Quantifying Variability in Four US Streams Using a Long-Term Data Set: Patterns in Water Quality Endpoints.

Temporal and spatial patterns of variability in aquatic ecosystems can be complex and difficult to quantify or predict. However, understanding this variability is critical to making a wide range of water quality assessment and management decisions effectively. Here we report on the nature and magnitude of spatial and temporal variation observed in conductivity, total phosphorus, and total nitrogen during a 15-year study of four U.S. stream systems receiving pulp and paper mill effluent discharges. Sampling locations included mainstem sites upstream and downstream of effluent discharge, as well as tributary sites. In all four stream systems, variability in conductivity as measured by the coefficient of variation was typically in the range of 10-50%, and was as low or lower than the variability in nutrient endpoints. The effect of effluent discharge was relatively minor overall, except in some site-specific instances. Some relatively large differences between tributary and mainstem variability were also observed. Flow variation tended to have a more consistent and larger effect on conductivity variation compared to the nutrient endpoints. After removing flow effects, significant relatively complex trends over time were observed at several sites. Changes in variability during the study also were observed. This paper highlights the importance of long-term studies to accurately characterize water quality variability used in water quality management decision-making.

Quantifying Variability in Four U.S. Streams Using a Long-Term Dataset: Patterns in Biotic Endpoints.

Effective water resources assessment and management requires quantitative information on the variability of ambient and biological conditions in aquatic communities. Although it is understood that natural systems are variable, robust estimates of long-term variation in community-based structure and function metrics are rare in U.S. waters. We used a multi-year, seasonally sampled dataset from multiple sites (n = 5-6) in four streams (Codorus Creek, PA; Leaf River, MS; McKenzie and Willamette Rivers, OR) to examine spatial and temporal variation in periphyton chlorophyll a, and fish and macroinvertebrate metrics commonly used in bioassessment programs. Within-site variation of macroinvertebrate metrics and benthic chlorophyll a concentration showed coefficient of variation ranging from 16 to 136%. Scale-specific variability patterns (stream-wide, season, site, and site-season patterns) in standardized biotic endpoints showed that within-site variability patterns extended across sites with variability greatest in chlorophyll a and lowest in Hilsenhoff's Biotic Index. Across streams, variance components models showed that variance attributed to the interaction of space and time and sample variance accounted for the majority of variation in macroinvertebrate metrics and chlorophyll a, while most variation in fish metrics was attributed to sample variance. Clear temporal patterns in measured endpoints were rare and not specific to any one stream or assemblage, while apparent shifts in metric variability related to point source discharges were seen only in McKenzie River macroinvertebrate metrics in the fall. Results from this study demonstrate the need to consider and understand spatial, seasonal, and longer term variability in the development of bioassessment programs and subsequent decisions.

Assessing the fit of biotic ligand model validation data in a risk management decision context.

Biotic ligand models (BLMs) have advanced the ability to predict the concentrations of metals in surface waters likely to harm aquatic organisms. BLMs have been developed for several metals including Cu, Zn, Cd, and Ag. Additionally, the US Environmental Protection Agency has published guidance on the use of a BLM to develop water quality criteria for Cu. To validate the predictive performance of many BLMs, model predictions based on test water quality have been compared with corresponding laboratory toxicity measurements. Validation results are typically described in the published literature in terms of the proportion of predicted effect concentrations that fall within a factor of 2 of measured values. In this article, an alternative is presented using a receiver operating characteristics approach and regression prediction limit analyses, quantifying the probabilities of true and false predictions of excess toxicity risk based on toxic unit calculations and a risk management threshold of 1. The approaches are applied to a published Zn BLM and 3 simulated data sets that reflect attributes of other published BLM validation data. The overall accuracy of the unified Zn BLM is estimated to be 80% to 90%, and analyses of simulated data suggest a similar level of accuracy for other published BLMs. Further application of these validation methods to other BLMs may provide more complete and transparent information on their possible predictive value when used in the management of risks due to aqueous metals.

Maximizing the accuracy of field-derived numeric nutrient criteria in water quality regulations.

High levels of the nutrients nitrogen and phosphorus can cause unhealthy biological or ecological conditions in surface waters and prevent the attainment of their designated uses. Regulatory agencies are developing numeric criteria for these nutrients in an effort to ensure that the surface waters in their jurisdictions remain healthy and productive, and that water quality standards are met. These criteria are often derived using field measurements that relate nutrient concentrations and other water quality conditions to expected biological responses such as undesirable growth or changes in aquatic plant and animal communities. Ideally, these numeric criteria can be used to accurately "diagnose" ecosystem health and guide management decisions. However, the degree to which numeric nutrient criteria are useful for decision making depends on how accurately they reflect the status or risk of nutrient-related biological impairments. Numeric criteria that have little predictive value are not likely to be useful for managing nutrient concerns. This paper presents information on the role of numeric nutrient criteria as biological health indicators, and the potential benefits of sufficiently accurate criteria for nutrient management. In addition, it describes approaches being proposed or adopted in states such as Florida and Maine to improve the accuracy of numeric criteria and criteria-based decisions. This includes a preference for developing site-specific criteria in cases where sufficient data are available, and the use of nutrient concentration and biological response criteria together in a framework to support designated use attainment decisions. Together with systematic planning during criteria development, the accuracy of field-derived numeric nutrient criteria can be assessed and maximized as a part of an overall effort to manage nutrient water quality concerns.

