API Publ 4751:2005 pdf download
API Publ 4751:2005 pdf download.Evaluation of Water Quality Translators for Mercury.
3.2 EPA PROPOSED TRANSLATION METHODS
The draft EPA guidance rccommcnds the following apprnachcs to critena translations, in order of preference: (I)
derive site-specific BAFs. (2) use a bioaccumulation model, or (3) use EPA’s national default translators tEPA
200-ta). Site-specific methodologies (i.e., EPA’s preferred approach I are discussed further in Section 3.3.
Bioaccuniulation models and default translators are addressed below
3.2.1 Bioaccumulation Models for Mercury
EPA 2004a) does not recommend any specific hioaccumulatioa models for mercury, although a modeling approach is ostensibly preferred over the use of default translators. Despite extensive study. there are no accurate, nationally applicable models for predicting mercury hioaccurnulation. Models developed for hydrophobic organic chemicals are not directly useful for this purpose. although conceptual modifications to address speciating chemicals such as mcrcuz-y hare been suggested (Ttxixc and Mackay 2004k. As noted by EPA (2004a. even the Mercury Cycling Model (Tetra Tech 19991 does not predict bioaccumulation. but rather uses BAF as an input. Along the same lines. Hope (2003) developed an elaborate, probabilistic food web model for mercury in conjunction with TMDL development for the Willarnetie River in Oregon. However, critical components such as the ratio of methyl- to total mercury in surface water and biuaccuniulation factors at the base of the food web were based on site-specific empirical data. As a result, this type of model is not transferable to other waterways without extensive data collection. Furthermore, the model assumes that site-specific fd and RAE levels are independent of total mercury loading, which may not be valid (Marvin-DiPa.squalc ci al. 2000: .Schaefer et al. 2004): thus without further validation over time, this model cannot substitute for continued direct sampling of fish tissue.
As an alternative to process-based or mechanistic mathematical models, the draft EPA guidance suggests that regression models incorporating such variables as pH. DOC. and tish age may be acceptable for criteria translation purposes. According to EPA’s Methodology for Deriving Ambient Water Quality Criteria for the Protection of Human Health (EPA 2()0c. such models should predict hioaccumulation based on the most biologically relevant chemical form (e.g.. dissolsed methylmercury t. while also allowing for translation between the bioasaibble form and the total concentration in water. Recent efforts to develop nationally applicable regression models have not produced highly accurate results, although they are probably more accurate than national default BAFs. Brumbnugh ci al. (2001) used data from the NAWQA program to develop a model predicting total mercury in fish (normalized to fish length) based on methylmercury concentrations in water tuntiltered. The most useful parameters in reducing variability in the model were p11. percent wetlands in the watershed, and acid volatile sulfide (AVS) in sediment. However, the model explained only 45% of the variation in fish tissue concentrations. Ongoing studies through NAWQA and other programs (e.g.. Brigham ci al. 20031 may eventually lead to improvements in the ability to model mercury hioaccumulation in stream ecosystems.
Regression models developed using regional data sets from lake sampling programs have provided greater accuracy than the NAWQA example cited above. Several examples are discussed below.
• (‘onadian laAes: Moore et al. (2003) were able to explain 84)% of the variability in mercury hioaccumulation aniong 38 Canadian lakes based on pH. DOC. and mcthylmercury in water. however, when this model was applied to a subset of the NAWQA stream data, only 32% of the variation was explained. The limited applicability of the Moore ci al. 20031 model to the NAWQA data is likely due to differences between lake and stream systems as well as regional geological differences. Ii is clear that mercury methylation is more variable over time in streams as compared to lakes. In an evaluation of within.site variability. Kelly et al. (1995) found significant correlations between total and methylmercury in water only for lakes and not for streams.
• Swedish lakes: Hakanson (2000) assessed mercury concentrations in fish collected from 39 Swedish lakes, His regression model explained 85% of the observed variation based on the following variables: lake inocphomeiry (?areaidcpth). mercury in sediment, pH. phosphorus, and lake depth. Hakanson (2000) proposes that the model can assist in the development o lake remediation strategies but cautions that it should not be applied to reservoirs or to lakes in different climatological zones.
• New Hampshire and Verninni lake.s: Another example of regional modeling for lake systems is a model developed by Kamnian et al. (2004) for lakes in New Hampshire and Vermont. Intereslingly. this model did not include any measure of mercury concentrations in water. Instead, data on acid neutralizing capacity, DOC, pH, conductivity, and lake flushing rate were used to generate binary predictions as to whether or not yellow perch.
