ADAPTIVE IMPLEMENTATION OF TMDLS USING BAYESIAN ANALYSIS
In their recent report on the USEPA's total maximum daily load (TMDL) program, a panel of the National Research Council (NRC) endorsed the program and stated that the science was sufficient for the program to move forward. However, the panel also acknowledged the limits of scientific
analysis, stating that “(s)cientific uncertainty is a reality within all water quality programs, including the TMDL program, that cannot be entirely eliminated.” The magnitude and consequence of scientific uncertainty is particularly evident in the water quality models necessary
for TMDL forecasting and development.
To address the possibility that the often-substantial uncertainty associated with water quality model forecasts may result in implementation of inefficient and flawed TMDL plans, the NRC TMDL panel recommended that an iterative implementation strategy,
termed “adaptive implementation,” be employed. In brief, adaptive implementation, or “learning by doing,” augments initial model forecasts of TMDLs with post-implementation monitoring that provides an assessment of actual response by the aquatic system of interest.
The need for integrated TMDL modeling and monitoring approaches is expressed in the following recommendation by the NRC panel:
In order to carry out adaptive implementation, EPA needs to foster the use of strategies that combine monitoring and modeling and expedite TMDL development (NRC
Report, page 10).
Fortunately, a number of analytic approaches already exist to perform the analysis for an adaptive implementation of a TMDL. The approach is essentially Bayesian although variations of the Bayesian theme have existed and been implemented for years (e.g., the Kalman filter
and data assimilation). In basic terms, the initial TMDL model forecast serves as the Bayesian prior, the post-implementation monitoring data serve as the sample information (the likelihood), and the posterior probability results from a pooling of the evidence and provides the basis for a
revised (and improved TMDL). A simple example of a Bayesian model for adaptive implementation is applied to a lake eutrophication assessment problem to illustrate the process. In addition, a nitrogen TMDL recently developed for the Neuse Estuary in North Carolina is examined for adaptive implementation.
These examples illustrate the feasibility of technical assessments for adaptive implementation, and they also serve to identify some scientific and analytic issues that need to be addressed to facilitate effective use of modeling-monitoring integration for adaptive implementation.
More about this publication?
Open access content
Free trial content