SEARCH

SEARCH BY CITATION

Keywords:

  • Biotic ligand model;
  • Ecological risk assessment;
  • Heavy metal;
  • Environmental standard;
  • Bayesian statistics

Abstract

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. DATA AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. SUPPLEMENTAL DATA
  8. Acknowledgements
  9. REFERENCES
  10. Supporting Information

Biotic ligand models (BLMs) have been broadly accepted and used in ecological risk assessment of heavy metals for toxicity normalization with respect to water chemistry. However, the importance of assessing bioavailability by using BLMs has not been widely recognized among Japanese stakeholders. Failing to consider bioavailability may result in less effective risk management than would be possible if currently available state-of-the-art methods were used to relate bioavailable concentrations to toxic effects. In this study, an ecological risk assessment was conducted using BLMs for 6 rivers in Tokyo to stimulate discussion about bioavailability of heavy metals and the use of BLMs in ecological risk management in Japan. In the risk analysis, a Bayesian approach was used to take advantage of information from previous analyses and to calculate uncertainties in the estimation of risk. Risks were judged to be a concern if the predicted environmental concentration exceeded the 5th percentile concentration (HC5) of the species sensitivity distribution. Based on this criterion, risks to stream biota from exposure to Cu were judged not to be very severe, but it would be desirable to conduct further monitoring and field surveys to determine whether temporary exposure to concentrations exceeding the HC5 causes any irreversible effects on the river ecosystem. The risk of exposure to Ni was a concern at only 1 of the 6 sites. BLM corrections affected these conclusions in the case of Cu but were moot in the case of Ni. The use of BLMs in risk assessment calculations for Japanese rivers requires water quality information that is, unfortunately, not always available. Integr Environ Assess Manag 2013; 9: 63–69. © 2012 SETAC


INTRODUCTION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. DATA AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. SUPPLEMENTAL DATA
  8. Acknowledgements
  9. REFERENCES
  10. Supporting Information

The Japanese government has recently begun to establish environmental quality standards to protect aquatic organisms. The first (and so far, the only) environmental quality standard to protect aquatic organisms was set for Zn in 2006. Seventeen other substances are currently listed as candidates for setting environmental quality standards (Central Environmental Council Environmental Water Group 2009), among them are 3 heavy metals: Cu, Ni, and Cd. Whether environmental quality standards will be set for these metals depends on whether forthcoming risk assessments by the environmental council conclude that risks from these metals are a concern in Japan.

The ecotoxicity of heavy metals is known to depend on water chemistry. Previous studies have suggested that the free ion forms of heavy metals are bioavailable and primarily responsible for toxic effects on organisms (Campbell 1995). Biotic ligand models (BLMs) are ecotoxicity models that assume that toxicity occurs when the metal-biotic ligand complex reaches a critical concentration. BLM models therefore consider that the apparent toxicity of heavy metals depends also on a competition for the biotic ligand between the toxic metal ion and the other cations (Di Toro et al. 2001; Niyogi and Wood 2004). BLMs have been broadly accepted and used in the ecological risk assessment of heavy metals (European Union 2008, 2009; van Sprang et al. 2009) because of their demonstrated ability to more accurately simulate the effects of water chemistry on toxicity than conventional methods such as equations that relate toxicity to water hardness. The importance of considering bioavailability in the risk assessment of heavy metals, however, has not been broadly recognized among Japanese stakeholders. This is partly because the relatively low hardness of river waters in Japan has not historically focused attention on the effects of water chemistry on toxicity. Failure to consider bioavailability can lead to inaccurate assessments of ecological risk and less effective risk management than could be provided by implementing currently available state-of-the-art methods that relate toxicity to bioavailability.

A constructive way to enhance discussion about bioavailability of heavy metals and the use of BLMs in Japan is to present an example of actual ecological risk assessment using BLMs for Japanese rivers. Hayashi and Kashiwagi (2011) is apparently the only such example. Their main goal was to use a Bayesian approach to conduct probabilistic risk assessments and risk comparisons of 9 substances in Tokyo rivers. The use of BLMs to estimate bioavailable concentrations of Ni and Cu revealed that taking bioavailability into account affected the risk estimates of the metals, especially in the case of Cu. The risk assessment calculations were limited, however, by the fact that only the median values of water chemistry parameters (e.g., pH, hardness, and total organic carbon [TOC]) from all Tokyo rivers were considered in the BLM normalizations. The analysis thus did not show how the risk assessment results for each river would change if the chemistry of each river were considered in the calculations. In terms of communication among stakeholders, risk assessments based on the chemistry of each river are clear and intuitive examples that show the realistic behavior of BLM corrections and their effects on risk assessment.

