Science Inventory

INTRODUCTION TO A COMBINED MULTIPLE LINEAR REGRESSION AND ARMA MODELING APPROACH FOR BEACH BACTERIA PREDICTION

Citation:

GE, Z. AND W. E. FRICK. INTRODUCTION TO A COMBINED MULTIPLE LINEAR REGRESSION AND ARMA MODELING APPROACH FOR BEACH BACTERIA PREDICTION. Presented at IAGLR's 50th Annual Conference on Great Lakes Research, University Park, PA, May 28 - June 01, 2007.

Impact/Purpose:

A main objective of this task is to combine empirical and physical mechanisms in a model, known as Visual Beach, that

  • is user-friendly
  • includes point and non-point sources of contamination
  • includes the latest bacterial decay mechanisms
  • incorporates real-time and web-based ambient and atmospheric and aquatic conditions
  • and has a predictive capability of up to three days to help avert potential beach closures.
The suite of predictive capabilities for this software application can enhance the utility of new methodology for analysis of indicator pathogens by identifying times that represent the highest probability of bacterial contamination. Successful use of this model will provide a means to direct timely collection of monitoring samples, strengthening the value of the short turnaround time for sampling. Additionally, in some cases of known point sources of bacteria, such as waste water treatment plant discharges, the model can be applied to help guide operational controls to help prevent resulting beach closures.

Description:

Due to the complexity of the processes contributing to beach bacteria concentrations, many researchers rely on statistical modeling, among which multiple linear regression (MLR) modeling is most widely used. Despite its ease of use and interpretation, there may be time dependences in the observations that cannot be explained by the MLR model, so that the residuals are not as random as they are assumed to be. In this case, an ARMA (auto-regressive moving average) model can be used to extract the possible deterministic time patterns from the MLR residuals. The ARMA-modeled deterministic part of the residual is then added to the MLR predictions as an adjustment, and the variance of the prediction errors can be considerably reduced. The whole modeling process is demonstrated with actual data from Huntington Beach, Ohio, in 2000-2004. Results show that the predictive capacity of the initial MLR model is significantly improved by making use of the supplemental ARMA technique. Supplemental ARMA modeling is an independent step that does not otherwise affect the existing MLR models, another attractive feature of this approach.

Record Details:

Record Type: DOCUMENT (PRESENTATION/ABSTRACT)
Product Published Date: 05/29/2007
Record Last Revised: 02/12/2007
Record ID: 163386

Organization:

U.S. ENVIRONMENTAL PROTECTION AGENCY

OFFICE OF RESEARCH AND DEVELOPMENT

NATIONAL EXPOSURE RESEARCH LABORATORY

ECOSYSTEMS RESEARCH DIVISION

REGULATORY SUPPORT BRANCH