River Invertebrate Prediction And Classification System Research Articles

Multitaxon distribution models reveal severe alteration in the regional biodiversity of freshwater invertebrates

Knowledge of current and historical conditions is needed to guide conservation and restoration policies, but such information is lacking for most taxa. This problem is especially severe for small, inconspicuous taxa, such as the thousands of invertebrate species that inhabit stream and other freshwater ecosystems. We describe a novel application of River Invertebrate Prediction and Classification System (RIVPACS)-type multitaxon distribution models that, when applied to the biological survey data routinely collected in support of water-quality monitoring programs, can quantify the regional biodiversity status of hundreds of taxa. We used models developed for 2 stream data sets (North Carolina and the Mid-Atlantic Highlands, USA) to illustrate the potential of this approach. The models were calibrated with data collected from sites considered to be in reference condition and predict how probabilities of capture for each taxon in a data set vary across natural environmental gradients. When applied to survey data from multiple sites, predicted probabilities of capture can be summed across sites to estimate the taxon-specific frequencies of collection (Fe) expected under reference conditions. Comparison of observed frequencies (Fo) with Fe provides a quantitative measure of how individual taxon frequencies of collection have shifted relative to estimated reference conditions. In these 2 data sets, Fe was statistically different from Fo for >70% of taxa, implying wholesale changes have occurred in the stream invertebrate biodiversity of both regions. Fo was <Fe for most taxa, including those predicted to have been the historically most common taxa, indicating significant biodiversity loss. This type of multitaxon, regional-scale assessment of biodiversity status complements the site-specific, community-level assessments typically used by water resource managers to assess the biological integrity of individual water bodies. Together, the 2 approaches can provide a more robust, multiscale understanding of the effects landscape and waterway alteration have had on the diversity of freshwater biota.

Role of periphyton in ecological assessment of lakes

Assessment of ecological integrity is the basis for sustainable management of the ecosystem services lakes provide. Periphyton is used in stream assessment, but lake assessment is based mostly on water-column variables. We addressed the use of periphyton to assess lakes, how factors influence responses of periphyton metrics, and whether periphyton provides unique information for determining biological condition and identifying risks. Much effort directed at using periphyton for assessment has focused on linear relationships between taxonomic metrics and single stressors. These relationships can be good proxies for water chemistry, but do not always represent biological conditions. Community diversity is difficult to relate to stressors because low values may result from natural conditions (grazing, disturbance). Quantile regression, River Invertebrate Prediction and Classification System (RIVPACS)-type models, and taxonomic distinctiveness may provide more precise indicators. Periphyton biomass and productivity are linked closely to trophic status but are rarely used in assessment. Responses of these variables over a trophic gradient are often nonlinear, and surrogate metrics that use quantile regression or weighted-averaging calibration/regression based on community composition may overcome this problem for assessment. Measures, such as alkaline phosphatase activity, number of N-fixing taxa, and periphyton nutrient stoichiometry can be used to assess littoral-zone ecosystem function and to help establish causative effects of stressors. The value added by periphyton-based assessment is that it provides higher-level understanding of the ecological status of the littoral zone. Assessment must be rooted in a strong understanding of the science and must provide guidance and cost-effective options for lake managers. We outline a hierarchy for periphyton assessment methods that assess risk at different levels of effort and precision.

