The organochlorine pesticide dieldrin is still frequently detected in soil, sediment, biota, and air in the United States (Nowell and others, 1999; Schmitt and others, 1999; Wong and others, 2000), even though its uses were discontinued 15–30 years ago. Dieldrin residues in the environment may originate from past application of either aldrin, which degrades readily in the environment to dieldrin, or dieldrin itself. Aldrin and dieldrin (like DDT) are both organochlorine pesticides that were once used extensively, but were discontinued because of their persistence, tendency to bioaccumulate, carcinogenicity, and hazard to wildlife. Aldrin and dieldrin were applied in agriculture in the United States until the early 1970s, mainly on corn, but also on fruits and nuts, other vegetables, tobacco, and cotton. They also were used in nonagricultural applications—especially in termite control, which continued until at least the late 1980s. Dieldrin, like many organochlorine pesticides, is hydrophobic (low in water solubility, with a tendency to sorb to soil and sediment and to partition to organic matter) and resistant to degradation in the environment.

Three national studies have monitored dieldrin and other organochlorine pesticides in whole freshwater fish from rivers and streams across the United States. These are the Fish and Wildlife Service’s (FWS) National Contaminant Biomonitoring Program (NCBP, 1969–1986), the U.S. Environmental Protection Agency’s (EPA) National Study of Chemical Residues in Fish (1986–1987), and the U.S. Geological Survey’s (USGS) National Water-Quality Assessment (NAWQA) Program (1992–2001). Evidence from these programs indicates that, nationally, organochlorine pesticide levels have declined since the use of these compounds was discontinued (Schmitt and others, 1999; Fuhrer and others, 1999), with the biggest declines occurring during the 1970s. Dieldrin residues in fish were highly variable during the 1970s, but were markedly lower after 1976 than before 1976; residues measured by the NCBP Program continued to decline during the 1980s (Schmitt and others, 1999; Nowell and others, 1999). NAWQA sampling data during 1992–2001 indicates the continued presence of dieldrin at low levels in fish and bed sediment from many streams in the United States (Wong and others, 2000; Nowell and Crawford, 2003)—dieldrin was detected in whole fish at over 50 percent of agricultural and urban streams, and in over 40 percent of streams in mixed land use areas (Nowell and Crawford, 2003). This suggests that dieldrin-contaminated soil is still being carried into receiving waters from both agricultural and urban land use settings.

Organochlorine pesticides have been shown to adversely affect the survival of various organisms (including aquatic invertebrates, fish, birds, and mammals) in laboratory tests; to disrupt the reproduction of fish and birds in the field; and to accumulate to high levels in fish-eating mammals (U.S. Environmental Protection Agency, 1975; Nowell and others, 1999). In addition, organochlorine pesticides (including dieldrin) induce the monooxygenase enzymes that hydroxylate testosterone (Haake and others, 1987), and they have been associated with endocrine and reproductive changes and immunosuppressive effects (Amdur and others, 1991). The biological significance of the low-level organochlorine residues detected in the environment today is more difficult to assess. Although the linkage between body burdens and effects in fish is not well understood, body burdens provide at least a direct indicator of exposure (Black and others, 2000). Organochlorine pesticide contamination in the field also has been associated with fish kills (Hunt and Linn, 1970; Madhun and Freed, 1990) and fish diseases (Myers and others, 1993). A few organochlorine pesticides (especially DDT and chlordane, and to a lesser extent dieldrin) are the subject of fish consumption advisories issued by the states for protection of human health from eating fish contaminated with bioaccumulative pollutants (U.S. Environmental Protection Agency, 2005).

Given the continued detection of low-level residues of organochlorine pesticides in fish, plus the uncertainty as to their potential effects on humans and wildlife from fish consumption, there is a need to ascertain the geographic extent and the degree of present-day exposure of aquatic populations to organochlorine pesticides. However, because of the high cost of monitoring and analysis, combined with decreasing interest in organochlorine pesticides as residues decline nationally, it is unlikely that substantial investment in monitoring will be made, except in areas where there is a strong probability of concern. An alternative to widespread direct monitoring is to predict concentrations using empirical regression models that relate measured concentrations in fish to watershed characteristics and estimates of past use.

A few prior studies have investigated the relationship between large-scale land use practices and the presence of organochlorine pesticides in sediment or biota (Truhlar and Reed, 1976; Munn and Gruber, 1997; Black and others, 2000). Black and others (2000) used logistic regression to develop predictive models for estimating the probability of detecting specific organochlorine pesticides in fish (sculpins) from streams in the Willamette and Puget Sound Basins of the Pacific Northwest. This regional-scale study predicted presence or absence, rather than specific concentration estimates. A number of studies have used linear regression or logistic regression models to predict concentrations of dissolved pesticides in stream water (Krueger and Tornqvist, 1996; Battaglin and Goolsby, 1997, 1998; Larson and Gilliom, 2001; Larson and others, 2004). Predictor variables in these models included pesticide use in the watershed, physical and chemical properties of the pesticides, and (or) various watershed characteristics. We are unaware of any studies that have developed models relating organochlorine pesticide concentrations in fish to land use and other watershed characteristics on a national scale.

Assessment of risk to humans or wildlife from pesticides in the environment requires evaluation of both exposure and effects. The purpose of this report is to investigate the suitability and effectiveness of a regression model for predicting, and explaining variability in, dieldrin concentrations in whole fish from United States streams on a national scale. Specific objectives are to:

• Develop a regression model to estimate dieldrin concentrations in whole fish on the basis of national data on watershed characteristics, including estimates of historical pesticide use;

• Assess how much variability in pesticide concentrations in whole fish can be explained by factors in the regression model;

• Predict dieldrin concentrations in whole fish in streams across the United States, as a function of lipid (fat) content, from the regression model on the basis of watershed characteristics and estimates of historical pesticide use.

This model was developed using NAWQA data on dieldrin in whole fish at 648 stream sites sampled during 1992–2001. These stream sites are located in most parts of the country and represent a variety of land uses. Ultimately, the method described in this report will be applied to other organochlorine pesticides detected in fish by NAWQA.

In this report, the following terms are used: Concentrations measured at NAWQA sites are “observed concentrations.” Application of the regression model using a given set of explanatory variables produces an “estimate” or “estimated concentration” for a site (in contrast to an observed, or measured, concentration). When the model is applied to an unmonitored site, the result is a “prediction” or “predicted concentration.” Finally, when predictions are made for unmonitored sites over a wide geographic area, this builds a “spatial extrapolation.”

The NAWQA data used in this study are the product of work by many people at the USGS who contributed to the study design, data collection, sample processing and analysis, and geographic information system (GIS) data analysis in the NAWQA Program. The authors are indebted to them all—especially to NAWQA study unit personnel, who developed the environmental framework and planned and conducted sampling for their study units, and provided basin-specific expertise for national land use classification and data analysis. The authors thank Larry R. Brown and David K. Mueller of the U.S. Geological Survey for their very helpful technical reviews of this report. The survreg procedure for the S language written by Terry M. Therneau of the Mayo Clinic was used to fit the regression model developed in this study. The authors are grateful to Michael Haverty (U.S. Dept. of Agriculture, Forest Service), Nan-Yao Su (University of Florida), and W. David Woodson (U.S. Geological Survey) for their assistance in obtaining termite distribution data and maps.