Environ. Sci. Technol. 2002, 36, 4593-4599
A Regional Atmospheric Fate and Transport Model for Atrazine. 2. Evaluation E L L E N J . C O O T E R , * ,† W I L L I A M T . HUTZELL,‡ WILLIAM T. FOREMAN,§ AND MICHAEL S. MAJEWSKI# NOAA, Atmospheric Research Laboratory, U.S. Environmental Protection Agency, National Exposure Research Laboratory, MD-E243-03, Research Triangle Park, North Caroline 27711, U.S. Geological Survey, National Water Quality Laboratory, Denver, Colorado 80225-0046, and U.S. Geological Survey, Sacramento, California 95819-6129
detected in soil, air, groundwater, and surface water adjacent to agricultural fields as well as at remote locations (4-11). Preliminary results for the atrazine version of CMAQ are consistent with observations and scientific theory regarding atrazine’s atmospheric fate (1). Models such as CMAQ facilitate the formulation of highly complex scientific hypotheses concerning natural processes that can be confirmed through comparison with observation but never truly validated. With this in mind, a preliminary diagnostic evaluation of the atrazine version of CMAQ is presented to gauge the strength and depth of our current understanding of natural processes. Model predictions of atrazine concentration in rainfall and air are compared to observations from two recent field studies, and the most likely sources of prediction uncertainty are discussed.
Methodology The Community Multiscale Air Quality (CMAQ) modeling system has been adapted to simulate the regional fate and transport of atrazine. Model modifications and simulations spanning April to mid-July 1995 are described in a previous paper. CMAQ results for atrazine concentrations in air and rainfall are evaluated against field observations taken along the Mississippi River and the shores of Lake Michigan in 1995. CMAQ results agree within 10% of published annual wet deposition load estimates for Lake Michigan and predicted annual dry deposition lies within published error bounds. Comparisons of weekly observed and predicted air and rainfall concentrations along the Mississippi River yield order-of-magnitude differences. Precipitation weighting of concentrations in rainfall good agreement for seasonal time frames. Weekly ambient gas form concentrations tend to be overpredicted by the CMAQ and semivolatile particulate fractions are underpredicted. Uncertainty in CMAQ predictions of air and rainfall concentrations for atrazine appear to derive primarily from uncertainty in emissions estimates, simulated precipitation, and spatial scale.
Introduction The U.S. Environmental Protection Agency (USEPA) has developed the Community Multiscale Air Quality (CMAQ) Eulerian modeling system to investigate scientific and regulatory questions related to acid deposition, tropospheric ozone, and particulate material over scales ranging from urban to intracontinental (2). Expansion of CMAQ to address a wider range of pollutants has been limited by the absence of physicochemical process detail necessary to simulate semivolatiles in the atmosphere. A previous paper (1) describes modifications required to include these processes and summarizes their implementation for atrazine. Atrazine is a relatively nonreactive and moderately long-lived semivolatile herbicide that has seen widespread use in the United States and Canada for more than 30 years (3). It is commonly * Corresponding author phone: (919)541-1334; fax (919)541-1379; e-mail:
[email protected]. On assignment to the U.S. Environmental Protection Agency, National Exposure Research Laboratory, MD-E243-01, Research Triangle Park, NC 27711. † NOAA, Atmospheric Research Laboratory. ‡ U.S. Environmental Protection Agency. § U.S. Geological Survey, Denver, CO. # U.S. Geological Survey, Sacramento, CA. 10.1021/es011372q Not subject to U.S. Copyright. Publ. 2002 Am. Chem. Soc. Published on Web 10/08/2002
The CMAQ vertical domain for this evaluation consists of 21 layers in the atmosphere. Most of these layers are concentrated near the earth’s surface, with eight layers assigned to the lowest ∼500 m of the atmosphere. The top of the model is bounded by the 100 mb pressure height, which typically ranges from 10 km to more than 15 km above the surface of North America and varies by season (higher during summer months) and geographic location (decreases from South to North). The horizontal domain consists of 5175 (75 × 69) 36 km × 36 km rectangular grid cells spanning the United States and southern Canada from the Rocky Mountains to the Atlantic ocean (1). Hourly atrazine volatilization from agricultural soils is provided by the Pesticide Emissions Model (PEM) (1, 12), and meteorological inputs to both CMAQ and PEM are provided by the Fifth Generation Pennsylvania State University/National Center for Atmospheric Research (NCAR) Mesoscale Meteorological model, coupled to a Land-Surface model (MM5-PX) (13). CMAQ simulations are performed from April 1 through July 18, 1995. Model output is in the form of hourly, grid-averaged chemical concentrations in the air and deposition fluxes from the lowest model atmosphere layer. For the present analysis, hourly model output is either averaged or summed over the appropriate observation time frame, e.g., 7 days. The midpoint values of the four grid cells surrounding the observation location are then bilinearly interpolated (14) to the sample site. Between-cell