COMBINING MULTIPLE MACHINE LEARNING ALGORITHMS TO PREDICT TAXA UNDER REFERENCE CONDITIONS FOR STREAMS BIOASSESSMENT

ABSTRACTIn the present study, we tested the potential of combining three machine learning techniques in a bioassessment tool to more accurately predict the pool of expected taxa at a site. This tool, the Hydra, uses the best performing technique from Support Vector Machines (SVM), Multi‐layer Perceptron and K‐Nearest Neighbour (KNN), to predict the taxa expected at a stream site, and further evaluates the quality of a site, though a classification system based on observed/expected values, similar to that used in River Invertebrate Prediction and Classification System (RIVPACS) models. To test the procedure, we used a dataset composed of 137 training sites, 15 validation sites and 174 test sites (potentially disturbed) from Portuguese streams. The combined use of three machine learning techniques was more effective in the prediction of invertebrate taxa at a site than their individual use. The three methods were always tested for all invertebrate taxa, but from the three techniques tested, SVM and KNN were most often the best performing techniques (the most accurate among the three for a higher number of taxa) in the prediction of invertebrate taxa with the present dataset. The combination of all algorithms implemented in Hydra resulted in good models for stream bioassessment (e.g. SD OE50 &lt; 0.2, regression of O vs E: R2 &gt; 0.6, Spearman correlations with global degradation &gt;0.7). We also found no advantage in removing rare taxa from the training dataset, and 50% accuracy is the most adequate accuracy level for calculation of OE ratios through Hydra. Future work should consist of comparing the performance of this technique with others, such as the RIVPACS models, using the same datasets. Considering the flexibility of this technique, self‐adjustment and easy implementation through a website (aquaweb.uc.pt), we expect it to be also useful in the prediction of other aquatic elements such as fishes and algae. Copyright © 2013 John Wiley &amp; Sons, Ltd.

The use of invertebrate body burdens to predict ecological effects of metal mixtures in mining-impacted waters

The present study investigated whether invertebrate body burdens can be used to predict metal-induced effects on aquatic invertebrate communities. Total dissolved metal levels and four invertebrate taxa (Leuctra sp., Simuliidae, Rhithrogena sp. and Perlodidae) were sampled in 36 headwater streams located in the north-west part of England. Using the River Invertebrate Prediction and Classification System (RIVPACS) taxonomic completeness of invertebrate communities was assessed. Quantile regression was used to relate invertebrate body burdens to a maximum (90th quantile) ecological response, both for all metals separately and in mixtures. Significant relations between Cu, Zn and Pb burdens in Leuctra sp. (Zn, Pb), Simuliidae (Zn, Pb), Rhithrogena sp. (Cu, Zn, Cu+Zn) and Perlodidae (Zn) and both taxonomic completeness (O/E taxa) and Biological Monitoring Working Party index scores (O/E BMWP) were observed. Correspondingly the obtained Cu–Zn mixture model an acceptable impact of 5% change in taxonomic completeness is expected at Rhithrogena sp. body burdens of 1.9μmolg−1 Cu (121μgg−1 Cu) in case of low Zn bioavailability (Rhithrogena sp. Zn body burden of 2.9μmolg−1 or 190μgg−1), which will drop to 0.30μmolg−1 Cu (19.1μgg−1 Cu) in case of higher Zn bioavailability (Zn body burden of 72.6μmolg−1 or 4747μgg−1). For Zn, 5% change in taxonomic completeness is expected at Rhithrogena sp. body burdens of 76.4μmolg−1 Zn (4995μgg−1 Zn) in case of low Cu bioavailability (Cu body burden of 0.19μmolg−1 or 12.1μgg−1), which will drop to 6.6μmolg−1 Zn (432μgg−1 Zn) at higher Cu bioavailability (Cu body burden of 1.74μmolg−1 or 111μgg−1). Overall, the present study concludes that invertebrate body burdens can be used to (1) predict metal-induced ecological effects and (2) to derive critical burdens for the protection of aquatic invertebrate communities.

Accounting for both local aquatic community composition and bioavailability in setting site-specific quality standards for zinc

Recent years have seen considerable improvement in water quality standards (QS) for metals by taking account of the effect of local water chemistry conditions on their bioavailability. We describe preliminary efforts to further refine water quality standards, by taking account of the composition of the local ecological community (the ultimate protection objective) in addition to bioavailability. Relevance of QS to the local ecological community is critical as it is important to minimise instances where quality classification using QS does not reconcile with a quality classification based on an assessment of the composition of the local ecology (e.g. using benthic macroinvertebrate quality assessment metrics such as River InVertebrate Prediction and Classification System (RIVPACS)), particularly where ecology is assessed to be at good or better status, whilst chemical quality is determined to be failing relevant standards. The alternative approach outlined here describes a method to derive a site-specific species sensitivity distribution (SSD) based on the ecological community which is expected to be present at the site in the absence of anthropogenic pressures (reference conditions). The method combines a conventional laboratory ecotoxicity dataset normalised for bioavailability with field measurements of the response of benthic macroinvertebrate abundance to chemical exposure. Site-specific QSref are then derived from the 5%ile of this SSD. Using this method, site QSref have been derived for zinc in an area impacted by historic mining activities. Application of QSref can result in greater agreement between chemical and ecological metrics of environmental quality compared with the use of either conventional (QScon) or bioavailability-based QS (QSbio). In addition to zinc, the approach is likely to be applicable to other metals and possibly other types of chemical stressors (e.g. pesticides). However, the methodology for deriving site-specific targets requires additional development and validation before they can be robustly applied during surface water classification.

