Hydrometeorological variables predict fecal indicator bacteria densities in freshwater: data-driven methods for...

Area of Science:

• Environmental science
• Water quality modeling
• Statistical ecology

Background:

• Microbial water quality monitoring is crucial for recreational safety.
• Current methods rely on extensive, site-specific data, posing cost challenges for risk managers.
• Developing efficient predictive models is essential for managing numerous water recreation sites.

Purpose of the Study:

• To evaluate data-driven models (tree regression, random forests) for selecting key hydrometeorological variables.
• To integrate selected variables into linear mixed effects (LME) models for predicting bacterial density.
• To assess the feasibility and performance of this alternative approach compared to resource-intensive methods.

Main Methods:

• Applied tree regression and random forests with conditional inference trees to identify predictive variables.
• Utilized variable importance from random forests for forward-step selection in LME models.
• Collected seasonal Escherichia coli and enterococci data from Chicago Area Waterway System and Lake Michigan (2007-2009).

Main Results:

• Tree regression reduced data dimensionality by over 50%.
• Two to three selected variables effectively predicted bacterial densities.
• LME models with selected variables showed reasonable performance (R² 0.335–0.658).
• Lake Michigan models achieved good prediction accuracy (72–77%) for the single sample maximum standard, with variable sensitivity (23–62%).

Conclusions:

• The proposed data-driven approach is feasible for predicting microbial water quality.
• This method offers a cost-effective alternative to resource-intensive water quality modeling.
• The findings support the use of readily available hydrometeorological data for risk management.