You are here:
Evaluation of Statistical Models to Predict the Dynamics of Fecal Indicator Bacteria in Tropical Waters
Citation:
Orizondo Lugo, M., M. Molina, A. Tseng, M. Bemis, AND J. Santo Domingo. Evaluation of Statistical Models to Predict the Dynamics of Fecal Indicator Bacteria in Tropical Waters. Coastal and Estuarine Research Federation 2023 Conference, Portland, OR, November 12 - 16, 2023.
Impact/Purpose:
This presentation will address the applicability of a water quality rapid tool (predictive modeling) to tropical streams to support TMDL development and revision.
Description:
Until recently, microbial water quality (MWQ) in Puerto Rico was measured with fecal coliforms (FC) as the fecal indicator bacteria (FIB). During this time, over 200 total maximum daily loads (TMDLs) have been developed using culturable fecal bacteria (i.e., FC). In 2017, the local government switched to enterococci as the primary FIB for TMDL purposes. However, there is very little data on how enterococci might relate to previously developed TMDLs. This study aims to (1) develop statistical models to predict FIB levels using culture-dependent (Enterolert) and culture-independent (quantitative polymerase chain reaction, qPCR) methods for MWQ assessments, (2) apply such developed models to historical data generated by USGS (nwis.waterdata.usgs.gov) to retroactively predict FIB levels and understand the impact of the new indicator on the current TMDLs. In addition, we will evaluate and compare urban and rural models for predicting the dynamics of fecal indicator bacteria in tropical waters. To achieve these objectives, we processed water samples collected from various rural and urban basins in Puerto Rico (i.e, Rio Grande de Loiza Caguas, Rio Espiritu Santo Rio Grande, Rio Piedras Hato Rey, and Rio Grande de Patillas) and analyzed them for FIB levels using both culture-based methods and qPCR assays and for various physicochemical parameters. Statistical models were developed to predict FIB concentrations based on environmental variables and other relevant factors using Multi Linear Regression (MLR) and Gradient Boosting Machine (GBM) learning techniques. We were able to predict FIB concentrations using environmental parameters with a high level of accuracy with both MLR and GBM approaches. Average GBM model accuracies ranged from 65 to 82%, while MLR R2 ranged from 0.16 to 0.4. The best model predictions were produced for FC when combined with enterococci. Temperature, pH, turbidity, dissolve oxygen, and conductivity were all important parameters for FIB prediction, but their influence depended on the type of watershed (urban vs rural). Our results indicate that these rivers are constantly exceeding local water quality criteria. Further studies are needed to identify the primary sources of fecal pollution to better comply with TMDL guidelines.