THE ASSESSMENT OF FINE SEDIMENT ACCUMULATION IN RIVERS USING MACRO‐INVERTEBRATE COMMUNITY RESPONSE

ABSTRACTIncreased fine sediment deposition and entrainment in rivers can arise from a combination of factors including low flows, habitat modification and excessive sediment delivery from the catchment. Physical and visual methods have traditionally been used to quantify the volume of deposited fine sediment (&lt;2 mm in size), but here we propose an alternative, the development and utilization of a sediment‐sensitive macro‐invertebrate metric (PSI — Proportion of Sediment‐sensitive Invertebrates) which provides a proxy to describe the extent to which the surface of river beds are composed of, or covered by, fine sediments. Where suitable biomonitoring data exists, the index can be calculated retrospectively to track trends in fine sediment deposition, and its ecological impact, through time. Furthermore, the utilization of reference condition models such as RIVPACS (River InVertebrate Prediction And Classification System), allows site‐specific unimpacted conditions to be defined, opening‐up the possibility of standard classification and assessment systems being developed. In Europe, such systems are vital if the Water Framework Directive is to be implemented. Knowledge regarding spatial differences in sediment/flow interdependencies may provide valuable information on diffuse sources of fine sediment to rivers and we illustrate this with an example from the UK (Laceby Beck). Further UK case studies are presented to show a range of applications, including the demonstration of improvements in habitat heterogeneity following river restoration (rivers Chess and Rib) and the detection of fine sediment impacts downstream of an impoundment (Eye Brook). The PSI metric offers a readily deployable, cost‐effective and hydroecologically relevant methodology for the assessment of fine sediment impacts in rivers. The technique has potential for application outside of the UK and an adaptation of the methodology for use in the Simandou Mountains (Guinea) is used to illustrate this. Copyright © 2011 John Wiley &amp; Sons, Ltd.

Evaluation of approaches for measuring taxonomic completeness of lake profundal macroinvertebrate assemblages

Summary 1. Benthic macroinvertebrates (MI) are commonly used to assess freshwater ecosystems with the reference condition approach. Such assessments necessitate control for natural community variation, either by categorical typologies or by predictive models that have been widely and successfully developed for running water biota but not previously for lake profundal invertebrates. 2. We evaluated four modelling techniques [multivariate regression tree (MRT), limiting environmental differences, nonparametric multiplicative regression (NPMR) and River Invertebrate Prediction And Classification System (RIVPACS) and the operative Finnish lake typology for assessing taxonomic completeness (observed-to-expected number of taxa, O/E) of profundal MI assemblages. We used data from 74 and 33 minimally disturbed reference lake basins for calibration and validation of the approaches, respectively, and 72 test basins subject to various anthropogenic pressures to evaluate sensitivity to detect impact. Either all predicted taxa (threshold probability of capture Pt = 0+) or only those predicted to be captured with ≥0.25 probability were used to calculate O/E. 3. With Pt = 0.25, all four modelling approaches were accurate (mean O/E = 0.966–1.053) but imprecise (SD of O/E = 0.279–0.304) in predicting the fauna actually observed in validation sites. All models were subtly more precise than a null model (mean 1.038, SD 0.343) or the typology (1.046, 0.327). The taxon-specific NPMR model was slightly more precise than the other three models based on site groupings. 4. The O/E values correlated relatively weakly (r = 0.55–0.86) among the approaches, which thus produced contrasting lake-specific assessments, despite their seemingly comparable performances. Indeed, typology, suggesting that MI assemblages were impaired in 56% of test sites, was more sensitive than the other approaches (26–46%) as an indicator of human-induced deterioration. However, this greater ostensible sensitivity seemed to be biased, as lake morphometry, a main driver of natural community variation, remained uncontrolled by the typology. 5. Generally, our exercise illustrates the inconclusiveness of the common validation criteria for the assessment methods. The apparent poor predictability of the profundal fauna, irrespective of the method, may partly stem from large observation error, which could be alleviated by more intensive sampling. However, instead of an O/E-taxa index, some other metric encompassing quantitative aspects might be preferable for assessing these species-poor communities.

