Grantee Research Project Results
2004 Progress Report: Adaptive Implementation Modeling and Monitoring for TMDL Refinement
EPA Grant Number: R830883Title: Adaptive Implementation Modeling and Monitoring for TMDL Refinement
Investigators: Reckhow, Kenneth H. , Stow, Craig A. , Shabman, Leonard A. , Arhonditsis, George B. , Borsuk, Mark E. , Qian, Song S. , Roessler, Chris , McMahon, Gerard
Current Investigators: Reckhow, Kenneth H. , Stow, Craig A. , Shabman, Leonard A. , Borsuk, Mark E. , Roessler, Chris , McMahon, Gerard
Institution: Duke University , United States Geological Survey , Swiss Federal Institute for Environmental Science & Technology (EAWAG) , North Carolina Division of Water Quality (DWQ) , Resources for the Future , University of South Carolina at Columbia
Current Institution: Duke University , Resources for the Future , United States Geological Survey
EPA Project Officer: Packard, Benjamin H
Project Period: June 1, 2003 through May 31, 2006 (Extended to October 30, 2007)
Project Period Covered by this Report: June 1, 2004 through May 31,2005
Project Amount: $660,171
RFA: Developing Regional-Scale Stressor-Response Models for Use in Environmental Decision-making (2002) RFA Text | Recipients Lists
Research Category: Ecological Indicators/Assessment/Restoration , Aquatic Ecosystems
Objective:
The overall objective of this research project is to develop an adaptive implementation modeling and monitoring strategy (AIMMS) for total maximum daily load (TMDL) improvement. AIMMS will allow us to integrate analytically TMDL modeling with postimplementation monitoring to refine and improve the TMDL over time. Two probabilistic models with the ability to support error propagation, the NeuBERN Bayes network estuary model and the U.S. Geological Survey Neuse Spatially Referenced Regressions on Watershed Attributes (SPARROW) model, will be linked in AIMMS. The case study for application and evaluation of AIMMS would be the Neuse Estuary nitrogen TMDL in North Carolina.
The specific objectives of this research project are to: (1) assess the value of information (value of additional monitoring) for TMDL compliance evaluation using the linked models in AIMMS; and (2) design and evaluate additional monitoring to update the TMDL forecast using AIMMS. Another objective of this project is to develop and test a process for engaging stakeholder decisionmakers in refining the format of the model outputs and endpoints. The anticipated result from the last objective is to ensure the model’s utility and credibility as an adaptive management decision support tool. In a general sense, this project should have broad applicability as a framework to update and improve model forecasts (and management actions) over time. In a specific sense, the Neuse application of AIMMS will provide an informative and useful case study, which will serve as a basis for the required 2006 North Carolina Division of Water Quality reevaluation of designated use support in the Neuse Estuary.
Progress Summary:
After the completion of the Bayesian analysis of SPARROW (Qian, et al., 2005), we focused on two fronts. We addressed the mismatch of temporal scales between the SPARROW model and the NeuBERN (Borsuk, et al., 2004a,b) model. As described in our last progress report, we proposed to use a probabilistic framework ( Box 1 of the 2004 progress report). The proposed framework is, however, problematic. This was because temporal scale mismatch is not the only problem in integrating the two models. The SPARROW model for the watershed was fitted using 1991-1992 data. Changes in the watershed since then are not reflected in the estimated nutrient loadings. As a result, even with the temporal scale mismatch resolved, we still were unable to integrate the watershed model and the estuary water quality model to evaluate changes in water quality over time in response to the reduced nitrogen loading resulting from the implementation of a TMDL program. With this discovery, we moved our focus on how to use the limited nitrogen loading data in the watershed since 1992 to provide updated estimates of nitrogen loading changes over time. We have worked in two different areas. First, we started a series of studies on the use of a simple Bayesian approach for updating nutrient loadings and nutrient concentrations. Second, we started to use the Neuse River Digital Watershed (from the National Science Foundation [NSF] Consortium of Universities for the Advancement of Hydrologic Science, Inc. [CUAHSI] program) to provide annual nutrient loading data and revise our Bayesian SPARROW model to update model parameters, as well as estimate nitrogen loadings when new data are available. In this report, we describe progress made in these two areas.