The main goal of this study is to promote discussion about bioavailability of heavy metals and the use of BLMs in Japan by presenting an ecological risk assessment using BLMs for 6 rivers in Tokyo. This ecological risk assessment also serves to delineate problems that currently exist in the application of BLMs to Japanese rivers before the forthcoming consideration of more sophisticated ecological risk assessments for the establishment of environmental water quality standards. This study focuses on the risk assessment of Cu and Ni because establishment of standards for these 2 metals is going to be considered in Japan in the near future.

A technical aim of this study is to develop an analytical approach for exposure and risk assessment even when informed calculations are constrained by a paucity of environmental monitoring data. At the time of writing, the available data for Cu and Ni in Tokyo rivers were so limited that conducting a statistical analysis with reasonable accuracy and precision was difficult. This study incorporates a framework of Bayesian statistics to overcome this difficulty by using information and distributions of data from a previous study. The statistical approach used in this study is applicable to exposure and risk assessment in a broader context.

DATA AND METHODS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. DATA AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. SUPPLEMENTAL DATA
  8. Acknowledgements
  9. REFERENCES
  10. Supporting Information

Data selection and compilation

Environmental monitoring data

Monitoring data for Tokyo rivers collected by the Tokyo Metropolitan Government during the period 2006–2009 (http://www.kankyo.metro.tokyo.jp/water/tokyo_bay/measurements/data/index.html) were used in the exposure analyses. Metal concentrations in this article are presented as total concentrations. Because the number of Cu and Ni data collected in 2006–2009 was limited, the results of a previous analysis (Hayashi and Kashiwagi 2011) based on data from 2000–2005 were also used to improve the precision of the exposure estimates.

Selection of risk assessment sites

Initially, 15 sites were selected as candidates for risk assessment because TOC concentrations, a key factor in BLM calculations, were available from monitoring studies at all 15 sites. BLM calculations typically use DOC rather than TOC data. However, I used TOC data because no DOC data were available in the Tokyo monitoring data set. The median TOC concentrations at these 15 sites ranged from 1.0 to 5.3 mg/L, and the median pH values from 7.25 to 8.75 (Supplemental Data Table A2 and Supplemental Data Figure A1a–c). With the exception of the Miyako bridge site on the Onda river, all median pH values were in the range 7.25 to 7.8. No hardness data were available at these 15 sites. From these 15 sites, 5 were finally selected as risk assessment sites (Table 1 and Supplemental Data Figure A2). I used the availability of both Cu and Ni data as a criterion for site selection so that I could conveniently compare the effects of the BLM corrections for these metals. When multiple sites that satisfied this criterion had similar TOC values, I chose only 1 site among them, because the purpose of this study was to determine the effects of BLM corrections at sites with different water chemistries rather than to provide ecological risk assessments at many sites with similar water chemistries. I did not select the Miyako bridge site because the pH of the water there was outside the applicable range of the BLMs. The median TOC values at these 5 sites spanned the range of values for all sites (1.0–5.3) and their median pH values (7.4–7.7) included the median pH values at 12 of the 15 sites (Supplemental Data Figure A1c). Monitoring data for both Ni and Cu were available at all sites except site A. Site A lacked Ni concentration data, but it was included because its median TOC (5.3 mg/L) was the highest among all sites.

Table 1. TOC, pH, and Cu and Ni concentration data for 5 rivers in Tokyo
CuRisk assessment siteTOC (mg/L)pHnOrder statisticsBayesian analysis
MedMax50th95th
  1. A = Uchitakumi bridge on Ayase River; B = Horikiri bridge on Ara River; C = Tsurumaichigou bridge on Sakai River; D = Denenchoufu dam on Tama River; E = Haijima bridge on Tama River; Med = median value; Max = maximum value; n = number of data; TOC = total organic carbon; — = no data; < = a concentration below the detection limit.