variability (interpolation error) is estimated as one-half the difference between maximum and minimum values of the four cells and does not explicitly include topographical and meteorological processes that may be poorly resolved at the 36 km scale. Two recent field studies were selected for comparison to model predictions: (1) the USEPA Lake Michigan Mass Balance (LMMB) Study and (2) the U.S. Geological Survey (USGS) Toxic Substances Hydrology (Toxics) Program, Agricultural Chemicals in the Midwest Project. Selection was based on completeness of observation record, temporal and spatial extent, documentation, and quality-control and quality-assurance criteria. Method Detection Limit (MDL) estimates were provided for the LMMB (15) and USGS (9, 16) rain methods. The USGS method assumes a 1-L sample size. An estimated reporting level, assumed in this paper to be equivalent to the MDL, was provided for the USGS air samples (17). The MDL describes the lowest level at which an analyte can be differentiated reliably from analytical method noise (18). The Reliable Quantification Level (RQL) describes a level above which reliable quantitative comparisons can be made and is usually assigned a value 9.5 times the MDL (18). VOL. 36, NO. 21, 2002 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
4593
FIGURE 1. Sample locations for observation sets (adapted from ref 9).
Lake Michigan The Great Lakes National Program Office (GLNPO) of the USEPA initiated the LMMB Study (19). Field sampling began in April, 1994 and continued through the end of October, 1995. Rainfall samples were collected at 10 sites: 8 along the Lake Michigan shoreline, 1 at Brule River, WI, and 1 at Bondville, IL (Figure 1). Rainfall was accumulated over a 4-week period at each site. Each cumulative sample was then analyzed for atrazine and used to represent a monthly averaged concentration at each site. Atrazine concentrations in the air were sampled under the LMMB Study but yielded insufficient observations above the MDL to support further quantitative analysis (15). Sampling equipment and procedures are described in greater detail in Miller et al. (10). Wet Deposition. Predicted and observed atrazine concentrations in rainfall for LMMB Study sites during April, May, and June 1995 are shown in Figure 2. The mean MDL, i.e., the average of individual sample MDLs adjusted for each sample’s collection volume, is 9.0 ng L-1 for Sleeping Bear 4594
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 36, NO. 21, 2002
Dunes and 3.0 ng L-1 at the remaining nine sites (15). This result yields RQL concentrations of 85.0 ng L-1 for Sleeping Bear Dunes and 28.5 ng L-1 for the remaining sites. A solid circle indicates an observation is less than the mean RQL but greater than or equal to the MDL. The observation month (April, May, or June) is shown for observations greater than the RQL. The horizontal bars reflect error associated with analytical activities and sample matrix effects (see Supporting Information for calculation). The vertical bars reflect the CMAQ spatial interpolation error discussed previously. Median observed and predicted concentrations for observations greater than the RQL are 140 ng L-1 and 119 ng L-1, respectively. Observed concentrations greater than the RQL are generally underpredicted during April and overpredicted for samples collected during May and June. Poorest April and May predictions are for samples collected at Beaver Island and Indiana Dunes (see Figure 2). Later discussion (see Prediction Uncertainty for Atrazine, atrazine emissions) suggests that uncertainty with regards to the timing of
FIGURE 2. Predicted and observed monthly atrazine concentrations in rainfall at Lake Michigan Mass Balance Study sites, April-June 1995. Horizontal and vertical bars represent measurement (analytical and matrix) and model interpolation error, respectively. Solid circles represent observations less than the Reliable Quantification Level (RQL). Letters denote the month of observation for concentrations greater than the RQL. chemical application is one factor that may be associated with these differences. Annual Lake-Wide Deposition. Annual atrazine load estimates produced by CMAQ (approximated by April-July, 1995 deposition) and others (10, 20, 21) are listed in Table 1. Miller et al. (10) estimate 1994-1995 annual atrazine load to Lake Michigan in rainfall to be 1.04 × 103 kg yr-1 and dry deposition of particle-associated atrazine to be between 0.23 and 1.00 × 103 kg yr-1. This dry deposition range reflects uncertainty based on near shore and over-lake average particle concentration of 0.2 ng m-3 (( 0.3 ng m-3) for the spring months (April through June) (10). Schottler and Eisenreich (20) estimate an average 1991-1994 atmospheric atrazine load to Lake Michigan in rainfall of 2.60 × 103 kg yr-1 and particulate deposition of 0.16 × 103 kg yr-1. A 2-year, 1990-1991 average atrazine load in rainfall to Lake Michigan of 3.01 × 103 kg yr-1 is estimated from Goolsby et al. (21). The CMAQ prediction and Miller et al. (10) agree (within 10%) that 1995 annual atrazine wet deposition is less than either the Schottler and Eisenreich (20) or Goolsby et al. (21) estimated averages and that the contribution of dry deposition is substantially greater than the Schottler and Eisenreich average, i.e., 27% (1995 CMAQ) and 18-50% (10) vs 6% (20).