Bayesian Updating Using a Conjugate Family of Prior Distributions
Information synthesis is often the motivation for adapting the Bayesian approach. The conventional application of a Bayesian approach, however, emphasizes the combination of prior information and a single set of data. Although it is shown in Bayesian statistics texts that sequential updating using the posterior probability from the previous step as the prior probability is equivalent to updating using all of the data together, sequential updating provides a means to investigate possible temporal patterns in the data, which is attractive for our proposed adaptive implementation for TMDL. In the past year, we have developed a series of sequential updating procedures, including procedures for updating our estimates of nutrient and chlorophyll α concentrations and updating parameters of an estuary chlorophyll α model.
Small sample sizes often characterize water quality data. Model estimated concentrations are often less reliable than measurements because of uncertainty associated with the model and model parameter estimates. Thus, using monitoring data to supplement model prediction can be a prudent strategy. In addition, because model parameters often are estimated based on historical data, predictions of water quality concentrations can be irrelevant to current or future conditions. For example, when a model is used for TMDL planning, the model’s parameters are inevitably estimated based on pre-TMDL data. Once the TMDL program is implemented, changes in waste generation and delivery processes may require a modified model or modified model parameters. Consequently, updating a model’s prediction using monitoring data may be essential.
We developed a series of computer programs to automate the process of updating water quality concentration estimation from model predictions and subsequent monitoring data. These programs use the Bayesian statistics results for (log) normal random variables and the conjugate family of prior distributions. The process has three steps. First, we developed a procedure for converting model predictions of water quality concentrations to a prior distribution of the underlying concentration distribution parameters (the “true” mean and variance). Second, we developed a program to produce the posterior distribution of the underlying concentration distribution parameters and the posterior predictive distribution of future observations. Third, the posterior distribution of the underlying concentration distribution parameters is converted to a prior distribution of the same parameters for the next time period, and the process repeats when new data are available.
Water Quality Model
In the past year, we focused on how the proposed model can fit into the Bayesian updating process described above. We have completed a preliminary study on the use of Markov Chain Monte Carlo (MCMC) simulation for sequential model parameter updating. This process is implemented under the newly available WinBUGS development software, which allows us to construct a moderately complex model (represented by a set of partial differential equations) and yet easily make use of the MCMC methods implemented under WinBUGS. We tested an algal growth model for the Neuse River Estuary. The model is structured as a sequence of five completely mixed stirred tank reactors. Within each compartment, two partial differential equations are used to describe: (1) the process of algal growth as a function of nutrient supply and sedimentation; and (2) changes in nitrogen concentration as a function of loading from upstream, plus consumption and sedimentation within the compartment. By implementing our water quality modeling work in WinBUGS, instead of AQUASIM and UNCSIM as reported in the 2004 progress report, we unify the modeling platform and enable advanced Bayesian model parameter updating.