  2. All concentrations are in µg/L. Values in parentheses denote the 90% credible interval.

CuA5.37.453610217.7 (7.0, 8.4)17.7 (15.9, 19.7)
 B4.27.4194114.9 (4.3, 5.6)9.7 (11.3, 13.1)
 C3.17.513<10107.3 (8.4, 9.7)19.0 (16.3, 22.3)
 D2.57.67<4104.4 (3.6, 5.5)10.2 (8.2, 12.8)
 E1.07.77<4<42.3 (1.7, 3.0)5.2 (3.9, 7.0)
NiA5.37.450
 B4.27.4472311.5 (8.0, 16.6)38.9 (26.3,58.7)
 C3.17.52<1<10.9 (0.6, 1.4)3.0 (1.9, 4.9)
 D2.57.63<5<50.9 (0.4, 2.0)3.2 (1.5, 7.2)
 E1.07.73<5<50.8 (0.4, 1.4)2.7 (1.5, 4.9)
Collection and compilation of ecotoxicity data

Ecotoxicity data were collected from European Union (EU) risk assessment reports for Cu (EU 2008) and Ni (EU 2009). Chronic EC10 (the effective concentration causing harm to 10% of test organisms) data with regard to endpoints that would potentially cause population-level effects (i.e., survival, growth, reproduction, hatching, and development) were used in the effect assessment. Chronic no-observed-effect concentrations (NOEC) were assumed to be equivalent to EC10 values. This equivalency of NOEC and EC10 concentrations is an assumption that appears in the EU risk assessment reports (EU 2008, 2009). If different endpoint EC10 data for a substance were available for a single species, the lowest EC10 (i.e., most sensitive endpoint) was used in the calculations. If there were multiple EC10 values measured for the most sensitive endpoint, I used the EC10 value measured under water quality conditions most comparable with the water quality at the risk assessment sites. The data I used as criteria were the min–max ranges of pH, TOC, and hardness at the assessment sites (7.4–7.8 for pH and 1–5.3 for TOC; Table A1). For hardness, I chose a range that included 90% of the hardness values in river waters in the entire Tokyo region by excluding the highest 5% and lowest 5% of the values. I used this range, 46 to 114 mg CaCO3/L, as a criterion because no hardness data were available at the risk assessment sites. I selected the EC10 data measured under test conditions that included the greatest number of water quality indices (i.e., pH, TOC, hardness) within these ranges as the “most comparable” data. If multiple EC10 data satisfied this most comparable criterion, I equated the geometric mean of those EC10 values to the EC10 value for the species.

Normalization of ecotoxicity data by BLM

The ecotoxicity data for Cu and Ni were normalized by converting them to free ion activities in accord with procedures in EU risk assessment reports (EU 2008, 2009). In the normalization procedure, I first transformed EC10 data in terms of total concentrations from the original tests into free ion activities. I then corrected the free ion activities of Cu and Ni at the EC10 for the water chemistry at the risk assessment sites by using BLM equations and activities of relevant species. Finally, I transformed the BLM-corrected free ion activities of Cu and Ni to total Cu and Ni concentrations by taking into consideration the water chemistry at the risk assessment sites. To make this transformation, I used the median hardness in river water in the entire Tokyo region (76 mg CaCO3/L) because no hardness data were available at the risk assessment sites. I have reported the effects of hardness on the BLM corrections in the Supplemental Material (Supplemental Data Table A4). I carried out the transformations between total concentrations and free ion activities with WHAM version 7 software (http://www.ceh.ac.uk/products/software/wham/index.html). Although this version of WHAM was different than those used in the EU risk assessment reports (i.e., versions 5 and 6 were used in the 2008 and 2009 EU reports, respectively), I used the latest version in this analysis because it incorporated the most improved binding models. A preliminary analysis with WHAM versions 5, 6, and 7 and vMINTEQ ver 2.61 (http://www.lwr.kth.se/English/OurSoftware/vminteq/) revealed that BLM normalizations with these models produced assessment results little different from those obtained with WHAM version 7 (i.e., HC5). The equations, parameters, and details of assumptions in this BLM normalization are presented in the Supplemental Material.

Methods for statistical analysis and risk assessment

Exposure assessment methods

Predicted environmental concentrations (PECs) were derived from environmental monitoring data. Given the limited quantity and quality of the monitoring data (i.e., many data were reported as less than the detection limit [LDL]), it was difficult to properly derive the PECs. To solve this problem, 2 complementary approaches were used in the analysis. The first was a simple and intuitive approach that gave point estimates, whereas the second was a complex approach that gave estimates of probability distributions of concentrations and associated uncertainties. In the first approach, PECs were derived from the median and maximum values of the ranked monitoring data, and these were used as the PECs (denoted as PECmed and PECmax, respectively) at each assessment site. This ordered statistics approach treated LDL data exactly as they were reported (i.e., <detection limit).