Mississippi River Valley Weekly composite samples of atrazine in air and rainfall were collected at three paired urban (Jackson, MS, Iowa City, IA, and Minneapolis, MN) and rural (Rolling Fork, MS, Cedar Rapids, IA, and Princeton, MN) locations along the Mississippi River and a rural background site at Eagle Harbor on Lake Superior (Figure 1) from early April through mid-September 1995 (9, 17). Ambient air-sampling equipment and procedures are discussed in Foreman et al. (17). An estimated reporting level, assumed in this analysis to be equivalent to the MDL, of 6 pg m-3 in an 850-m3 air volume was provided for the USGS air samples (17). These reporting levels were estimated from water (rain) method MDLs, adjusting for average air volume and method recovery. Each sample of atrazine in rainfall is the sum of precipitation events occurring during a 1-week period. Majewski et al. (9) report an MDL of 1 ng L-1. Atrazine recoveries ranged from 88 to 100% in stability studies conducted using spiked rainwater stored from 5 days (9) to 3 weeks (22) at ambient temperature.
These recoveries are in the range of expected performance for the rainwater method and indicate excellent atrazine stability under storage conditions more extreme than the refrigerated field storage conditions employed by Majewski et al. (9). 7-Day Air Concentrations. Predicted and observed concentrations of particulate atrazine in the air are shown in Figure 3 and summarized in Table 2. In Figure 3, the dashed vertical lines mark the estimated analytical MDL and RQL. Detections reported as less than the MDL have been assigned ∼MDL value. Complete agreement between observations and predictions is represented by the 1:1 line. Model spatial interpolation error for each sample (not shown) ranges from 10 to 120% of the model estimate. Ninety-five percent analytical and laboratory matrix error bounds for particulate samples are estimated to be ( (sample value × 0.25) (see Supporting Information). Model predictions associated with sample concentrations less than the estimated MDL are generally biased high, with larger predicted values most often associated with samples from urban locations. Median observed particulate concentrations greater than the RQL are underpredicted by a factor of 1.40 (Table 2). The range of predicted particulate form concentrations is a factor of 10 less than the observations. Strength of association is tested using a Spearman rank correlation statistic, a robust and resistant measure of correlation (23). A value of 0.63 (Table 2) indicates that individual predicted and observed particulate concentrations are significantly correlated (n ) 39, R < 0.005 (24)). A similar comparison for observed and predicted gas form concentrations is shown in Figure 4 and summarized in Table 2. Model spatial interpolation errors (not shown) range from 10 to 80% of the model estimates. Ninety-five percent analytical and laboratory matrix error bounds for gas form samples are estimated to be ( (sample value × 0.25) (see Supporting Information). The CMAQ generally overpredicts gas form concentrations for observations less than the MDL. Most of these values are associated with observations from urban locations. Table 2 indicates that the CMAQ model overpredicts the observed median concentration by a factor of 2 and underpredicts the range of gas form concentrations greater than the RQL by nearly 50%, but a Spearman value of 0.60 suggests predicted and observed concentrations are correlated (n ) 22, R < 0.005 (24)). The availability of both gas and particulate atrazine samples facilitates the preliminary evaluation of the CMAQ partitioning algorithm (1). Results of such a comparison should be viewed with caution in the absence of field measurement error estimates. Figure 5 shows observed and predicted particulate fraction (the ratio of particulate concentration to total, i.e., gas plus particulate concentration) for observations in which both gas and particulate concentrations are greater than the RQL (n ) 22). With the exception of Iowa City (red symbols), only samples from rural locations meet this criteria. The mean and standard deviation of predicted values (0.19, 0.11) are one-half those of observations (0.40, 0.22). Predicted fractions are less than observed, with the exception of Rolling Fork, MS 7-day samples ending May 16th and 23rd and the Cedar Rapids, IA sample ending July 10. 7-Day Wet Deposition. Predicted and observed atrazine concentrations in rainfall are shown in Figure 6 and summarized in Table 2. Model spatial interpolation error ranges from 10% to 200% of the model estimate. Ninety-five precipitation analytical and laboratory matrix error bounds for particulate samples are estimated to be ( (sample value × 0.20) (see Supporting Information). The observed median and range of concentrations are underpredicted by CMAQ (Table 2), but the Spearman statistic (0.54) indicates observed and predicted values are correlated (n ) 67, R < 0.005 (24)). There is no apparent rural/urban prediction bias. VOL. 36, NO. 21, 2002 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
4595
TABLE 1. Annual Lake Michigan Atmospheric Load Estimates for Atrazined (Dry Deposition Is the Sum of Gas and Particulate Forms) community multiscale air quality model (CMAQ) (April 1-July 18, 1995)
July 1994-July 1995a
1991-1994 average annualb
1990-1991 average annualc
1.14 × 103 0.43 × 103 1.57 × 103
1.04 × 103 0.23 to 1.00 × 103 1.27 to 2.04 × 103
2.60 × 103 0.16 × 103 2.76 × 103
3.01 × 103 N/Ae N/Ae
wet deposition (kg) dry deposition (kg) total (kg)
a Miller et al. (10) estimate that 98% of annual wet deposition occurs from April through July, 1995. Dry deposition estimate range is based on average over-lake particle-bound concentration ( one standard deviation and single, fixed particle size and settling velocity. b Reference 20. c Reference 21. Samples were collected from March through December, 1990 and from January through September, 1991. d Note: surface area of Lake Michigan ∼ 57800 km2. e N/A ) information not available.
