Assessing Bacteria Level with Machine Learning (ML) Models in the Neches River, Texas

Bacteria released from the on-site sewer facilities (OSSFs) elevated fecal bacteria indicators in the Neches River Tidal with increasing flood frequency and intensity. It is necessary to monitor the bacteria levels on time as elevated fecal bacteria bring a public health concern. The aim of the study is to collect water samples after rain events at hotspots of OSSFs to assess the bacteria impacts, apply innovative lab technology to track fecal bacteria sources from the selected water samples, and develop the classification ML model utilizing random forest (RF) for estimating bacteria levels in the Neches River Tidal to assist water monitoring. The field and lab measurements of the water sample showed higher Escherichia coli at hotspots of OSSF increased the bacteria level in the river, especially in winter. The RF determined bacteria levels in river depending on pH, temperature, dissolved oxygen (DO), alkalinity, phosphorus, sulfate, conductivity, chloride, total suspended solid (TSS), total nitrogen (TN), Secchi-Depth, ammonia-N, and flow rate from five monitoring stations. RF achieved accuracy=0.74, precision=0.79, recall=0.86, and F1-score=0.83 with less overfitting, superior to the Decision Tree (DT) model. Flow rate was identified as the most important feature. The probability of elevated bacterial level significantly increased when the flow rate was greater than 250 m3/s although pH, temperature, and DO were also key biochemical factors. In conclusion, fixing failed OSSFs or applying green infrastructures are recommend approaches to eliminate the nonpoint source and decrease the bacteria level in Neches River Tidal.