In the second approach, environmental concentration distributions (ECDs) were estimated using a Bayesian framework and the results from a previous analysis. Information from the previous analysis was introduced by using prior distributions, and current monitoring data were used to update the prior distributions. The update was numerically conducted based on the following equation

  • equation image(1)

Here, θ and σ are the mean and standard deviation of the common logarithm of environmental concentration data, and p(θ) and p(σ) are their prior distributions. X represents environmental monitoring data in the current analysis (i.e., data from 2006 to 2009), which are assumed to follow normal distributions N(θ, σ). All environmental monitoring data were treated in the common logarithm form in the calculation. LDL data were incorporated through MCMC simulations by iterative replacement with sampled numerical values (Supplemental Data). In the analysis, normal distributions were assumed for p(θ) and p(σ). The prior distributions of θ and σ at each risk assessment site were taken from Hayashi and Kashiwagi (2011) and converted to normal distributions for purposes of this analysis (Supplemental Data). p(θ, σ|X) is the updated (posterior) distribution of the parameters. In this approach, the 50th and 95th percentiles in the estimated ECDs were used as PECs (denoted by PEC50 and PEC95, respectively). Note that this Bayesian approach was intended not to be a surrogate for but rather a complement to the first (i.e., ordered statistics) approach.

The posterior distributions of the ECD parameters (i.e., θ and σ) were calculated by Markov-chain Monte Carlo (MCMC) simulations (Gelman 2004) using WinBUGS (Lunn et al. 2000). The incorporation of LDL data in the ECD estimation was conducted by using a WinBUGS code (Supplemental Material). In the implementation, the first 10 000 iterations were discarded as a burn-in period and then 100 000 consecutive iterations were conducted. Parameter values were drawn every 10 iterations, which resulted in posterior distributions of the parameters with 10 000 MCMC samples. All MCMC samples were judged to have converged to the corresponding posterior distributions because the index of convergence ranged between 1.0 and 1.1 in all cases (Gelman 2004). The analyses of ECDs were done with R (R Development Core Team 2007) and WinBUGS. All source codes for the calculations in this study are available at https://www.sugarsync.com/pf/D852288_4182053_74576.

Effect assessment methods using species sensitivity distributions

The 5th percentile concentration (HC5) of the species sensitivity distribution (SSD) was used as an effect index of Cu and Ni. SSDs were derived assuming that the common logarithm of the EC10 followed a normal distribution. The distribution parameters (i.e., mean and variance) of the SSDs were calculated by WinBUGS so that the posterior sample values could be used for the later uncertainty analysis in the risk calculation. For the prior distributions, a normal distribution with mean 0 and variance 106 was assumed for the mean parameter. A γ distribution with both scale and shape parameters being 0.001 was assumed for the variance parameter. These prior distributions are sufficiently flat that they convey virtually no information (i.e., they are noninformative prior distributions). Given these assumptions, the SSDs for Ni and Cu were effectively derived with no prior information other than the ecotoxicity data used in the analysis.

Risk calculation methods

Hazard quotients (HQs) were used as risk indices. In this study, a hazard quotient was defined as

  • equation image(2)

where x specifies which PEC was used among the 4 PECs (i.e., PECmed, PECmax, PEC50, or PEC95). In general an HQ greater than 1 is considered to imply that risk is potentially a concern (Posthuma et al. 2002), and that same assumption was adopted in this analysis.

The probability of an environmental concentration exceeding HC5 (ExHC5) was also calculated as a probabilistic risk index. The value of ExHC5 can be interpreted as the fraction of time that environmental concentrations exceed HC5 at a site. The ExHC5 × 100 at a site equals 100% minus the percentile of the ECD corresponding to the HC5 at the site.

Uncertainty analysis

When the estimates from the Bayesian approach (i.e., PEC50, PEC95, and ExHC5) were used as exposure indices, the medians and uncertainties of the HQs and ExHC5s were calculated by Monte Carlo simulation by using the 10 000 posterior numerical samples of SSD and ECD parameters as input to the simulation (see the program code for details of the implementation).

RESULTS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. DATA AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. SUPPLEMENTAL DATA
  8. Acknowledgements
  9. REFERENCES
  10. Supporting Information

Exposure assessment

The highest Cu concentrations were reported at site A (Table 1). More than half of the Cu concentrations were LDL at sites C and D, and all Cu concentrations were LDL at site E. For Ni fewer than 5 data were reported at any of the sites. The highest concentrations were reported at site B (Table 1). All of the Ni concentrations were LDL at sites C, D, and E (note that site A had no data for Ni). The median values from the current data were outside of the 90% confidence interval of the 50th percentile from the Bayesian analysis in some cases (Table 1). This could be caused by a potential discrepancy between the current data (2006–2009) and the previous data (2000–2005).