FIGURE 3. Predicted and observed 7-day particulate form atrazine concentrations in the air for U.S. Geological Survey sites AprilJuly 1995 (MDL ) Method Detection Limit; RQL ) Reliable Quantification Level).
FIGURE 4. Predicted and observed 7-day gas form atrazine concentrations in the air at U.S. Geological Survey sites AprilJuly 1995 (MDL ) Method Detection Limit; RQL ) Reliable Quantification Level).
TABLE 2. Community Multiscale Air Quality (CMAQ) Predicted and U.S. Geological Survey Observations for Atrazine in the Mississippi River Valley and Eagle Harbor, MI, April through July 1995a concn in the air concn in gas particulate rainfall rainfallb -3 -3 (ng m ) (ng m ) (µg L-1) (cm) minimum observed 25th percentile median 75th percentile maximum
0.08 0.18 0.53 1.17 23.00
0.06 0.17 0.29 0.50 22.80
0.01 0.04 0.10 0.20 2.04
0.16 1.40 2.42 4.93 20.16
minimum predicted 25th percentile median 75th percentile maximum
0.03 0.44 0.99 2.32 10.30
0.00 0.07 0.17 0.65 2.27
0.00 0.01 0.04 0.20 1.04
0.01 0.99 1.85 3.20 14.90
0.60
0.63
0.54
0.32
Spearman rank correlation
a All statistics are for observations greater than the RQL. Air concentrations are weekly averages. Concentration in rainfall represents a weekly integrated sample. b Includes all precipitation samples.
Prediction Uncertainty for Atrazine Uncertainty in model predictions stem from computational precision, incorrect or incomplete parameter estimates or process representation, and model inputs. Computational uncertainty in the CMAQ is discussed elsewhere (25-27). Uncertainty associated with the CMAQ modifications and assumptions for semivolatiles are discussed in Cooter and Hutzell (1). The present discussion focuses on prediction 4596
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 36, NO. 21, 2002
FIGURE 5. Predicted and observed particulate fraction [particulate concentration/(particulate + gas concentration)] for cases in which both particulate and gas form observed concentrations are greater than the RQL. Green symbols are computed when June NH3 emissions are increased by 25%. Red symbols indicate urban sites (Iowa City). One hidden value (6/20, predicted). uncertainty arising from model input error and uncertainty in gas to particulate matter partitioning, chemical transformation, emissions, and precipitation. Gas to Particulate Partitioning. The sensitivity of the partitioning algorithm to selected chemical properties is
FIGURE 6. Predicted and observed atrazine concentrations in rainwater at U.S. Geological Survey sites April through July, 1995 (MDL ) Method Detection Limit; RQL ) Reliable Quantification Level). presented in ref 1. Chemical parameters for atrazine are reported to have uncertainties of up to 50% (28). Significant partitioning changes result (1) when the subcooled liquid vapor pressure, Po (1.75 × 10-3 Pa), is increased or decreased by a factor of 5, or the difference between enthalpies of desorption and vaporization, ∆E (6.28 × 103 J mol-1), is reduced to zero or increased by a factor of 3. No substantial response to smaller changes, more representative of parameter uncertainty reported for atrazine (28), were noted. Wet scavenging of gas form atrazine depends on the Henry’s Law constant, water content of the cloud, and precipitation rate (29). The Henry’s Law constant, Ho, was increased by 20% to test predicted wet scavenging response to uncertainty in this parameter. No significant change in wet removal of atrazine was noted, indicating maximum wet scavenging efficiency is already achieved with the original value of H0 (2.48 × 10-4 Pa m3 s-1). Chemical Fate. Transformation, i.e., degradation of gas form atrazine in the air is modeled as the reaction of atrazine with OH radicals in the atmosphere (1, 30). There is no firm evidence suggesting temperature dependency, and structural analysis (31, 32) provides a temperature independent rate constant of 2.73 × 10-11 cm3 molecule-1 s-1. Although other estimates (28, 30) suggest this rate constant may vary by more than a factor of 5, a clear bias cannot be inferred from the comparison results presented here. The loss of atrazine in surface runoff is estimated in the PEM as a fixed percentage (0.5%) of the chemical applied to grid cells throughout the model domain (1, 30). Capel and Larson (33) report a median atrazine load to streams as a percentage of chemical use (LAPU) for watersheds of size 102-105 ha of 0.47%. Larger values of LAPU are expected for extreme (heavy) rainfall immediately following application. Sample calculations (see Supporting Information) suggest that underestimation of surface runoff loss could result in slight overestimation of subsequent atrazine emissions and concentrations in rainfall but is not a significant factor contributing to prediction errors in air and rainfall. Atrazine Emissions. The time dependent volatilization of atrazine from agricultural soils estimated by the PEM (12) is assumed to represent the majority of regional-scale atrazine emissions (1). The PEM uses physically based algorithms to simulate the behavior of chemicals in agricultural soils and has been validated against observations for bare field conditions (12). Uncertainty in these estimates is related primarily to PEM model inputs, including chemical proper-