Influence of geographical extent on typology‐ and model‐based assessments of taxonomic completeness of river macroinvertebrates

Summary1. River InVertebrate Prediction and Classification System (RIVPACS)‐type predictive models are increasingly used to assess the biological condition of freshwaters, but management schemes may also be based on a priori groupings of similar water bodies (typologies) to control for natural variation in biota. The two approaches may lead to disagreements of the biological status of a site, depending on, for example, the spatial scale at which assessments are conducted.2. We used data from 96 reference and 134 potentially impacted sites from Western and Central Finland to compare RIVPACS‐type models and a simple size‐based typology of rivers for the assessment of taxonomic completeness (the quotient of the Observed‐to‐Expected number of predicted taxa, O / E) of riffle macroinvertebrates. We specifically examined how geographical extent influences bioassessment performance (accuracy, precision and sensitivity to detect impact) of the two approaches. To fully examine the behaviour of the O / E‐index with the two approaches at differing spatial scales, we performed all assessments with a full range of thresholds for predicted taxa occurrence probabilities (pt from 0+ to 0.9).3. Both approaches performed consistently better than the corresponding null models. At the larger extent (i.e. assessment encompassing the whole study area), the RIVPACS‐approach performed in all aspects better than the typology‐approach. However, at the smaller extent (i.e. regional assessments) the RIVPACS‐type models and the typologies showed similar accuracy to predict the actual fauna (mean E), similar precision (SD) of cross‐validated O / E and similar sensitivity to detect sites with human impairment.4. SD(O / E) decreased (i.e. precision increased) consistently with increasing pt. However, both approaches were most sensitive at intermediate pt:s (c. 0.2–0.6) when taxa with low predicted occurrence probabilities were excluded.5. Our results show that RIVPACS‐type predictive models are less susceptible to variations in spatial scale, whereas the performance of a priori typologies increases with decreasing spatial extent. Thus, RIVPACS‐type models are more efficient for large‐scale bioassessments, but at restricted spatial scales, or with an otherwise biologically meaningful stratification, simple a priori classifications can be equally useful for the assessment of taxonomic completeness of river macroinvertebrates.

Calibration and evaluation of five indicators of benthic community condition in two California bay and estuary habitats

Many types of indices have been developed to assess benthic invertebrate community condition, but there have been few studies evaluating the relative performance of different index approaches. Here we calibrate and compare the performance of five indices: the Benthic Response Index (BRI), Benthic Quality Index (BQI), Relative Benthic Index (RBI), River Invertebrate Prediction and Classification System (RIVPACS), and the Index of Biotic Integrity (IBI). We also examine whether index performance improves when the different indices, which rely on measurement of different properties, are used in combination. The five indices were calibrated for two geographies using 238 samples from southern California marine bays and 125 samples from polyhaline San Francisco Bay. Index performance was evaluated by comparing index assessments of 35 sites to the best professional judgment of nine benthic experts. None of the individual indices performed as well as the average expert in ranking sample condition or evaluating whether benthic assemblages exhibited evidence of disturbance. However, several index combinations outperformed the average expert. When results from both habitats were combined, two four-index combinations and a three-index combination performed best. However, performance differences among several combinations were small enough that factors such as logistics can also become a consideration in index selection.