Effect assessment

For Cu, there is a clear tendency that increasing TOC increases HC5 (Table 2). The HC5 derived from the SSD without BLM normalization was 4.3 µg/L, which was slightly higher than the HC5 (2.5 µg/L) at the site with the lowest TOC (site E, TOC = 1 mg/L). For Ni, the tendency for increasing TOC to increase the HC is less pronounced. The HC5 derived from the SSD without BLM normalization was 5.3 µg/L (Table 3). This value was close to the HC5 (5.2 µg/L) at site E. The details of the estimated SSD parameters are shown in Supplemental Data Table A2.

Table 2. BLM-normalized HC5 concentrations at the risk assessment sites
SiteTOC (mg/L)pHHC5
Cu (µg/L)Ni (µg/L)
  1. A = Uchitakumi bridge on Ayase River; B = Horikiri bridge on Ara River; C = Tsurumaichigou bridge on Sakai River; D = Denenchouhus dam on Tama River; E = Haijima bridge on Tama River; BLM = biotic ligand model; HC5 = 5th percentile of species sensitivity distribution; TOC = total organic carbon.

  2. Values in parentheses denote the 90% credible interval.

A5.37.4512.8 (8.5, 17.2)8.6 (4.6, 13.9)
B4.27.410.6 (6.9, 14.6)8.4 (4.4, 13.6)
C3.17.57.7 (4.9, 10.5)7.4 (3.9, 12.1)
D2.57.66.2 (4.0, 8.6)7.5 (3.9, 12.4)
E1.07.72.5 (1.4, 3.8)5.3 (2.8, 8.8)
No BLM normalization4.3 (2.4, 6.6)5.2 (2.5, 9.0)
Table 3. Risk assessment results for Cu and Ni
 SiteTOC (mg/L)pHHQmedHQmed (no BLM)HQmaxHQmax (no BLM)HQ50HQ50 (no BLM)HQ95HQ95 (no BLM)ExHC5ExHC5 (no BLM)
  1. A = Uchitakumi bridge on Ayase River; B = Horikiri bridge on Ara River; C = Tsurumaichigou bridge on Sakai River; D = Denenchouhus dam on Tama River; E = Haijima bridge on Tama River; ExHC5 = probability of the environmental concentration exceeding HC5; HQmed, HQmax, HQ50, HQ95 = hazard quotients based on PECmed, PECmax, PEC50, and PECmax, respectively; “no BLM” = specifies the case without BLM normalization; TOC = total organic carbon.

  2. Values in parentheses denote the 90% credible interval.

CuA5.37.450.782.31.644.90.60 (0.44, 0.91)1.8 (1.1, 3.2)1.4 (1.0, 2.1)4.1 (2.6, 7.4)0.16 (0.052, 0.43)0.87 (0.61, 0.99)
 B4.27.40.380.931.02.60.46 (0.33, 0.73)1.1 (0.74, 2.0)1.1 (0.75, 1.7)2.6 (1.7, 4.8)0.065 (0.013, 0.27)0.60 (0.27, 0.92)
 C3.17.5<1.3<2.31.32.31.1 (0.77, 1.7)1.9 (1.2, 3.5)2.5 (1.7, 4.0)4.4 (2.8, 8.0)0.58 (0.30, 0.87)0.91 (0.67, 0.99)
 D2.57.6<0.65<0.931.62.30.72 (0.48, 1.2)1.0 (0.6, 1.9)1.7 (1.1, 2.7)2.4 (1.5, 4.4)0.26 (0.075, 0.62)0.52 (0.19, 0.89)
 E1.07.7<1.6<0.93<1.6<0.930.91 (0.55, 1.7)0.53 (0.32, 1.0)2.1 (1.3, 3.9)1.2 (0.72, 2.3)0.43 (0.12, 0.85)0.10 (0.011, 0.50)
NiA5.37.45
 B4.27.40.661.352.174.41.4 (0.75, 2.8)2.2 (1.1, 4.9)4.7 (2.5, 9.8)7.5 (3.8, 17)0.67 (0.35, 0.92)0.86 (0.57, 0.99)
 C3.17.5<0.13<0.190.130.190.12 (0.062, 0.25)0.17 (0.084, 0.38)0.4 (0.21, 0.89)0.59 (0.28, 1.3)0.0024 (6.6 × 10−5, 0.036)0.0094 (3.5 × 10−4, 0.10)
 D2.57.6<0.81<0.96<0.81<0.960.13 (0.048, 0.33)0.18 (0.068, 0.52)0.44 (0.17, 1.2)0.63 (0.24, 1.8)0.0030 (2.1 × 10−5, 0.078)0.012 (1.4 × 10−4, 0.20)
 E1.07.7<2.0<0.96<2.0<0.960.15 (0.068, 0.35)0.15 (0.068, 0.37)0.52 (0.23, 1.2)0.53 (0.23, 1.3)0.0061 (1.3 × 10−4, 0.084)0.0064 (1.2 × 10−4, 0.097)

