Last year, MVPC received funding as part of the Merrimack River Clean-up Initiative, providing financial backing for clean-up events and trash collection in the river. As part of this grant, we also received additional funding to further progress the Early Alert Tool, a piece of machine-learning based technology intended to provide a look into the Merrimack River’s fecal bacterial load.

Fecal bacteria is a contaminant introduced from many sources, including natural ones like waterfowl and riverside wildlife. However, it can also come from unnatural sources, like improperly disposed pet waste. The largest single source of fecal bacteria can be traced to combined sewer overflow (CSO) events. These events are when raw sewage, mixed with stormwater and runoff, is discharged into the river as a side-effect of combined sewer systems.

Most modern cities are built today with separate stormwater and sewer systems. The cities along the Merrimack River, dating to the industrial era and before, are instead burdened with the relics of those times – a combined sewer and stormwater system. Both runoff from rainstorms and sewage from homes and businesses flow into the same pipes. In most situations, this does not cause an issue, but in heavy rainstorms, these pipes can be overwhelmed and wastewater plants cannot keep up. To avoid sewage backing up into homes and businesses, there are relief pipes that instead discharge it into the nearest waterbody. In our case, this is the Merrimack and its tributaries.

While separation of these sewers and stormwater systems is ongoing, it is exceptionally costly and time consuming. In the meantime, we have invested time and research into a tool that may help the public understand the effects of CSOs and judge their ability to interact with the river safely. Using fecal bacterial sampling data collected and provided by the Merrimack River Watershed Council over the past four years, combined with publicly-available CSO information from our local wastewater managers, and other information like weather and river conditions, we have been able to produce a machine learning model that provides a prediction of the current bacteria level in the river.

This is the third iteration of this model and the second using machine learning technology. The prior model, built last year in partnership with GEI Consultants, provided results that were not as accurate as we hoped. This year, we once again partnered with GEI using a new approach that we feel is far more accurate, providing better results that we hope can be a part of the public’s toolkit.