best performing family and mixed-level E-PSI indices. CoFSI is a mixed-level (predominantly genus-and species-level) biomonitoring tool, and as such cannot be applied to present and historic datasets recorded at family-level.
Most recently, in North America, Hubler et al. (2016) developed a statistically based index for Oregon streams (the Biological Sediment Tolerance Index; BSTI) using weighted averaging to assign tolerance values, and found a moderate correlation (r 2 = 0.58) between the index and fine sediment in their independent test dataset (n = 50). Similarly, Zweig and Rabeni (2001) found a moderate correlation (r s = −0.59, P < 0.01) between their Deposited Sediment Biotic Index (DSBI) and visual estimates of fine sediment, using data from four streams in Missouri. The authors used the sediment values that corresponded with each taxon's median abundance, to assign tolerance ratings, and used the same data for testing, therefore lacking an independent test dataset. In addition, the DSBI is currently limited by the small development dataset, which is likely to restrict its spatial applicability. A further example from North America is the Fine Sediment Biotic Index (FSBI), which again used empirical data to assign sensitivity weights to taxa (Relyea et al., 2012). The FSBI showed a large amount of variation, with the data forming a wedge-shaped response that has since been suggested as a typical response to multiple pressures (Friberg 2014); when sediment pressure is low, it has minimal impact on the invertebrate community and other pressures dominate. This wedge-shaped response is also observed in PSI, E-PSI and CoFSI (see Supplementary material D), supporting this generalisation. Some of the variation in these relationships is also inevitably associated with invertebrate sampling and quantifying fine sediment pressures. For example the standardised, three minute kick-sample method has been shown to collect only approximately 50% of the species and 60% of the families present in six replicate samples (Furse et al., 1981). The challenges of quantifying fine sediment pressures in a way that is biologically relevant and accurately represents spatial variation is a further limitation to the development and/or testing of sediment-specific biomonitoring tools. Visual estimates of substrate composition were used in this study due to the data being from an existing database, which had the benefits of a wide geographic coverage. Although the technique is subjective (Clapcott et al., 2011) it provides a measure of fine sediment surface drape, which is likely to directly influence macroinvertebrate com-munities and as such is likely to be a biologically relevant metric (Conroy et al., 2016), relating to niche theory and habitat suitability (Hirzel and Le Lay 2008) over the entire reach.

Taxonomic resolution
In contrast to previous findings (Relyea et al., 2012;Turley et al., 2014) the taxonomic resolution had little effect on the indices correlations with fine sediment; E-PSI fam69 and E-PSI mixed had similar strong correlations with fine sediment. Relyea et al. (2012) found that sensitivities varied within invertebrate families and concluded that family level was insufficient for pressure-specific tools. This within-family variation of sensitivities to fine sediment was expected to result in large differences in terms of the correlation between family and mixed level indices and fine sediment. Despite the similar correlations observed, it is likely that the mixed level tool will provide more diagnostic information, particularly at more impacted sites. The taxonomic sufficiency is a key consideration when developing an invertebrate-based biomonitoring tool, as the taxonomic resolution can affect the accuracy of predictions (Schmidt-Kloiber and Nijboer, 2004) as well as the costs associated with processing biotic samples (Jones, 2008;Marshall et al., 2006). The higher the taxonomic resolution, the more resource intensive biomonitoring becomes (identification requires more time and expertise/training). For macroinvertebrates, some studies have shown that family level is often sufficiently able to detect environmental change (Marshall et al., 2006;Mueller et al., 2013), whilst others conclude that a higher resolution is preferential, particularly for genera-rich families or when attempting to identify subtle environmental changes (Monk et al., 2012;Waite et al., 2004). Furthermore, although our knowledge of species level biology is incomplete, intra-generic environmental requirements have been shown to vary significantly for various families of macroinvertebrates (reviewed in Lenat and Resh, 2001). In terms of a pressure-specific biomonitoring tool, the optimum taxonomic resolution in terms of model predictions is likely to be dependent on the scale or gradient of environmental pressure, but will inevitably be a compromise between the costs associated with high taxonomic resolution and the ecological information lost at lower resolutions. Nevertheless, the development of new pressure-specific tools should ideally align with national/international monitoring agencies, to ensure that they can be applied to existing protocols and historical data.