Land use and the structure of western US stream invertebrate assemblages: predictive models and ecological traits

AbstractInferences drawn from regional bioassessments could be strengthened by integrating data from different monitoring programs. We combined data from the US Geological Survey National Water-Quality Assessment (NAWQA) program and the US Environmental Protection Agency Wadeable Streams Assessment (WSA) to expand the scope of an existing River InVertebrate Prediction and Classification System (RIVPACS)–type predictive model and to assess the biological condition of streams across the western US in a variety of landuse classes. We used model-derived estimates of taxon-specific probabilities of capture and observed taxon occurrences to identify taxa that were absent from sites where they were predicted to occur (decreasers) and taxa that were present at sites where they were not predicted to occur (increasers). Integration of 87 NAWQA reference sites increased the scope of the existing WSA predictive model to include larger streams and later season sampling. Biological condition at 336 NAWQA test sites was...

Effects of regionalization decisions on an O/E index for the US national assessment

AbstractWe examined the effects of different regionalization schemes on the performance of River InVertebrate Prediction and Classification System (RIVPACS)-type predictive models in assessing the biological conditions of streams of the US for the National Wadeable Streams Assessment (WSA). Three regionalization schemes were considered: a single national predictive model (MOD1), separate predictive models for each of the 9 WSA aggregated Omernik level III ecoregions (MOD9), and 3 predictive models roughly corresponding to the western US, the Appalachian Mountains, and the Central and Coastal Plains (MOD3). The goal of the WSA was to assess stream condition at the national scale and at the scale of WSA aggregated ecoregions, so we compared the performance of the ratio of the observed number of taxa to the expected number of taxa (O/E) index estimated using different regionalization schemes at both of these spatial scales. We assessed model performance with a randomized resampling procedure, in which we set...

Predicting the stream macroinvertebrate fauna across regional scales: influence of geographical extent on model performance

AbstractPredictive modeling is used increasingly often in freshwater bioassessment. These models are routinely applied across broad geographical areas, yet the influence of spatial extent on model performance has not been assessed directly. We examined the influence of changing geographical extent on the performance of River InVertebrate Prediction and Classification System (RIVPACS)–type predictive models of benthic macroinvertebrates in boreal headwater streams representative of reference conditions. Separate models were constructed for 2 hierarchical extents: within ecoregions (middle boreal and north boreal) and across ecoregions. We assessed the accuracy and precision of the models using root mean squared error (RMSE) of the observed/expected (O/E) ratio of the number of predicted taxa. We compared the RMSE of the within-ecoregion models to that of the across-ecoregions models. We tested the models using 2 thresholds for probability of species occurrence (Pt): Pt = 0+ (all predicted taxa) and Pt = 0....

An index of compositional dissimilarity between observed and expected assemblages

The reference-condition approach to bioassessment often uses the observed/expected (O/E) ratio to indicate anthropogenic alteration of aquatic macroinvertebrates, fish, or periphyton assemblages. Given a list of taxa found at ≥1 minimally disturbed reference sites, E is the number of those taxa that would expected in a sampled assemblage if the sampled stream were in reference condition, and O is the number of those taxa observed in the sample. An O/E value significantly <1.0 indicates that a stream has lost taxa relative to its reference-condition expectation, possibly because of anthropogenic stress. However, the O/E index can be insensitive to stress-induced shifts in taxonomic composition that leave assemblage richness unchanged. As an alternative to O/E, I propose using BC, an adaptation of Bray–Curtis distance, to measure the compositional dissimilarity between an observed and expected assemblage directly. I compared the performance of BC and O/E at 5685 stream and lake sites throughout the contiguous 48 states of the US using 1 of 10 River Invertebrate Prediction and Classification system (RIVPACS)-type models to predict expected assemblages. Percentages of independently determined nonreference sites that were declared to be nonreference by BC exceeded the percentage declared to be nonreference by O/E by an average of 6 to 16 percentage points, depending on whether the 2 indices included low-probability taxa, whether a 1-sided or 2-sided O/E criterion was used to declare nonreference, and whether predictive or null models were used to predict expected assemblages. Correlations between BC scores and anthropogenic stressor variables were stronger than correlations between O/E scores and anthropogenic stressor variables in 18 of 25 cases. In contrast to O/E, BC can include low-probability taxa without reducing its power to detect nonreference conditions.