Assessing the predictive performance of risk-based water quality criteria using decision error estimates from receiver operating characteristics (ROC) analysis.

Field data relating aquatic ecosystem responses with water quality constituents that are potential ecosystem stressors are being used increasingly in the United States in the derivation of water quality criteria to protect aquatic life. In light of this trend, there is a need for transparent quantitative methods to assess the performance of models that predict ecological conditions using a stressor-response relationship, a response variable threshold, and a stressor variable criterion. Analysis of receiver operating characteristics (ROC analysis) has a considerable history of successful use in medical diagnostic, industrial, and other fields for similarly structured decision problems, but its use for informing water quality management decisions involving risk-based environmental criteria is less common. In this article, ROC analysis is used to evaluate predictions of ecological response variable status for 3 water quality stressor-response data sets. Information on error rates is emphasized due in part to their common use in environmental studies to describe uncertainty. One data set is comprised of simulated data, and 2 involve field measurements described previously in the literature. These data sets are also analyzed using linear regression and conditional probability analysis for comparison. Results indicate that of the methods studied, ROC analysis provides the most comprehensive characterization of prediction error rates including false positive, false negative, positive predictive, and negative predictive errors. This information may be used along with other data analysis procedures to set quality objectives for and assess the predictive performance of risk-based criteria to support water quality management decisions.

Estimating the designated use attainment decision error rates of US Environmental Protection Agency's proposed numeric total phosphorus criteria for Florida, USA, colored lakes.

The utility of numeric nutrient criteria established for certain surface waters is likely to be affected by the uncertainty that exists in the presence of a causal link between nutrient stressor variables and designated use-related biological responses in those waters. This uncertainty can be difficult to characterize, interpret, and communicate to a broad audience of environmental stakeholders. The US Environmental Protection Agency (USEPA) has developed a systematic planning process to support a variety of environmental decisions, but this process is not generally applied to the development of national or state-level numeric nutrient criteria. This article describes a method for implementing such an approach and uses it to evaluate the numeric total P criteria recently proposed by USEPA for colored lakes in Florida, USA. An empirical, log-linear relationship between geometric mean concentrations of total P (a potential stressor variable) and chlorophyll a (a nutrient-related response variable) in these lakes-that is assumed to be causal in nature-forms the basis for the analysis. The use of the geometric mean total P concentration of a lake to correctly indicate designated use status, defined in terms of a 20 µg/L geometric mean chlorophyll a threshold, is evaluated. Rates of decision errors analogous to the Type I and Type II error rates familiar in hypothesis testing, and a 3rd error rate, E(ni) , referred to as the nutrient criterion-based impairment error rate, are estimated. The results show that USEPA's proposed "baseline" and "modified" nutrient criteria approach, in which data on both total P and chlorophyll a may be considered in establishing numeric nutrient criteria for a given lake within a specified range, provides a means for balancing and minimizing designated use attainment decision errors.

Using Monte Carlo analysis to characterize the uncertainty in final acute values derived from aquatic toxicity data.

Many ambient water quality criteria established to protect aquatic life from acute toxicity are calculated using a procedure described in the US Environmental Protection Agency's "1985 Guidelines" (USEPA 1985). The procedure yields a final acute value (FAV) from acceptable median lethal or effective concentrations (LC50 or EC50, respectively) that is a single-point, deterministic estimate of the concentration of a chemical substance that will protect 95% of aquatic species from >50% mortality or other acute toxic effects. However, because of variation and uncertainty associated with toxicity test results, uncertainty in the estimated FAV exists that is not accounted for by the 1985 Guidelines procedure. Here, Monte Carlo analysis is used to characterize this uncertainty. The analysis uses Cu EC50 values adjusted for differences in test water chemistry obtained from USEPA's final freshwater Cu criteria guidance published in 2007. Additional Monte Carlo simulations illustrate Cu FAV distributions obtained using a subset of tested species and assuming fewer replicate tests. The deterministic procedure yields an FAV of 4.68 µg/L for the complete data set. By comparison, 3 replicate Monte Carlo simulations yielded mean FAVs of 4.66 µg/L. The 5th and 95th percentiles of the distribution of calculated FAVs were 4.14 µg/L and 5.20 µg/L, respectively. Reducing the number of tested genera from 27 to 8 (the minimum recommended by the 1985 Guidelines) and setting the number of tests per species equal to 3 yielded 5th and 95th percentiles of 1.22 µg/L and 6.18 µg/L, respectively, compared to a deterministic estimate of 2.80 µg/L. Results of this study indicate that Monte Carlo analysis can be used to improve the understanding and communication of uncertainty associated with water quality criteria derived from acute toxicity data using the 1985 Guidelines. This may benefit the development, revision, and application of these criteria in the future.