Purpose: The purpose of this study was to investigate (i) how well lognormal distributions fit empirical data; (ii) how estimates of microbial criteria are affected by detection limits; (iii) if 20 samples are sufficient to characterize the water quality; (iv) how sensitive the microbial criteria are to shifts in water quality; and (v) the predictive ability of Escherichia coli for the presence of Salmonella spp.
Methods: This study used 540 samples from six irrigation ponds measuring E. coli concentrations, Salmonella spp. presence, turbidity, and other physicochemical parameters. Objectives were analyzed by (i-ii) fitting distributions to data using maximum likelihood estimation, while considering censoring; (iii) analyzing data subsets to simulate limited sampling; (iv) simulating shifts in water quality using generated data and measuring the time until microbial criteria reflect the shift; (v) using logistic regression.
Results: Lognormal distributions provided an adequate fit and accounting for censoring due to detection limits increased the spread of fitted distributions. Due to high variability in E. coli counts, 20 samples were not sufficient to characterize the water quality and sudden shifts in water quality were not detected using the prescribed sampling scheme for as many as six years. Escherichia coli was found to be an adequate predictor of Salmonella spp. presence, with turbidity as an additional significant variable.
Significance: When bacteriological quality of irrigation ponds has high variability, as in this study, alternative approaches to ensuring water quality should be considered.