Future Activities:
After careful research, we believe that the optimal method for linking the watershed and estuarine models is the use of the digital watershed developed under the CUAHSI project. CUAHSI is an NSF-funded project to foster advancements in the hydrologic sciences. A prototype digital watershed was developed through CUAHSI using data from the Neuse River Basin. We experimented with the ways to link a GIS-based data retrieval system and the Bayesian SPARROW model using a Python script. From the Python script, we are able to pass the data collected by the GIS system to R and from R to WinBUGS. Once WinBUGS finishes updating the SPARROW model, results are passed back to R, then to Python, and then back to GIS. This system combines the sophisticated statistical modeling and the advanced spatial presentation of GIS. Once this combined GIS-WinBUGS system is completed, we will be able to produce updated annual prediction of summer nitrogen loading to the estuary. As a result, it is then possible to link the watershed nitrogen loading model to the water quality model for the estuary. Once the models are linked, the integrated watershed-waterbody response model will be updated with the use of our extensive historic database (covering both TMDL pre- and postimplementation periods). Using data from the postimplementation period will allow us to assess the TMDL effectiveness and make refinements to attain the criterion (chlorophyll) and meet designated uses. Furthermore, we intend to use the linked model as a means for optimizing the sampling network and identifying needs for additional monitoring in the Neuse drainage network and the Neuse Estuary. Based on this assessment of the value of information (value of additional monitoring), we will design a sampling network that focuses on sites where the greatest forecast uncertainty occurs but also minimizes sample redundancy caused by spatial or temporal correlation. The collection of this additional information (if necessary) will allow us to update the linked model, update the TMDL forecast, and reassess the TMDL effectiveness. Also important for refining model’s credibility would be the role of decisionmakers and stakeholders. We expect that their suggestions will help us to extend the tools of analysis to better characterize the certainty and uncertainty of stressor-response relationships (and stressor reduction-response) relationships for complex water bodies. Finally, the stakeholder contribution would be vital in selecting meaningful water quality (and modeling) end points, and thus one of our future objectives would be the use of our modeling framework for supporting decisions on water quality standard setting by stakeholders.
Journal Articles on this Report : 5 Displayed | Download in RIS Format
Other project views: | All 9 publications | 6 publications in selected types | All 5 journal articles |
---|
Type | Citation | ||
---|---|---|---|
|
Borsuk ME, Stow CA, Reckhow KH. Integrated approach to total maximum daily load development for the Neuse River Estuary using Bayesian probability network model (Neu-BERN). Journal of Water Resources Planning and Management 2003;129(4):271-282. |
R830883 (2004) |
Exit |
|
Borsuk ME, Stow CA, Reckhow KH. A Bayesian network of eutrophication models for synthesis, prediction, and uncertainty analysis. Ecological Modelling 2004;173(2-3):219-239. |
R830883 (2003) R830883 (2004) |
Exit Exit Exit |
|
Borsuk ME, Stow CA, Reckhow KH. Confounding effect of flow on estuarine response to nitrogen loading. Journal of Environmental Engineering 2004;130(6):605-614. |
R830883 (2003) R830883 (2004) |
Exit |
|
McMahon G, Alexander RB, Qian S. Support of total maximum daily load programs using spatially referenced regression models. Journal of Water Resources Planning and Management 2003;129(4):315-329. |
R830883 (2003) R830883 (2004) |
Exit |
|
Qian SS, Reckhow KH, Zhai J, McMahon G. Nonlinear regression modeling of nutrient loads in streams: a Bayesian approach. Water Resources Research 2005;41;W07012, doi:10.1029/2005WR003986. |
R830883 (2004) R830883 (2005) |
Exit |
Supplemental Keywords:
water, watersheds, risk, integrated assessment, ecological effects, Bayesian, modeling, southeast, GIS,, RFA, Economic, Social, & Behavioral Science Research Program, Scientific Discipline, Water, Ecosystem Protection/Environmental Exposure & Risk, Water & Watershed, Monitoring/Modeling, Regional/Scaling, Environmental Monitoring, decision-making, Ecology and Ecosystems, Economics & Decision Making, Watersheds, risk assessment, ecosystem modeling, aquatic ecosystem, watershed, ecology, ecosystem assessment, Bayesian approach, decision analysis, decision making, environmental decision making, ecological variation, TMDL, regional scale impacts, water quality, assessment endpoint mechanistic research, ecological indicators, ecology assessment models, ecosystem stress, watershed assessment, ecological models, decision support tool, environmental risk assessment, Bayesian classifiers, water monitoringProgress and Final Reports:
Original AbstractThe perspectives, information and conclusions conveyed in research project abstracts, progress reports, final reports, journal abstracts and journal publications convey the viewpoints of the principal investigator and may not represent the views and policies of ORD and EPA. Conclusions drawn by the principal investigators have not been reviewed by the Agency.