Risk calculation

For Cu, HQmax (i.e., PECmax/HC5) exceeded or was close to 1.0 at all sites except E (Table 3). At site E, all data were LDL so that HQmax was less than 1.6. HQmed (i.e., PECmed/HC5) was less than 1.0 at sites A, B, and D. At sites C and E, all data were LDL, so that HQmed values were less than 1.3 and less than 1.6. The median HQ95 (i.e., PEC95/HC5) exceeded 1.0 at all sites. The median HQ50 was less than 1.0 at all sites except site C, although the HC50 values at sites D and E were relatively high (0.72 and 0.91), and the error bounds included 1.0 (Table 3). The ExHC5 (i.e., the probability of the environmental concentration exceeding HC5) was highest at site C (0.58). The implication is that environmental concentrations exceed HC5 for a total of approximately 7 months per year at site C. At the other sites ExHC5 was approximately 0.1 to 0.4. HQs derived from the original ecotoxicity data without BLM normalization were up to approximately 4 times those derived using BLM normalization. This result suggests that risk assessment without BLM normalization is conservative in the sense that it generally tends to overestimate risk in the case of Cu.

For Ni, all of the HQs (i.e., HQmax, HQmed, HQ95 and HQ50) were less than 1.0 except at site B (Table 3). ExHC5 was also exceptionally high (0.67) at site B, whereas ExHC5 was no greater than 0.01 at the other sites. HQs derived from the original ecotoxicity data without BLM normalization were not much different from the HQs derived from the BLM-normalized data. This result suggests that risk estimates are much less sensitive to BLM normalization for Ni versus Cu, at least within the range of water chemistry treated in this study. For Ni, water hardness may have larger effects on toxicity than do DOC and pH (Supplemental Data).

DISCUSSION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. DATA AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. SUPPLEMENTAL DATA
  8. Acknowledgements
  9. REFERENCES
  10. Supporting Information

Conclusion from the risk assessments

The result of the Cu risk assessment showed that HQmax (=PECmax/HC5) and the median HQ95 (=PEC95/HC5) exceeded 1.0 at most risk assessment sites, whereas HQmed (=PECmed/HC5) and the median HQ50 (=PEC50/HC5) did not exceed 1.0 at most of the sites. ExHC5 values indicated that Cu concentrations would be a concern during approximately 7 months of the year at site C and for several months per year at the other 4 sites. Thus the current Cu risk at these sites does not seem to be very severe, but it would be desirable to get further information and carry out some field surveys to determine whether concentrations that temporarily exceed the HC5 at these sites have any permanent effects on the river ecosystem.

For Ni, all HQs (i.e., HQmax, HQ95, HQmed, and HQ50) exceeded 1.0 at site B, and the ExHC5 (Table 3) implies that Ni concentrations would be a concern approximately 70% of the time at that site. It would be desirable to conduct further risk assessments including more detailed environmental monitoring and field surveys at site B. In contrast, even HQmax and the median HQ95 were less than 1.0 at all the other sites (Table 3). This suggests that the risk of Ni toxicity is not a concern at these sites. However, uncertainties remain about the representativeness of the estimated PECmax and PEC95 values because the monitoring data for Ni were severely limited. Further environmental monitoring would be desirable for Ni to check the accuracy of the current exposure and risk assessment.

