Director’s Forum
Kenneth H. Reckhow, Director
Water Resources Research Institute

Scientific issues in the development of total maximum daily loads (TMDLs)

One of the relatively new strategies for meeting ambient water quality standards is the establishment of a Total Maximum Daily Load, or TMDL. In brief, the TMDL program requires that each state identify waterbodies which fail to achieve water quality standards based on point source controls, and then develop a plan for additional (point and nonpoint source) controls to meet the standards. The November/December 1996 WRRI News or the EPA TMDL website (http://www.epa.gov/OWOW/tmdl) should be consulted for additional information.

Of particular interest here are: (1) the scientific basis for the identification of the waters in a state in need of a TMDL, and (2) the methods of scientific assessment for the development of the TMDL for a particular waterbody.

Waterbodies, or segments of waterbodies, that require a TMDL are to be identified on the 303(d) list for each state. This list may include “evaluated waters” which are based on old ambient monitoring or on non-monitoring assessments such as fish surveys or predictive modeling. Alternatively, the list may include “monitored waters” which are assigned to the list based on recent ambient monitoring, usually as part of the 305(b) assessment. Each of these has drawbacks.

For example, “evaluated waters” reflect an indirect or dated assessment, neither of which may signify current water quality standard violations. Therefore, the preference may be to rely on recent 305(b) monitoring. However, 305(b) monitoring is undertaken to assess the extent to which the population of waters of a state support their designated uses. Monitoring to make inferences concerning the water quality in the population of a state’s streams, lakes, and estuaries should be based on probability sampling. In contrast, monitoring to identify waterbodies with standard violations should be targeted to waterbodies suspected to have standard violations. Thus, these competing objectives for the 305(b) program will lead to different designs, and most state 305(b) monitoring programs are not optimized to meet either objective.

Once the TMDL need is identified, an assessment is required to provide a scientific basis for the allowable pollutant loading. This assessment is likely to involve predictive modeling, so EPA recently released the BASINS (Better Assessment Science Integrating Point and Nonpoint Sources) model for that purpose. BASINS is packaged with a large database and runs in Windows within an GIS framework. This approach by EPA, to support TMDL assessment with a comprehensive modeling package, seems like a good idea. Unfortunately, the actual models contained in BASINS lack essential features for management applications.

BASINS includes the mechanistic models HSPF and QUAL2E. The mechanistic modeling approach reflects the seemingly reasonable belief that the best predictive models result from detailed mathematical characterization of basic scientific processes. However, nature is complex, and even the most descriptive simulation model is extremely simple in comparison. At some point, additional detail exceeds our ability to simulate and predict with reasonable error levels. Based on the few comprehensive, rigorous error propagation studies available in the literature, this point appears to have been exceeded for HSPF.

This observation should not be interpreted as an indictment of the science in HSPF; rather, it is a statement about the difficulty in predicting spatial/temporal responses in an extremely complex environment. So what should be done?

We must first recognize that the objective of model building in this case is not to try to correctly describe all processes (an impossible task), but to provide useful or reliable predictive assessments. An essential measure of the reliability or validity of a TMDL assessment is the prediction uncertainty in that assessment. Unfortunately, studies indicate that detailed process description and low prediction uncertainty are often incompatible in surface water quality models. Predictive models should not be promoted for management applications until rigorous, thorough error propagation studies, reflecting all error terms, confirm acceptable prediction errors.

It seems likely that emphasis on prediction uncertainty will lead to the development and application of models for TMDL assessment that are aggregated in space, time and process characterization and thus are less detailed than is HSPF. In fact, this is already occurring, as some users of BASINS have substituted simpler mechanistic models in BASINS in place of HSPF.

In summary, two improvements are recommended for the scientific assessment of TMDLs. First, the 305(b) monitoring program within a state should have clearly expressed objectives and then be optimized to meet those objectives. Thus, if the 305(b) report is to characterize all waterbodies using a sample, it should be a probability sample; if the 305(b) report is to identify ambient standard violations, it should be targeted toward suspected violations. Second, TMDL prediction should be accompanied by a complete uncertainty analysis; this provides the essential measure of validity of the predictive assessment. Rigorous testing to assess reliability is standard practice for most technology; why should it be otherwise for water quality simulation models?