ties, county-level chemical use, application timing, soil texture and organic content, farm-management practices, and meteorology. County-level chemical use estimates are available from the U.S. Geological Survey (34). About 10% error is associated with these estimates (33), which are based on annual state-level pesticide-use estimates representing general county-wide patterns of use over a 4-year period on 1992 Census of Agriculture county crop acreage (35). Atrazine use during 1995 in Southern Canada was supplied by Y. F. Li of Environment Canada (personnel communication). No error estimates were provided with these data. The timing of agrichemical application varies with crop, local soil texture, and weather conditions that affect trafficability, soil moisture, and soil temperature. It is assumed that atrazine is applied at or near to crop planting (36). State level crop planting dates are abstracted from Weekly Crop Progress Reports for 1995 (37) and are provided to the PEM (see Supporting Information (1)). If, for instance, chemical application dates provided to the PEM are later in the spring than actually occurred, then atrazine available for wet removal by rainfall could be underestimated early in the spring and overestimated later in the season. For the LMMB study region, chemical application was assumed to begin on April 29 in Wisconsin, Illinois, Indiana, and Ontario, Canada (38) and May 6 in Michigan. Crop progress reports for 1995 indicate only 7%, 4%, and 1% of all corn acres were planted during April in Illinois, Indiana, and Michigan, respectively, and no agricultural activity is reported during April in Wisconsin. This is as compared to 44%, 55%, 69%, and 72% for Illinois, Indiana, Michigan, and Wisconsin, respectively, during May, 1995. Canadian dates reflect climatology only because no year-specific crop progress information is available to the public. Additional study is needed to determine if the omission of April planting activity in Illinois and Indiana and the lack of year-specific information for Canada are sufficient to account for the concentration prediction errors highlighted in Figure 2. Resuspension (emission) of particulate form atrazine via wind erosion is not included in the present analysis but could have implications for the underprediction of particulate form atrazine. Atrazine is often applied as a wetable powder that is very susceptible to wind erosion. Glotfelty et al. (6) indicates that wind erosion is not a significant loss process relative to volatilization, but their work was performed exclusively in the Chesapeake Bay region and may not reflect conditions representative of other geographic and climatic regimes. This emission process should be explored further in future studies. Other Emissions. The amount and composition of particulate matter with diameter less than 10 µm (PM10) determines the partitioning of atrazine between gas and particulate forms (1). Uncertainty regarding primary emission and secondary production of atmospheric particulate matter in CMAQ is currently undergoing evaluation (39). Early results suggest that the CMAQ generally underestimates PM10 between 20% and 50%. Sensitivity results presented in Cooter and Hutzell (Figure 1C in (1)) suggest such underestimation could lead to underestimation of the particulate fraction. The means of properly assessing the importance of this bias with regard to CMAQ partitioning is unclear, because the error depends on both primary emission and secondary production processes. Ammonia (NH3), sulfate, and nitrate emissions are modeled as interacting in the aqueous aerosol material and, thus, affect partitioning behavior (1). Major point source emissions of sulfate are known within a factor of 1.5 (27, 40) and nitrogen oxide emissions are generally known within a factor of 3 to 5, but emission estimates for NH3 are particularly suspect with regard to magnitude and seasonality (41). Under humid conditions, the CMAQ semivolatile partitioning algorithm is quite sensitive to the presence of NH3 (see ref 1, Figure 1A). VOL. 36, NO. 21, 2002 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
4597
