Development of a RIVPACS-type predictive model for bioassessment of wadeable streams in Wyoming

RIVPACS models produce a community-level measure of biological condition known as O/E, which is derived from a comparison of the observed (O) biota with those expected (E) to occur in the absence of anthropogenic stress. We used benthic macroinvertebrate and environmental data collected at 925 stream monitoring stations, from 1993 to 2001, to develop, validate, and apply a RIVPACS model to assess the biological condition of wadeable streams in Wyoming. From this dataset, 296 samples were identified as reference, 157 of which were used to calibrate the model, 46 to validate it, and 93 to examine temporal variability in reference site O/E-values. We used cluster analyses to group the model development reference sites into biologically similar classes of streams and multiple discriminant function analysis to determine which environmental variables best discriminated among reference groups. A suite of 14 categorical and continuous environmental variables best discriminated among 15 reference groups and explained a large proportion of the natural variability in biota within the reference dataset. Eleven of the predictor variables were derived from GIS. As expected, mean O/E-values for reference sites used in model development and validation were near unity and statistically similar. Temporal variability in O/E-values for reference sites was low. Test site values ranged from 0 to 1.45 (mean = 0.73). The model was accurate in both space and time and precise enough (S.D. of O/E-values for calibration data = 0.17) to detect modest alteration in biota associated with anthropogenic stressors. Our model was comparable in performance to other RIVPACS models developed in the United States and can produce effective assessments of biological condition over a broad, ecologically diverse region. We also provide convincing evidence that RIVPACS models can be developed primarily with GIS-based predictor variables. This framework not only simplifies the extraction of predictor variable information while potentially reducing expenditures of time and money in the collection of predictor variable information, but opens the door for development and/or application of RIVPACS models in regions where there is a paucity of local-scale, abiotic information.     

Implications of community concordance for assessing stream integrity at three nested spatial scales in Minnesota,U.S.A.

1. Fish and invertebrate assemblage data collected from 670 stream sites in Minnesota (U.S.A.) were used to calculate concordance across three nested spatial scales (statewide, ecoregion and catchment). Predictive taxa richness models, calibrated using the same data, were used to evaluate whether concordant communities exhibited similar trends in human‐induced taxa loss across all three scales. Finally, we evaluated the strength of the relationship between selected environmental variables and the composition of both assemblages at all three spatial scales. 2. Significant concordance between fish and invertebrate communities occurred at the statewide scale as well as in six of seven ecoregions and 17 of the 21 major catchments. However, concordance was not consistently indicative of significant relationships between rates of fish and invertebrate taxa loss at those same scales. 3. Fish and invertebrate communities were largely associated with different environmental variables, although the composition of both communities was strongly correlated with stream size across all three scales. 4. Predictive taxa‐loss models for fish assemblages were less sensitive and precise than models for invertebrate assemblages, likely because of the relatively low number of common fish taxa in our data set. Both models, however, distinguished reference from non‐reference sites. 5. The importance of concordance, geographic context and scale are discussed in relation to the design and interpretation of stream integrity indicators. In particular, our findings suggest that community concordance should not be viewed as a substitute for an evaluation of how assemblages respond to environmental stressors.     

Selecting discriminant function models for predicting the expected richness of aquatic macroinvertebrates

1. The predictive modelling approach to bioassessment estimates the macroinvertebrate assemblage expected at a stream site if it were in a minimally disturbed reference condition. The difference between expected and observed assemblages then measures the departure of the site from reference condition. 2. Most predictive models employ site classification, followed by discriminant function (DF) modelling, to predict the expected assemblage from a suite of environmental variables. Stepwise DF analysis is normally used to choose a single subset of DF predictor variables with a high accuracy for classifying sites. An alternative is to screen all possible combinations of predictor variables, in order to identify several ‘best’ subsets that yield good overall performance of the predictive model. 3. We applied best‐subsets DF analysis to assemblage and environmental data from 199 reference sites in Oregon, U.S.A. Two sets of 66 best DF models containing between one and 14 predictor variables (that is, having model orders from one to 14) were developed, for five‐group and 11‐group site classifications. 4. Resubstitution classification accuracy of the DF models increased consistently with model order, but cross‐validated classification accuracy did not improve beyond seventh or eighth‐order models, suggesting that the larger models were overfitted. 5. Overall predictive model performance at model training sites, measured by the root‐mean‐squared error of the observed/expected species richness ratio, also improved steadily with DF model order. But high‐order DF models usually performed poorly at an independent set of validation sites, another sign of model overfitting. 6. Models selected by stepwise DF analysis showed evidence of overfitting and were outperformed by several of the best‐subsets models. 7. The group separation strength of a DF model, as measured by Wilks’Λ, was more strongly correlated with overall predictive model performance at training sites than was DF classification accuracy. 8. Our results suggest improved strategies for developing reliable, parsimonious predictive models. We emphasise the value of independent validation data for obtaining a realistic picture of model performance. We also recommend assessing not just one or two, but several, candidate models based on their overall performance as well as the performance of their DF component. 9. We provide links to our free software for stepwise and best‐subsets DF analysis.     