Long‐term changes in macroinvertebrate communities of a heavy metal polluted stream: the river Nent (Cumbria, UK) after 28 years

AbstractThe macroinvertebrate fauna of the river Nent, a heavy metal polluted upland stream in northern England, first surveyed in 1976, was re‐examined in 2004 to assess long‐term changes in relation to activities in the catchment which have included, a river restoration project, works associated with the development of a mine heritage site and the removal of a chicken farm and more effective organic pollution control measures. The degree of change between years in total abundance, numbers of taxa and community composition varied considerably between sites. Two tributary sites showed marked changes between years due to a reduction in acidity in one and changed substratum in the other but in the main river no significant difference in total taxa and total abundance was observed between years, although seasonal differences were significant. Multivariate community analyses grouped 1976 samples with those taken in 2004 but revealed a clear separation between upper and lower Nent sites. This observed difference may be related to increased algal cover in the lower sites in conjunction with geomorphological features of the lower Nent. Environmental assessment methodology River Invertebrate Prediction and Classification System (RIVPACS) identified ‘sensitive’ faunal groups and indicated that the extensive impact of zinc pollution in the main river has remained practically the same between 1976 and 2004. This persistence of community structure despite the heavily disturbed nature of the river is attributed to relative constancy in instream habitat conditions. The recent activities in the catchment have had little effect on the overall controllers of faunal communities in this system, zinc concentrations and geomorphological characteristics of the stream. However, further disturbances involving movement of spoil heaps and channel alterations in the upper catchment may increase both sediment and heavy metal loadings to the river. Copyright © 2007 John Wiley &amp; Sons, Ltd.

Modeling natural environmental gradients improves the accuracy and precision of diatom-based indicators

AbstractDiatom-based indicators can contribute significantly to comprehensive assessments of stream biological conditions. We used modeling to develop, evaluate, and compare 2 types of diatom-based indicators for Idaho streams: an observed/expected (O/E) ratio of taxon loss derived from a model similar to the River InVertebrate Prediction And Classification System (RIVPACS) and a multimetric index (MMI). Modeling the effects of natural environmental gradients on assemblage composition is a key component of RIVPACS, but modeling has seldom been used for MMI development. Diatom assemblage structure varied substantially among reference-site samples, but neither ecoregion nor bioregion accounted for a significant portion of that variation. Therefore, we used Classification and Regression Trees (CART) to model the variation of individual metrics with natural gradients. For both CART and RIVPACS modeling, we restricted predictors to natural variables unaffected by or resistant to human disturbances. On average,...

Comparing apples with apples: use of limiting environmental differences to match reference and stressor‐exposure sites for bioassessment of streams

AbstractHuman impacts on freshwater biota are often assessed by comparisons with regional reference sites judged to be free or nearly free from the influence of stressors of concern. These comparisons may involve predictive models that extrapolate from reference‐site data to estimate the biota that would occur at each assessment site in the absence of the stressors. This extrapolation often involves selection or weighting of data from particular reference sites, but it is seldom demonstrated whether this process results in a more accurate, precise and sensitive assessment than would be achieved from comparison with a simple unweighted combination of data from all reference sites, that is a null model. In addition, predictive models often rely on additive combinations of environmental predictor variables, but such combinations may poorly represent the natural control of spatial variation in biological communities by multiple limiting factors. In this paper, we describe a different type of reference‐site approach, based on the concept of limiting environmental differences (LEDs) as natural constraints on biological similarity among sites. In this method, an assessment site is compared with a sub‐set of environmentally matched reference sites selected by application of LED criteria to individual environmental variables. We illustrate this approach by its application to a potentially subtle impact: the effect of water abstraction on fish assemblages in unregulated streams in northeastern New South Wales, Australia. We compared (1) a LED‐based predictive model, (2) a null model and (3) a model based on cluster analysis and multiple discriminant analysis in the style of the widely followed River Invertebrate Prediction and Classification System (RIVPACS). The LED‐based model was about as accurate as the RIVPACS‐type model and the most precise and sensitive of the three, being best able to distinguish sites where fish assemblages departed from reference status. Copyright © 2007 John Wiley &amp; Sons, Ltd.

