Thursday, June 25, 1998 9:30 a.m.

R. Peter Richards

Senior Research Scientist

Water Quality Laboratory

Heidelberg College

310 East Market Street

Tiffin, OH 44083

Mr. Chairman and members of the Subcommittee, thank you for the opportunity to testify before you today regarding the Food Quality Protection Act (FQPA). My remarks and written testimony will focus on the estimation of exposures to pesticides through drinking water, one of the major pathways of pesticide exposure evaluated under the FQPA.

The Water Quality Laboratory (WQL) has been studying pollution in the major tributaries to Lake Erie since 1975. The land use in these tributaries is dominated by row-crop agriculture, so the pollution is mostly from agricultural sources. The exception is the Cuyahoga River basin, which is dominated by urban-suburban and forested land uses. We have been studying concentrations of more than a dozen currently-used pesticides in these rivers since 1983, during which time we have analyzed more than 4000 samples of river water to determine their pesticide concentrations. The pesticides we quantify account for about 90% by weight of the pesticides used in the basin.

Since 1988 we have operated a voluntary private rural well testing program, which includes analysis for the major herbicide groups: the triazines and acetanilides. We have analyzed more than 10,000 well water samples for acetanilides and more than 12,000 for triazines. These samples are from locations throughout the United States, but concentrated in the midwestern states of Ohio, Indiana, West Virginia, Kentucky, and Illinois.

We have also participated in projects which have attempted to reproduce observed pesticide concentrations using computer models, and in projects to examine the distribution of human exposures to herbicides via drinking water, using the best available monitoring data to estimate drinking water concentrations.

These extensive experiences with pesticides have revealed the following general patterns, which should be valid in general terms for many other pesticides for which monitoring data are lacking, particularly if adjusted appropriately by taking into account the known properties of the unmonitored pesticides, such as degradation rate, water solubility, and tendency to attach to sediment particles or organic matter. These patterns should also be valid in general terms in other parts of the country, when differences are considered in the seasons of use, quantities applied, and regional climate.

By the time of the hearing we will have provided the committee with copies of peer-reviewed journal articles to support the following statements. Many of these statements are also generally supported by published results of other studies, including studies by the U.S. EPA and U.S. Geological Survey.

Pesticide concentrations in rivers and streams are seasonal in nature. The highest concentrations occur over a short period of time following application to crops. This period typically lasts from several weeks to about two months depending on the annual climate and the degradation rate of the pesticide. During this "pesticide runoff season", concentrations are highly variable from day to day, with high concentrations associated with runoff from the land following precipitation events. Lower concentrations are associated with low flow conditions between rainfall runoff events.

During the rest of the year, concentrations of pesticides are quite low. Many pesticides are not detected at all outside of the pesticide runoff season; others are detected at concentrations which are usually below part-per-billion levels. The distribution of pesticide concentrations in several of our study rivers is shown in Table 1.

In general, higher concentrations are associated with smaller streams whose watersheds are dominated by agricultural land use. Larger rivers have lower concentrations. In the rivers in our study area which are used as sources for drinking water, no annual average concentration of any pesticide has exceeded the corresponding health advisory level or maximum contaminant level for that pesticide during the 15 years of our study. According to EPA risk assessment procedures, the annual average concentration is the concentration measure which is appropriate for comparison with long-term health advisory levels, which are concentrations which the EPA believes provide a large margin of safety against adverse effects from lifetime human consumption.

Concentrations high enough to cause acute, short-term effects have not been observed in any sample for any of the compounds studied. Since drinking water concentrations are never higher than the raw water source from which they are drawn (see below), this means that acute health effects due to pesticide exposure through drinking water are not expected to occur.

Lakes and reservoirs serve to dampen the concentration fluctuations that characterize the rivers and streams that flow into them. Consequently, the concentrations in lakes and reservoirs are relatively stable over time periods of days to weeks, but may fluctuate over longer time periods of months, years, or decades, depending on the volume of the lake or reservoir compared to the volume of inflow and outflow.

In general, pesticide concentrations in large lakes and reservoirs are low relative to health advisory levels. Pesticide concentrations in most small lakes and reservoirs are also low, but the likelihood of encountering concentrations near or above health advisory levels increases as the size of the lake or reservoir (and its watershed) decreases, and as the percent of the watershed in agricultural land use increases.

