The US Gulf of Mexico Coast is highly susceptible to extreme storm events which result in frequent flooding of urban areas and increased potential for enteric pathogenic contamination in surface waters. Unfortunately, coastal flooding events are expected to increase due to hurricanes becoming more frequent and severe. Due to the risks of sampling during severe weather conditions and the time required to process water samples for microbial analysis using lab-based methods, implementation of methodologies that allow rapid determination of microbial water quality are highly needed. In this study, we explore predictive modeling as a rapid alternative methodology to accurately predict fecal indicator bacteria (FIB) both under normal conditions and in the aftermath of tropical storms or hurricanes.
We applied Gradient Boosting Machine (GBM) and Multiple Linear Regression (MLR) methods to weekly measurements of enterococci concentration and other water quality parameters collected from the years 2015-2023 at 6 sites in the Lake Pontchartrain estuary, New Orleans, LA. Three sites were located on the highly urbanized south shore, while three were on the more rural north shore. We compared water quality measurements collected in the weeks following storm events with baseline data and generated separate models for the north and south shores.
We observed increased enterococci concentrations during the days following storms with significant rainfall, which returned to initial levels approximately 2 weeks after a storm. GBM results indicated that the best predictors of enterococci levels were dissolved oxygen (DO) and water temperature on the north shore, and turbidity and precipitation on the south shore. We found that GBM methods outperformed MLR, with prediction accuracies of ~0.9 for both shores, compared to MLR adjusted R2 values of ~0.3. Future investigations into models incorporating fate and transport, and resuspension, could yield further progress developing rapid and accurate water quality prediction methods.