comparison to other CWSs in the baseline study. The subset of CWSs to continue monitoring in future years should also include CWSs that had low and moderate levels of pesticides in the baseline study as they may just be in a low range of a temporal trend and future years will exhibit higher pesticide levels.
Instead, some form of interpenetrating survey design might be used. In a simple example, the total number of CWSs is reduced by a fraction, say 20%, and an equal fraction of sites are selected for multi-year sampling, keeping the total number of samples the same. If a longer time view is selected, that is a true monitoring plan is desired, CWSs could be sampled for a couple of years then replaced with other CWSs with no one CWS being in the sample for more than say 3 years. This allows the monitoring plan to cover the whole region fairly uniformly over time while also collecting the needed information on temporal variability.
The question to the Agency is this, "How do you plan to interpret the resulting distributions of parameter estimates if no temporal sampling is performed?" In agricultural research, it is traditional that studies be replicated over three years, five is preferred, if the researcher wishes to make believable inferences to expected results over time. Something similar to this needs to be incorporated into this study as well. Uncertainty estimates in annual average pesticide levels in CWSs would be underestimated and lead to biased confidence intervals for parameters of concern in this study if year-to-year variability is not included.
Even if additional funding is not provided, the Agency should not adopt a single year sampling scheme. The resulting exposure estimates will not be credible because of the significant year-to-year variability in the factors that contribute to pesticide occurrence. Three years should be the minimum duration considered; five years would be preferred.
8.) EFED recognizes that pesticide concentrations in drinking water are dependent on factors including watershed characteristics, pesticide use, pesticide fate properties, surface water hydrology, and water treatment processes. Interpretation of the monitoring data will be dependent on the collection of such related ancillary data.
What types of ancillary information does the SAP believe would assist in the
interpretation of the monitoring data, and application of the data to model development and validation?
While exposure estimates could be made without use of ancillary data, model construction requires it. The Agency has identified a reasonable set of watershed properties and ancillary data. A greater concern expressed by the SAP was not which data but the quality and scale of these data. It is critical that these be collected as part of the study. Development of models (and their successive refinements) is an important component for determining vulnerability. An aspect that hasn't been discussed is to validate the data layers used in developing models--in other words, how accurate are the data derived from county-level crop surveys when aggregated by watershed. Much of the relevant information (e.g., physical characteristics of the watershed) may already be