Exclusion of rare taxa affects performance of theO/Eindex in bioassessments

The contribution of rare taxa to bioassessments based on multispecies assemblages is the subject of continued debate. As a result, users of predictive models such as River InVertebrate Prediction and Classification System (RIVPACS) disagree on whether to exclude locally rare taxa from the O/E index, where O is the number of taxa observed in a sampled assemblage, and E is the number that would be expected if the site were in a minimally disturbed reference condition. We assessed how the bioassessment performance of O/E was affected by the exclusion of taxa with site-specific, model-predicted occurrence probabilities that did not exceed thresholds of PT = 0+, 0.1, 0.2, … , 0.7. We assessed O/E performance for each of 10 predictive models applied to a total of 5685 stream and lake samples from throughout the contiguous USA. For 5 of the 10 cases, the standard deviation (SD) of O/E across reference sites was reduced by at least 0.02 O/E units when locally rare and uncommon taxa were excluded (PT = 0.5) from O/E, as compared with all taxa being included (PT = 0+). These reductions in SD denote increases in precision of the O/E index. We also assessed the sensitivity of O/E, measured by the % of test sites (that is, sites independently assessed as not being in reference condition) that were declared to be outside the reference distribution of O/E scores. Five of our 10 cases showed increases in sensitivity of ≥10% at PT = 0.5 as compared with PT = 0+. All cases that did not show increases in sensitivity or precision also showed no decrease in either of these performance measures as PT increased. A comparison of observed occurrence frequencies of taxa at reference and test sites qualitatively explained the size of the O/E sensitivity response in all 10 cases. This result suggests that effects of rare-taxa exclusion are a direct consequence of these occurrence frequencies rather than of predictive-model structures. Thus, we predict that other assemblage-based bioassessment tools are likely to show improved sensitivity when rare taxa are excluded.

Effects of locally rare taxa on the precision and sensitivity of RIVPACS bioassessment of freshwaters

Summary1. An overall aim in freshwater bioassessment is to use biological methods, metrics and forms of indices which are precise, in that they give repeatable results between replicate samples, but which are also sensitive to changes in environmental impacts and stresses. Here we studied the effects of excluding taxa with site‐specific River Invertebrate Prediction and Classification System (RIVPACS)‐type model expected probabilities less than (or equal to) a threshold Pt (0.0, 0.1, 0.2,…,0.9) on the value, precision and power to detect biological effects of environmental stress using the observed to expected ratios (O/E) of biotic indices used to assess the ecological status of U.K. river sites.2. Amongst the 614 high quality GB RIVPACS reference sites, excluding taxa with low expected probabilities of occurrence gave less total variation (i.e. lower SD) in the estimates O/E for number of taxa (O/ETAXA) and the average score per taxon (O/EASPT).3. A separate analysis of a replicated sampling study of sites from a wide range of physical types and qualities revealed that sampling variances in O/E for reference condition sites decreased as more locally rare taxa were excluded (but only up to Pt = 0.5 for O/EASPT). However, for moderately impacted and poor quality sites, estimates of both O/ETAXA and O/EASPT based on all (Pt = 0.0) or most taxa (i.e. Pt ≤ 0.3) had lower sampling variances and were more precise.4. Within a very large independent set of test sites with a wide range of perceived levels of environmental stress, increasing the threshold Pt led to systematic compression of the realised O/E scale towards unity. Specifically, with increasing threshold, O/E values &gt;1 are on average reduced, while O/E values &lt;1 have a tendency to be higher and closer to unity (with the exception of O/EASPT for the most severely stressed sites).5. Accuracy and statistical power to detect environmental stress (measured by the percentage of stressed sites with O/E below the lower 10‐percentile value for reference sites) was very similar using O/ETAXA for Pt up to 0.7. Using O/EASPT, power to detect overall general stress decreased slowly as Pt was increased; the rate of fall in power was slightly faster when restricted to sites subject to moderate or severe stress from organic inputs.6. Taxa which are more sensitive to (organic) stresses [i.e. have high Biological Monitoring Working Party (BMWP) scores] tend to be naturally less widespread (i.e. amongst reference sites) and thus were found to have considerably lower average site‐specific expected probabilities; this may explain why the use of higher thresholds Pt can exclude more such sensitive taxa and lead to underestimation of the extent of impacts.7. The standard U.K. RIVPACS sampling and sample processing procedures aim to identify all taxa within a sample. This may lead to a longer distribution tail of rarer (low probability) taxa than sampling methods based on a fixed count subsample and influence the practical effects of excluding rare taxa with low expected probabilities from bioassessments.