In addition, as the E-PSI index is not a purely statistically based tool, it is able to include relatively rare taxa (low abundances or small distribution range), which are often removed in the development stages of statistically based indices (Murphy et al., 2015;Relyea et al., 2012). These rare taxa, specifically those with small distribution ranges, may be important indicators of subtle environmental change as they are likely to be more specialised, having a narrower range of optimal conditions, and as such may be the first taxa to indicate the effects of a pressure (Cao et al., 1998;Schmidt-Kloiber and Nijboer, 2004). Whilst these assumptions are intuitive, there is some debate as to the suitability of rare taxa in multivariate analyses particularly relating to their redundancy and their influence on observed versus expected scores (Cao et al., 2001;Marchant 2002;Van Sickle et al., 2007). Furthermore, when conservation issues become a concern for these rare species, their collection should be carefully considered. The E-PSI index only uses taxa that are collected in routine samples by the monitoring agencies, and as such makes the most use of existing data. The River Invertebrate Classification Tool (RICT) used in the UK to derive expected index scores for sites, includes rare species, calculating scores based on their probability of occurrence (Clarke et al., 2003).

Application of the E-PSI index
In application, most indices should be used within an Ecological Quality Index (EQI) framework, based on observed versus expected invertebrate community composition (Glendell et al., 2014a), particularly when being used for regulatory purposes. This requires a predictive model that is independent of the pressure of concern. At present, in the UK, the predictive model "RICT" (previously RIVPACS IV) is not independent of fine sediment, as substrate composition is a predictor variable (Clarke et al., 2011). Furthermore, it is over-simplistic to assume that deviation of observed from expected community composition (and therefore index score) can be explained by a single measurement at one point in time. Aside from the fact that the structure of macroinvertebrate communities is determined not simply by the contemporaneous fine sediment, but by local sediment dynamics spanning the preceding days, weeks and months, there are multiple reasons for this not being a simple cause-effect relationship. Firstly, fine sediment has numerous direct and indirect effects on invertebrates (reviewed in Bilotta and Brazier, 2008), many of which are poorly understood. Secondly, not only are there often dissociated pressures present, but studies have also shown a number of common sediment-associated physical, chemical and biological pressures ( Table 1) that depend on catchment land use, in-stream influences, geomorphology, hydrology and so on. For example, fine sediment pressure is commonly associated with stream flow, due to the influence of stream velocity on the transport, deposition and re-suspension of fine sediment (Wood and Armitage, 1997). Additionally, depending on the sorptive properties of the sediment (i.e. organic carbon content, particle size, clay content and cation exchange capacity) hydrophobic contaminants often become associated via adsorption or absorption, potentially acting as confounding pressures (Warren et al., 2003). Lastly, in disturbed sites, the biological communities have often been subjected to this range of pressures over an extended period of time and therefore the observed biological community composition at any point in time is partly reflective of the legacy of these pressures (Allan, 2004).
The common occurrence of diverse pressures in rivers and streams, places a greater importance on a mechanistic linkage and understanding of biotic response to the pressure of concern. As a result, it is likely that most pressure-specific indices will be limited in terms of their ability to make accurate predictions of the degree of impact across all river and stream ecosystems. Instead, a more achievable and realistic goal is to use a suite of different pressurespecific indices to identify potential areas of ecological degradation and the likely causes of any deviations. The family and mixed level E-PSI indices are two such tools, which could be used alongside other indices to inform more targeted monitoring and mitigation measures.
ter Biological Association, Scotland and Northern Ireland Forum for Environmental Research, Scottish Environment Protection Agency, Scottish Executive, Scottish Natural Heritage, South West Water, Welsh Assembly Government. The authors would also like to thank the anonymous reviewers and the journal editor for comments that helped to improve the manuscript.
Appendix A.