Most wells do not show detectable concentrations of pesticides: in our studies, more than 90% of wells were free of acetanilide herbicides and more than 95% were free of triazines. Only one well per thousand (0.1%) contained triazines in excess of the maximum contaminant level for atrazine. About one well per hundred (1.1%) had concentrations of acetanilides in excess of the alachlor health advisory level, but the compound in these wells which was detected in the tests was usually not alachlor but ethane sulfonate, a much less toxic breakdown product.

Concentrations in properly constructed, properly maintained wells change only slowly over time. Wells which are shallow, old, constructed in sandy soils, and/or developed in aquifers unprotected by an overlying impermeable geologic stratum are more likely to have detectable pesticide concentrations than wells which have the opposite characteristics. Private wells are more likely to have detectable pesticide concentrations than municipal wells.

Both in surface water and in ground water, the frequency of detection of a pesticide, and its observed concentrations, are related to the amount used in the area, to the breakdown rate of the pesticide, and to the affinity of the pesticide for attachment to soil particles. Pesticides which are widely used and in large amounts are more frequently detected in the water. Pesticides which break down quickly are less frequently seen than those which break down slowly. Pesticides which attach strongly to soil particles rarely occur in groundwater, but may occur in surface water when sediment is in transport. When sediment is deposited, the pesticide tends to be deposited with the sediment.

Because insecticides are used in smaller quantities and on fewer acres than herbicides, especially in the Midwest corn-soybean belt, and because they tend to degrade more rapidly, they are detected less frequently and in smaller concentrations than herbicides, as can be seen from Table 1.

Concentrations of pesticides in finished drinking water are, on average, lower than those in the raw water entering the treatment plant. Concentrations of some pesticides in finished water may be approximately equal to those in the raw water, but it is hard to imagine a situation where they would be higher on average. The extent to which they are lower depends on the nature of the treatment they receive and the properties of the pesticides.

Pesticides which adhere strongly to sediment particles get filtered out of the finished water during clarification. Such pesticides are almost totally removed from the finished water. With some important exceptions, current-generation insecticides tend to adhere to sediment particles more strongly than herbicides.

Some water treatment plants use activated carbon as part of the purification process. Depending on the pesticide's concentration and properties, the amount of activated carbon added, and the details of the process, between half and nearly all of the pesticide will be removed by this treatment. Some pesticides are broken down by chlorination or other treatments applied at most water treatment plants.

The importance of these relationships lies in the fact that many monitoring programs which have produced pesticide concentration data have been carried out for other purposes than drinking water quality, and they have measured concentrations in natural waters rather than finished drinking water. These concentrations can be used as conservative (somewhat high) estimates of the equivalent drinking water concentrations. They can be used as is to provide an estimate of drinking water quality, or adjusted by knowledge of the pesticide properties and treatment plant operations to provide a more precise estimate of drinking water quality.

When data are not available or are of unknown quality, it is tempting to turn to models to estimate concentrations in drinking water. This is not a good idea, in our opinion, for several reasons. Unless models are calibrated using good data, and then validated to show that they can reproduce the important properties of other data sets of good quality not used in calibrating them, no one can say whether their predictions are accurate. Many modelers contend that models should never be used to predict actual concentrations. They contend that the only reliable use of models is to explore different scenarios, and see which ones indicate higher concentrations and which ones indicate lower concentrations. In other words, models should be used to explore relative differences but not absolute quantities or differences.

We participated in an ecosystem risk study of the herbicide atrazine, in which the computer model called PRZM was used. This is the same model that EPA has been using for assessing water tolerances under the Food Quality Protection Act (FQPA). In our study, PRZM was calibrated to reproduce the water flow accurately for conditions monitored at several small research study sites. It was then used to predict concentrations of atrazine during the same period of time, based on the application rates and chemical properties of atrazine. In several runs of the model for different study sites, PRZM consistently predicted atrazine concentrations about ten times higher than were observed to occur.