The sensitivity of CMAQ particulate fraction predictions to NH3 emission uncertainty is assessed by increasing June NH3 emissions by 25% (42) over existing mean annual inventory values. CMAQ particulate fractions were repredicted and compared with USGS air samples collected during June for which both gas and particulate concentrations are greater than the RQL (Figure 5, green symbols, n ) 10). Increased NH3 emissions increase the median June particulate fraction and decrease median gas form concentrations slightly. Although these changes are not statistically significant (as determined by a Wilcoxon rank-sum test) and do not change the predicted air or rainfall concentrations significantly, the direction of change suggests that NH3 emission uncertainty could be an important contributor to particulate fraction prediction error. Precipitation. The relationship between rainfall volume (a function of precipitation amount and frequency) and weekly sample concentration for herbicides is well documented (21, 43). The first part of a precipitation event tends to scavenge most of the herbicide from the atmosphere, especially that associated with particulate material. Rainfall occurring later in the event dilutes the previous concentration and contributes little to the total mass of herbicide deposited. Plots of weekly observed and predicted precipitation against observed and predicted concentration show similar relationships (see Supporting Information, Figure SI1). An absence of low volume/high concentration samples obscures this relationship for longer accumulation intervals, i.e., LMMB (see Supporting Information, Figure SI2). Goolsby et al. (21) speculate that, because of the large temporal and spatial variation in the amount of rainfall across a geographically diverse sampling network, it is difficult to make meaningful comparisons of herbicide concentrations on the basis of individual weekly samples and advocates the use of precipitation-weighted (PW) concentrations. A similar argument can be made when considering concentrations associated with model-generated precipitation estimates. Table 2 shows important differences between the distribution of predicted and observed weekly precipitation totals for USGS monitoring sites, but the Spearman statistic indicates the data sets are correlated (n ) 76, R < 0.05). A PW concentration is the sum of the product of precipitation and concentration over a period of time, divided by the total precipitation (21). This calculation was performed for weekly samples and predictions between April 1 and July 18, and results are shown in Supporting Information Figure SI3. Precipitation weighted results still indicate model underprediction of concentrations at most sites, but all predicted/ observed pairs lie much closer to the 1:1 line. Predicted as well as observed 1995 PW concentrations from USGS locations in Iowa, Minnesota, and on Lake Superior are within the 1990 and 1991 range for these regions reported by Goolsby et al. (21). Similar agreement with Goolsby et al. (21) occurs for all PW CMAQ predictions and for observations for LMMB sites reporting at least one detection greater than the RQL (Beaver Island, Sleeping Bear Dunes, MI, Indiana Dunes, MI, Chiwauke Prairie, WI, Bondville, IL, and Brule River, WI). Predicted and observed PW values for these sites agree within a factor of 2. PW concentrations for LMMB sites reporting values between the MDL and RQL (South Haven, MI, Manitowoc, WI, Muskegon, MI, and Chicago, IL) depart significantly from the Goolsby et al. (21) ranges and show poor agreement with CMAQ PW concentrations. Observations at these four locations should be investigated further to more clearly identify the primary source of such differences. This analysis has shown that present emission inventories and meteorological models support seasonal to annual predictions of regional (Lake-wide) atrazine wet deposition (within 10%) and dry deposition (within the published range of uncertainty) for the LMMB study area. Precipitation 4598
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 36, NO. 21, 2002
weighting of 1995 monthly LMMB observations greater than the RQL, observed rainfall concentrations at all USGS sites, and all CMAQ PW predictions show good agreement with previously reported estimates (21). In most cases, these PW predictions and observations agree within a factor of 2 or better. Errors in ammonia and primary particulate emission inventories, secondary particulate material production, and the unknown contribution of wind erosion have been identified as likely contributors to CMAQ underprediction of observed particulate fraction atrazine. Observed gas form concentrations in the air are, for the most part, overpredicted by CMAQ. The source of this error is not clear, but the urban bias noted in Figures 3 and 4 suggest the possible presence of scale issues. That is, the present 36 km horizontal grid may lack or poorly represent subgrid scale processes that are important for the proper characterization of volatilization and gas and particulate form atmospheric fate (e.g., transformation and deposition). Additional process research and further model evaluation in these areas are clearly indicated.