Effects of fixed-count size on macroinvertebrate richness,site separation,and bioassessment of Chinese monsoonal streams

Subsampling that counts and identifies a random subset of individuals from field samples is widespread yet controversial. We evaluated the effects of fixed-count size on macroinvertebrate richness, site separation, and performance of modeled and null (i.e., natural variation adjusted and unadjusted, respectively) biological indices in Chinese monsoonal stream sites. To do so, we estimated the fixed-count size that was adequate to collect 95% of the estimated true regional macroinvertebrate taxa richness, and we also evaluated the effects of fixed-count size on site and group (reference vs test) separation, and the precision, sensitivity and responsiveness of modeled and null multimetric indices (MMI) and observed/expected (O/E) indices. Random subsamples of individual fixed-count sizes ranged from 50 to 500. Mean cumulative taxa richness continued to increase with increasing fixed-count size. We found that 150 and 200 individuals were needed to collect 75% of estimated true species richness 75% and 95% of the time, respectively. We estimated that at least 1500 individuals per site were required for collecting 95% of estimated true species richness. Site and group separation and classification strength also improved with increased fixed-count size. Larger fixed-count sizes improved the performance of modeled and null O/E₅₀ (O/E calculated using taxa with probabilities of capture ≥0.5); however, they showed no significant difference for modeled and null MMIs and O/E₀ (O/E calculated using all taxa). Overall, we found that fixed-counts affected richness and site/group separation, but not index performance. Until China develops standard sampling methods, we recommend using fixed-count sizes of 500 individuals and rarefaction of ≥200 individuals to limit the effects of sampling error for site and group separation and for precise and accurate bioassessment, respectively.     

Estimates and comparisons of the effects of sampling variation using ‘national’ macroinvertebrate sampling protocols on the precision of metrics used to assess ecological status

The Water Framework Directive (WFD) of the European Union requires all member countries to provide information on the level of confidence and precision of results in their river monitoring programmes to assess the ecological status class of river sites. As part of the European Union project STAR, the overall effects of sampling variation for a wide range of commonly used metrics and sampling methods were assessed. Replicate samples were taken in each of two seasons at 2–6 sites of varying ecological status class within each of 18 stream types spread over 12 countries, using both the STAR-AQEM method and a national sampling method or, where unavailable, the RIVPACS sampling protocol. The sampling precision of a combination of sampling method and metric was estimated by expressing the replicate sampling variance as a percentage P_samp of the total variance in metric values with a stream type; low values of P_samp indicate high precision. Most metrics had percentage sampling variances less than 20% for all or most stream types and methods. Most national methods including RIVPACS had sampling precisions at least as good as those for the STAR-AQEM method as used in their country at the same sites; the main exceptions were the national methods used in Latvia and Sweden. The national methods used in the Czech Republic, Denmark, France, Poland and the RIVPACS method used in the UK and Austria all had percentage sampling variances of less than 10% for the majority of metrics assessed. In contrast, none of the metrics had percentage sampling variances less than 10% when based on either the Italian (IBE) method, which used bank-side sorting, or the Latvian national method which identifies only a limited set of taxa. P_samp was lowest on average for the two stream types sampled in the Czech Republic using either the PERLA national method or the STAR-AQEM method. Averaged over all stream types and methods, the three Saprobic-based metrics had the lowest average percentage sampling variances (3–6%) amongst the 26 metrics assessed. These estimates of sampling standard deviation can be used to help assess the uncertainty in single or multi-metric systems for estimating site ecological status using the general STAR Bioassessment Guidance Software (STARBUGS) developed within the STAR project.     