Limitations to the risk assessment

Despite the effort to use currently available data and methods as effectively as possible, the risk assessment in this study has several important limitations. First, both the quantity and quality of the environmental monitoring data were insufficient to conduct a highly dependable risk analysis. Especially in the case of the monitoring data for Tokyo rivers, there was no site at which both TOC and water hardness data were available. The risk assessment was thus compromised by the fact that the BLM normalizations used the median water hardness data for all Tokyo rivers as the surrogate for the water hardness at each site. The use of TOC instead of DOC was a source of uncertainty in the BLM corrections. In addition, the number of Ni concentration data was small, and most of the data were reported as LDL, which also limits the accuracy of the exposure assessment and the process of PEC derivation. The use of median values of other water quality data (i.e., temperature, TOC, and pH) for each site also constrained the assessment of effects caused by temporal variability in water chemistry. In the upcoming risk assessments for the determination of new environmental standards, a strategic monitoring strategy will be needed to compensate for the limitations resulting from the low quality and quantity of currently available data. Moreover, although the BLMs used in this assessment have been reasonably validated in Europe (EU 2008, 2009), the applicability of these same BLMs to Japanese rivers is problematic. The BLMs and chemical speciation models used here were developed and validated based on studies using European or North American waters and organisms. It is unclear whether the models can properly determine the effect of Japanese river waters on the organisms in Japanese rivers (see below for further discussion).

Advantages of Bayesian analysis

In this study the risk assessment was conducted using Bayesian methods. In the exposure analysis, prior distributions were used from a previous analysis, and current monitoring data were used to update those distributions. This approach enabled a stable and robust estimation of the ECD from limited data. The analysis, however, included some uncertainties with respect to the approximation of posterior sample values based on a past analysis in which distributions were described by a normal distribution function (Supplemental Material). Another concern is the age of the past data, which might have biased the estimation of current ECDs in the rivers if the ECDs were changing with time. Thus, it would be unwise to regard this Bayesian approach as a complete surrogate for more conventional approaches. In this study, the Bayesian approach was regarded as a complement to the basic approach based on ordered statistics, and results from both approaches were examined throughout the article. The analysis based on the Bayesian approach enabled an examination of the risk based on multiple sources of evidence.

The other advantages of using a Bayesian framework in this study were the incorporation of LDL data in the ECD estimation without arbitrary data manipulation, and the use of the sample values from the posterior distribution created by the MCMC simulations as input to a Monte Carlo analysis to calculate and present uncertainty in the risk estimation.

Application of BLM to rivers in Tokyo and throughout Japan

The effect of BLM normalizations on the risk assessment results were clear in the case of Cu but less obvious in the case of Ni (Table 3). This suggests that BLM normalization is important to Cu risk assessment in Tokyo river waters but is relatively unimportant in Ni risk assessment. The EU (2008, 2009) risk assessment reports showed that BLM normalizations led to important changes in HC5s for both Cu and Ni. This apparent discrepancy in the effects of BLM normalization for Ni between this study and EU (2009) can be explained by the difference in the range of water chemistry properties in the 2 studies. In this study the range of TOC was 1.0 to 5.3 mg/L, whereas the range of DOC in EU (2009) was 2.5 to 12.0 mg/L. In general, organic pollution in Japanese rivers has dramatically decreased since the 1970s as a result of improved sewage treatment; for example at site A, the 75th percentile of BOD was approximately 30 mg/L in 1988 and approximately 5 mg/L in 2006 (Edogawa River Office 2006). The range of TOC in this study (1–5.25 mg/L) reasonably captures the current range of TOC in Tokyo rivers, although further monitoring is certainly needed. Ignorance of differences in water hardness would be another reason that this risk assessment showed only small effects of BLM normalization in the case of Ni. The effect of water hardness may be more important for Ni than Cu (Supplemental Data).

One of the difficulties in adopting BLMs for risk assessment in Japan is the lack of an assessment and management framework that would allow different environmental standards to be set based on the water chemistries of the sites (Central Environmental Council Environmental Water Group 2003). Given this situation, the adoption of conservative estimates that err on the side of safety might be a realistic compromise in the risk assessment of heavy metals in Japan. In general, the risk assessment in this study showed that TOC and HC5 were positively correlated, and HC5s without BLM normalization were relatively close to the HC5 at the site with the lowest TOC (Table 2). These results suggest that risk assessment without BLM correction does not cause a large problem if overestimation of risk is acceptable. In general, however, too much overestimation of risk tends to consume human and monetary resources ineffectively without the expected benefit to organisms. It is thus still important in the management process to appreciate the possible degree of overestimation of risk by considering BLM normalizations, even if only conservative estimates of risk are used to set environmental standards.