Acknowledgments The authors thank Kenneth Rygwelski of the U.S. Environmental Protection Agency, Large Lakes Research Laboratory, Gosse Ile, Louis Blume of the U.S. Environmental Protection Agency, Great Lakes National Program Office, Chicago, and Marcia Kuehl of MAKuehl Company for providing insights regarding Lake Michigan hydrology and Lake Michigan Mass Balance Study data quality assurance. Supporting Information was provided by Louis Blume and Judy Schofield and Ken Miller of DynCorp, I&ET, Inc. The information reported here has been funded in part under an Interagency Agreement from the U.S. Environmental Protection Agency, Great Lakes National Program Office (DW13947769-01). It has been subjected to Agency review and approved for publication. Mention of trade names or commercial products does not constitute endorsement or recommendation for use.
Supporting Information Available Information is provided regarding the estimation of LMMB and U.S. Geological Survey sample uncertainties, the impact of surface runoff loss uncertainty on atrazine volatilization estimation, LMMB and U.S. Geological Survey concentration/ rain volume relationships and U.S. Geological Survey observed vs predicted rainfall-weighted concentrations. This material is available free of charge via the Internet at http:// pubs.acs.org.
Literature Cited (1) Cooter, E. J.; Hutzell, W. T. Environ. Sci. Technol. 2002, 36, 40914098. (2) Ching, J. K. S.; Byun, D. W. In Science Algorithms of the EPA Models-3 Community Multiscale Air Quality (CMAQ) Modeling System; Byun, D. W., Ching, J. K. S., Eds.; U.S. Environmental Protection Agency, Office of Research and Development: Washington, DC, 1999; EPA-600/R-99/030. (3) Ciba Crop Protection In U.S. EPA Pesticide Registration Records; Ciba Crop Protection: Greensboro, NC, U.S.A., 1994. (4) Cassarett and Doull’s Toxicology: The Basic Science of Poisons; Amdur, M. O., Doull, J., Klassen, C. D., Eds.; Pergamon Press: New York, 1991. (5) Barbash, J. E.; Resek, E. A. Pesticides in ground water: Distribution, trends, and governing factors; Ann Arbor Press: Chelsea, MI, 1996. (6) Glotfelty, D. E.; Leech, M. M.; Jersey, J.; Taylor, A. W. J. Agric. Food Chem. 1989, 37, 546. (7) Larson, S. J.; Capel, P. D.; Majewski, M. S. Pesticides in surface waters: Distribution, trends, and governing factors; Ann Arbor Press: Chelsea, MI, 1997. (8) Majewski, M. S.; Foreman, W. T.; Goolsby, D. A.; Nakagaki, N. Environ. Sci. Technol. 1998, 32, 3689. (9) Majewski, M. S.; Foreman, W. T.; Goolsby, D. A. Sci. Total Environ. 2000, 248, 201.
(10) Miller, S. M.; Sweet, C. W.; DePinto, J. V.; Hornbuckle, K. C. Environ. Sci. Technol. 2000, 34, 55. (11) Thurman, E. M.; Cromwell, A. E. Environ. Sci. Technol. 2000, 34, 3070. (12) Scholtz, M. T.; Voldner, E.; McMillan, A. C.; Van Heyst, B. J. Atmos. Environ. 2002, 36, 5011. (13) Pleim, J. E.; Xiu, A. J. Appl. Meteor. 1995, 34, 16. (14) Press, W. H.; Teukolsky, S. A.; Vertterling, W. T.; Flannery, B. P. Numerical Recipes in FORTRAN: The art of scientific computing; Cambridge University Press: New York, 1992. (15) Keuhl, M. Lake Michigan Mass Balance Quality Assurance Report, Atrazine; 905R-01-013; U.S. Environmental Protection Agency, Great Lakes National Program Office: Chicago, 2001. (16) Zaugg, S. D.; Sandstrom, M. W.; Smith, S. G.; Fehlberg, K. M. Methods of analysis by the U.S. Geological Survey National Water Quality Laboratory- -Determination of pesticides in water by C-18 solid-phase extraction and capillary-column gas chromatography/mass spectrometry with selected-ion monitoring; OpenFile Report 95-181; U.S. Geological Survey: 1995. (17) Foreman, W. T.; Majewski, M. S.; Goolsby, D. A.; Wiebe, F. W.; Coupe, R. H. Sci. Total Environ. 