PREDICTIVE MODELS ATTRIBUTE EFFECTS ON FISH ASSEMBLAGES TO TOXICITY AND HABITAT ALTERATION

Biological assessments should both estimate the condition of a biological resource (magnitude of alteration) and provide environmental managers with a diagnosis of the potential causes of impairment. Although methods of quantifying condition are well developed, identifying and proportionately attributing impairment to probable causes remain problematic. Furthermore, analyses of both condition and cause have often been difficult to communicate. We developed an approach that (1) links fish, habitat, and chemistry data collected from hundreds of sites in Ohio (USA) streams, (2) assesses the biological condition at each site, (3) attributes impairment to multiple probable causes, and (4) provides the results of the analyses in simple-to-interpret pie charts. The data set was managed using a geographic information system. Biological condition was assessed using a RIVPACS (river invertebrate prediction and classification system)-like predictive model. The model provided probabilities of capture for 117 fish species based on the geographic location of sites and local habitat descriptors. Impaired biological condition was defined as the proportion of those native species predicted to occur at a site that were observed. The potential toxic effects of exposure to mixtures of contaminants were estimated using species sensitivity distributions and mixture toxicity principles. Generalized linear regression models described species abundance as a function of habitat characteristics. Statistically linking biological condition, habitat characteristics including mixture risks, and species abundance allowed us to evaluate the losses of species with environmental conditions. Results were mapped as simple effect and probable-cause pie charts (EPC pie diagrams), with pie sizes corresponding to magnitude of local impairment, and slice sizes to the relative probable contributions of different stressors. The types of models we used have been successfully applied in ecology and ecotoxicology, but they have not previously been used in concert to quantify impairment and its likely causes. Although data limitations constrained our ability to examine complex interactions between stressors and species, the direct relationships we detected likely represent conservative estimates of stressor contributions to local impairment. Future refinements of the general approach and specific methods described here should yield even more promising results.

Comparison of the performance of different bioassessment methods: similar evaluations of biotic integrity from separate programs and procedures

Regional bioassessment programs of states, various federal agencies, and other governmental and private groups often use different methods to collect and analyze stream invertebrate samples. This lack of uniformity has created concern and confusion over the comparability of disparate sources of data, but few studies have attempted to evaluate differences in performance between methods or to reconcile the results produced from different methods. We conducted concurrent sampling at 40 sites in the eastern Sierra Nevada of California using 3 bioassessment methods to obtain directly comparable data sets. The riffle-based methods (University of California Sierra Nevada Aquatic Research Laboratory [UC-SNARL, Lahontan Water Board], California Stream Bioassessment Protocol, and US Forest Service Region 5) differed at each stage from field sample collection to laboratory processing and data analysis. We used a performance-based methods system to compare precision, uniformity, discrimination, accuracy, and correlations among multimetric Index of Biotic Integrity (IBI) scores and multivariate River Invertebrate Prediction and Classification System (RIVPACS)-type observed/expected (O/E) ratios. Reference and test sites were identified using local and upstream-watershed disturbance criteria, and invertebrate community measures and models were then developed to discriminate between reference and test sites. The more-intensive UC-SNARL method showed slightly, but consistently, greater sensitivity for discriminating impairment than the other 2 methods. The UC-SNARL method produced greater differences between reference- and test-site means relative to lower reference-site standard deviations than the other 2 methods. However, assessment scores were highly correlated among methods and distinguished reference from test sites with similar accuracy among methods despite the slight differences in performance. Our results show that differing bioassessment methods can yield very similar, effective discrimination of impaired biological condition even though they have multiple differences in field and laboratory protocols (mesh size, replication, area sampled, taxonomic resolution, total counts). Moreover, this conclusion did not depend on the approach taken to data analysis because both multimetric IBIs and multivariate RIVPACS-type O/Es were in close agreement. Methodological uniformity is important when coordinating monitoring programs, but our results suggest that data from multiple sources could potentially be used interchangeably and for cross-validation of assessments of stream biological integrity.