The question of whether the calculations in the BLMs and the assumptions about metal speciation are appropriate for Japanese rivers and organisms has already been raised. The stakeholders in Japan may not accept BLM normalizations in the absence of validation of the models in Japan. For example, metal speciation may be different in Japanese water because of potential differences in the composition of organic matter and some interacting substances (Nagai 2011). To encourage the consideration of bioavailability in risk assessments, and the use of BLMs in Japan, there is a need for further basic studies of issues relative to BLM normalizations (e.g., strategic monitoring of DOC, pH, and water hardness, and examination of metal speciation in Japanese river waters) and ecotoxicity tests using Japanese river waters. It would also be desirable to develop a BLM based on the important species in Japanese ecosystems such as Medaka (Oryzias latipes) (Kamo and Hayashi 2011). The work presented here is the first study focused on the effects of BLM normalization on risk assessment in Japanese rivers and will hopefully promote and guide proper consideration of bioavailability in ecological risk management of heavy metals in Japan in the near future.

Acknowledgements

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. DATA AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. SUPPLEMENTAL DATA
  8. Acknowledgements
  9. REFERENCES
  10. Supporting Information

This work was supported by the Steel Industry Foundation for the Advancement of Environmental Protection Technology. I thank 3 anonymous reviewers for very helpful comments on the manuscript. I also thank M. Kamo and W. Naito for providing very helpful information about the correction of ecotoxicity data with biotic ligand models.

REFERENCES

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. DATA AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. SUPPLEMENTAL DATA
  8. Acknowledgements
  9. REFERENCES
  10. Supporting Information
  • Campbell PGC. 1995. Interactions between trace metals and organisms: critique of the free-ion activity model.: In: Tessier A, Turner D, editors. Metal speciation and bioavailability in aquatic systems. Chichester (UK): Wiley. p. 45102.
  • Central Environmental Council Environmental Water Group. 2003. Setting of environmental quality standards for the conservation of aquatic organisms. Tokyo, Japan: Environmental Agency.
  • Central Environmental Council Environmental Water Group. 2009. 19th report 6. Tokyo, Japan: Environmental Agency.
  • Di Toro DM, Allen HE, Bergman HL, Meyer JS, Paquin PR, Santore RC. 2001. Biotic ligand model of the acute toxicity of metals. 1. Technical basis. Environ Toxicol Chem 20:23832396.
  • Edogawa River Office. Temporal change in water quality in monitoring sites [Internet]. [cited 2011 August 4]. Available from: http://www.ktr.mlit.go.jp/ktr_content/content/000015349.pdf
  • [EU] European Union. 2008. European Union risk assessment report. Voluntary risk assessment of Cu, copper II sulphate pentahydrate, copper(I)oxide, copper(II)oxide, dicopper chloride trihydroxide. Available from: http://echa.europa.eu/copper-voluntary-risk-assessment-reports/
  • [EU] European Union. 2009. European Union risk assessment report. Nickel and nickel compounds. Available from: http://esis.jrc.ec.europa.eu/doc/risk_assessment/REPORT/nickelreport311.pdf
  • Gelman A. 2004. Bayesian data analysis. Boca Raton (FL): Chapman & Hall/CRC. p. 696.
  • Hayashi TI, Kashiwagi N. 2011. A Bayesian approach to probabilistic ecological risk assessment: risk comparison of nine toxic substances in Tokyo surface waters. Environ Sci Pollut Res 18:365375.
  • Kamo M, Hayashi TI. 2011. Biotic ligand model: historical overview and perspectives. Jpn J Environ Toxicol 14:2538. Japanese.
  • Lunn DJ, Thomas A, Best N, Spigelhalter D. 2000. WinBUGS—A Bayesian modeling framework: Concepts, structure, and extensibility. Stat Comput 10:325337.
  • Nagai T. 2011. Metal speciation and bioavailability in natural aquatic environment. Jpn J Environ Toxicol 14:1323. Japanese.
  • Niyogi S, Wood CM. 2004. Biotic ligand model, a flexible tool for developing site-specific water quality guidelines for metals. Environ Sci Technol 38:61776192.
  • Posthuma L, Suter GW, Traas TP. 2002. Species sensitivity distributions in ecotoxicology. Boca Raton (FL): Lewis Publishers. p. 616.
  • R Development Core Team. 2007. R: Language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing.
  • van Sprang PA, Verdonck FAM, van Assche F, Regoli L, De Schamphelaere KAC. 2009. Environmental risk assessment of zinc in European freshwaters: A critical appraisal. Sci Total Environ 407:53735391.

Supporting Information

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. DATA AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. SUPPLEMENTAL DATA
  8. Acknowledgements
  9. REFERENCES
  10. Supporting Information

Additional Supporting Information may be found in the online version of this article.

FilenameFormatSizeDescription
ieam_1326_sm_SupplMaterial.doc2362KSupplementary Material

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.