2000, 248, 213. (18) Adams, N. H. Quality Assurance 1998, 5, 257. (19) U.S. Environmental Protection Agency. Lake Michigan Mass Budget/Mass Balance Work Plan; EPA-905-R-97-016; Great Lakes National Program Office: Chicago, IL, 1997. (20) Schottler, S. P.; Eisenreich, S. J. Environ. Sci. Technol. 1997, 31, 2616. (21) Goolsby, D. A.; Thurman, E. M.; Pomes, M. L.; Meyer, M. T.; Battaglin, W. A. Environ. Sci. Technol. 1997, 31, 1325. (22) Goolsby, D. A.; Scribner, E. A.; Thurman, E. M.; Pomes, M. l.; Meyer, M. T. Data on selected herbicides and two triazine metabolites in precipitation of the midwestern and northeastern United States, 1990-1991; Open-file Report 95-469; U.S. Geological Survey: Denver, 1995. (23) Wilks, D. S. Statistical Methods in the Atmospheric Sciences; Academic Press: San Diego, CA, 1995. (24) Olds, E. G. Annuals Mathematical Statistics 1938, 9. (25) Byun, D.; Pleim, J. E. In Air Pollution Modeling and Its Application XIV; Gryning, S.-E., Schiermeier, F. A., Eds.; Kluwer Academic/ Plenum: New York, 2001; p 203. (26) Science Algorithms of the EPA Models-3 Community Multiscale Air Quality (CMAQ) Modeling System; Byun, D., Ching, J., Eds.; EPA/600/R-99/030; United States Environmental Protection Agency, Office of Research and Development: Washington, DC, 1999. (27) Russell, A.; Dennis, R. L. Atmos. Environ. 2000, 34, 2283. (28) Liu, C.; Bennett, D. H.; Kastenburg, W. E.; McKone, T. E.; Brown, D. Reliability Eng. System Safety 1999, 63, 169.
(29) Roselle, S. J.; Binkowski, F. S. In Science Algorithms of the EPA Models-3 Community Multiscale Air Quality (CMAQ) Modeling System; EPA-600/R-99/030; Byun, D. W., Ching, J. K. S., Eds.; United States Environmental Protection Agency, Office of Research and Development: Washington, DC, 1999. (30) Howard, P. H. Handbook of Environmental Fate and Exposure; Lewis Publishers: Chelsea, MI, 1991; Vol. 3. (31) Cohen, M.; Commoner, B.; Bartlett, P.; Cooney, P.; Eisl, H. Exposure to Endocrine Disruptors from Long-Range Air Transport of Pesticides; Report to the W. Alton Jones Foundation, Flushing, NY; CBNS, Queens College, City University of New York: New York, 1997. (32) Howard, P. H.; Meylan, W. M. Atmospheric Oxidation Rate Program; Lewis Publishers: Syracuse, NY, 1992. (33) Capel, P. D.; Larson, S. J. Environ. Sci. Technol. 2001, 35, 648. (34) Theilin, G. P.; Gianessi, L. P. Method for estimating pesticide use for county areas of the coterminous United States; Open-File Report 00-250; U.S. Department of the Interior, U.S. Geological Survey: Sacramento, CA, 2000. (35) U.S. Department of Commerce, Bureau of the Census, Data User Services Division. http://usda.mannlib.cornell.edu/usda (accessed June 2002). (36) United States Department of Agriculture, National Agricultural Statistics Service. http://usda.mannlib.cornell.edu/usda (accessed June 2002). (37) United States Department of Agriculture, National Agricultural Statistics Service, Agricultural Statistics Board. http://usda. mannlib.cornell.edu/usda (accessed June 2002). (38) Li, Y.-F. Atmospheric Environment Service (ARQI), Environment Canada, Toronto, ON, Canada, personal communication, 1999. (39) Mebust, M.; Eder, B. K.; Binkowski, F. S.; Roselle, S. J. J. Geophys. Res. in press. (40) Hanna, S. R.; Lu, Z.; Frey, H. C.; Wheeler, N.; Vukovich, J.; Arunachalam, S.; Fernau, M.; Hansen, D. A. Atmos. Environ. 2001, 35, 891. (41) Pierce, T. E.; Bender, L. W. In Proceedings of the A&WMA Emissions Conference; Air & Waste Management Association: Raleigh, NC, October 26-28, 1999; 2000. (42) Gilliland, A. B.; Dennis, R. L.; Roselle, S. J. Scientific World J. 2001, 1, 356. (43) Capel, P. D. Wet deposition of herbicides in Minnesota; Water Resources Investigations Report 91-4034; U.S. Geological Survey Toxics Substances Hydrology Program: Monterey, CA, 1991.
Received for review October 17, 2001. Revised manuscript received August 8, 2002. Accepted August 21, 2002. ES011372Q
VOL. 36, NO. 21, 2002 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
4599