The EPA has been using a modified version of PRZM called GENEEC, which involves modeling pesticide concentrations in a farm pond adjacent to a field, to which the pesticide in question is applied according to standard practices. The model then simulates the concentrations in the pond in response to runoff from the field following rainfall. A number of extreme assumptions are made, including that all of the runoff from the field enters the pond, no other water enters the pond, and that is no dilution of the pesticide in the runoff by water already in the pond. The maximum concentration is used as the estimated concentration for short-term drinking water exposure, and various kinds of average concentrations are used as estimates of long-term exposure.

What we have learned from our monitoring studies indicates that a pond in the middle of a field will have much higher concentrations than any reasonable drinking water source. This is partly because small water bodies tend to have high concentrations, partly because the pond is totally surrounded by the field and its pesticide load (not the usual situation for drinking water supplies), and partly because there is no intervening land use to interrupt or dilute the effects of pesticide runoff from the field. Edge of field concentrations are always much higher than those in water bodies of a size likely to be chosen for water supplies.

The EPA assessment approach involves using a model which over-predicts concentrations, applying it to a scenario which itself would lead to much higher concentrations than would occur in any reasonable water supply, and making a number of assumptions about the details of the runoff process which further inflate the estimated concentrations. This stringing together of extremely improbable conditions will inevitably lead to an extraordinarily high and unreasonable reference concentration.

The likely consequence of this approach is that nearly every compound evaluated will show a hazard of acute health effects. This may trigger direct evaluation of the acute exposure risks of the compound, by specially-designed drinking-water monitoring studies. These studies will be expensive, time consuming, and will in most and probably all cases reveal that this risk is actually very nearly zero. Alternatively, EPA may attempt to withdraw the tolerance and effectively stop the use of beneficial compounds, based only on model-derived concentrations that appropriate monitoring would demonstrate never occur in drinking water.

The extent to which the modeled concentrations are inflated is not known. It would be very useful to use the model to estimate concentrations for a well-studied compound such as atrazine, and compare the model results with the existing source water and drinking water data. This would provide some indication of the extent to which predicted concentrations for other compounds may be exaggerated.

There is a large body of data on pesticide concentrations in water, both in drinking water and in raw water sources. EPA already has much of this data. Some compounds have much larger databases than others. But much can be learned about the likely behavior even of unmonitored compounds by reference to data for well-studied compounds, adjusting for differences in use patterns and amounts, rates of degradation, water solubility, and affinity for attachment to soils and sediments. This monitoring data should be used in drinking water assessments for the FQPA, and in most cases should be given more credibility than results of modeling.

Most drinking water sources contain concentrations of studied herbicides which present no health risk under the provisions of the Safe Drinking Water Act, and there is every reason to believe that those which are not monitored also present no health risk. Levels which would indicate acute risk are not seen in monitoring studies. There are infrequent situations in which pesticides exceed long-term (chronic) health advisory levels on an annual average basis; in these cases remediation is called for to provide assurance that health risks from chronic exposure have been reduced to established acceptable levels. The "aggregate risk" provisions of the FQPA change the way these situations would be evaluated, but the basic pattern of no acute risk and infrequent localized chronic risk is likely to remain valid.

Where new monitoring studies are needed to assess drinking water risks, the monitoring design should take into account the known patterns of occurrence of pesticides in rivers and streams, lakes and reservoirs, and groundwater, and the differences in concentrations in these source waters as compared to the finished water derived from them, described above. When no previous data exists, targeted and carefully designed monitoring programs are the only way to provide accurate and appropriate data of known validity for assessment of drinking water within the framework of the FQPA. These data can also be used to calibrate appropriate and realistic models, which can then be used to simulate time frames longer than the monitoring period, other land use distributions, etc.

Models can provide useful insights, especially when employed in a comparative mode of analysis. To provide useful absolute information such as estimated concentration distributions, models must be carefully calibrated and repeatedly validated on a compound-by-compound basis. They must be used to model realistic scenarios which accurately reflect the issues of concern. The use of overly simplistic models with unreasonable scenarios involving multiple worst-case assumptions will inevitably lead to dire predictions which have no correspondence to reality and therefore are useless. Even when used as a screening tool, such models are useless because they probably won't screen anything out!