Prediction of swimmability in a brackish water body

Abstract

Purpose

The primary objective of this study is to develop a predictive model that will predict the swimmability of certain areas of a brackish water body (Lake Pontchartrain) based on physicochemical and meteorological conditions.

Design/methodology/approach

Samples were collected and analyzed for bacteria indicator organisms at 13 sites along and adjacent to Lincoln Beach for four years. Physicochemical parameters and meteorological data were also recorded. A logistic regression model and an artificial neural networks (ANNs) model were both used to predict whether a lake condition is “safe to swim” or “not safe to swim”, given only physicochemical and meteorological parameters.

Findings

Both models predicted very well the results observed when lake conditions were “safe to swim” (97.7 percent of time the statistical model predicted correctly and an average >99.5 percent of the time for the ANNs model). However, for conditions under which the lake water quality was “not safe to swim”, the statistical model predicted correctly only 5.6 percent of the time. The ANNs model successfully predicted the “not safe to swim” conditions for an average 98.5 percent of the time. However, this percentage decreases to 53.9 percent when ANNs is used for forecasting “not safe to swim” conditions.

Research limitations/implications

The poorer performance of both models for “not safe to swim” conditions is probably due to the fact that most data (84.5 percent) were collected during “safe to swim” conditions. The limited database for “not safe to swim” conditions resulted in a poorer forecasting success rate.

Originality/value

The ANNs model might serve as a useful tool for public beach management with increased data on “not safe to swim” conditions.

Citation

Jin, G. and Englande, A.J. (2006), "Prediction of swimmability in a brackish water body", Management of Environmental Quality, Vol. 17 No. 2, pp. 197-208. https://doi.org/10.1108/14777830610650500