A comparison of the European Water Framework Directive physical typology and RIVPACS-type models as alternative methods of establishing reference conditions for benthic macroinvertebrates

The EU Water Framework Directive requires European Union Member States to establish ‘type-specific biological reference conditions’ for streams and rivers. Types can be defined by using either a fixed typology (System-A), defined by ecoregions and categories of altitude, catchment area and geology, or by means of an alternative characterisation (System-B) that can use a variety of physical and chemical factors. Several European countries also have existing RIVPACS-type models that give site (rather than stream type) specific predictions of benthic macroinvertebrate communities. In this paper we compare the Water Framework Directive (WFD) System-A physical typology and three existing European multivariate RIVPACS-type models as alternative methods of establishing reference conditions. This work is carried out in Great Britain – using RIVPACS, Sweden – using SWEPAC_SRI and the Czech Republic – using PERLA. We found that in all three countries, all seasons and season combinations, and for all biotic indices tested, RIVPACS-type models were more effective (lower standard deviations of O/E ratios) than models based solely on the WFD System-A variables or null models (based on a single expectation for all sites). We also investigated the explanatory power of whole groups of WFD System-A variables and RIVPACS-type model variables, and the explanatory power of individual variables. We found that variables used in the RIVPACS-type models were often better correlates of macroinvertebrate community variation than the WFD System-A variables. We conclude that this is primarily because while the latter use very broad categories of map-derived variables, the former are based on continuous variables selected for their ecological significance.     

Macrophytes in boreal streams: Characterizing and predicting native occurrence and abundance to assess human impact

Macrophytes are a structurally and functionally essential element of stream ecosystems and therefore indispensable in assessment, protection and restoration of streams. Modelling based on continuous environmental gradients offers a potential approach to predict natural variability of communities and thereby improve detection of anthropogenic community change. Using data from minimally disturbed streams, we described natural macrophyte assemblages in pool and riffle habitats separately and in combination, and explored their variation across large scale environmental gradients. Specifically, we developed RIVPACS-type models to predict the presence and abundance of macrophyte taxa at stream sites in the absence of human influence and, used data from impacted streams to explore the responses of three biotic indices to anthropogenic stress. The indices used, taxonomic completeness (O/E-taxa), a measure of compositional dissimilarity (BC-index) and an index taking into account the abundance of species (AB-index), are based on predicted and observed macrophyte communities. We found that size of the catchment area, altitude, latitude and percentage of lakes in the catchment were the large scale environmental variables that best predicted the natural variation of assemblages. The RIVPACS approach substantially improved both the precision and accuracy to predict the natural communities and the sensitivity to human disturbance. O/E-taxa performed best in relation to the null model decreasing the variation by 20% in pools, 29% in riffles and 32% in combined data. In general, models based on the riffle assemblages performed better than models based on pool assemblages, but including both habitats and predicting abundances instead of only presence/absence yielded the greatest accuracy and sensitivity. Our results support the use of multivariate modelling techniques in predicting reference condition to assess status of stream macrophyte communities.     

Structure and Dynamics of a Benthic Invertebrate Community in an Intertidal Area of the Tagus Estuary, Western Portugal: A Six Year Data Series

A predictive model, incorporating macroinvertebrate and environmental data, similar to that developed for Australian rivers (AUSRIVAS) and British rivers (RIVPACS) was constructed using a dataset collected from 23 reference (least altered) wetlands on the Swan Coastal Plain, Western Australia, sampled in summer and spring, 1989 and spring, 1990. Four main groups of reference wetlands were identified by UPGMA classification (using the Bray–Curtis dissimilarity measure). Distinguishing environmental variables identified by Stepwise Multiple Discriminant Function Analysis were: calcium, colour (gilvin), latitude, longitude, sodium and organic carbon. Observed to expected ratios of taxa with a >50% chance of occurrence (OE50) derived from the model for a suite of 23 test wetlands sampled in spring, 1997, were significantly correlated with pH and the depth of the sampling sites. Greater discrimination between the test wetlands was provided by the OE50 ratios than either raw richness (number of families) or a biotic index (SWAMPS). Results obtained for a subset of 11 test wetlands sampled with both a rapid bioassessment protocol (incorporating field picking of 200 invertebrates collected in 2 min sweeps from selected habitats) and a semi-quantitative protocol (incorporating laboratory picking of all invertebrates collected in sweeps along 10 m transects at randomly allocated sites) were not significantly different, indicating that the former could be used to reduce the time and costs associated with macroinvertebrate-based wetland monitoring programs. In addition to providing an objective method of assessing wetland condition, predictive modelling provides a list of taxa expected to occur under reference conditions, which can be used as a target in wetland restoration programs. The probable impediment to widespread adoption of predictive modelling for wetland bioassessment is the need to produce models tailored to specific geographic regions and specific climatic conditions. This may incur significant costs in countries, such as Australia, which span a wide range of climatic zones.