2019 - Real-time responsive nutrient loading management in urban catchments through sewer-embedded sensing and controls

Grant Award Year: 
Principal Investigator: 
Amy Mueller
Edward Beighley
Research Description: 
The Charles River, which receives stormwater runoff from the City of Cambridge, is impaired by eutrophication. To meet Total Maximum Daily Load (TMDL) allocations for phosphorous, the City is executing its Long-Term Control Plan which includes separating sanitary and stormwater sewers, however structural and non-structural Best Management Practices are still needed to treat stormwater prior to its entering receiving water. To achieve this, flow controls are being considered to selectively direct (ideally high P) portions of stormwater to the Deer Island Wastewater Treatment Plant, but this raises two key questions: (i) which fraction(s) of stormwater flows should be diverted to the WWTP, given limits on how much additional flow the WWTP can handle, and (ii) how can a real-time control system be implemented to automatically manage this process? Previously best understanding indicated that P loads are maximum during a storm’s 'first flush' and associated with a specific particle size fraction, conditions which could be selected for by measuring velocity. However, preliminary high-resolution data collection conducted by this team in 2018 over a number of storm events has shown this is not universally true, geographically or temporally, and therefore a more complete understanding and control strategy is needed. In response, this project envisions a pathway to real-time estimation of P flux in storm waters through (1) installation of real-time sensing modalities for estimation of water chemistry at storm sewer diversion points and (2) using these data to calibrate both hydraulic and machine learning models to identify reliable real-time predictors of P flux. While an obvious next step, achieving these goals requires overcoming the lack of commercial real-time P sensors and difficulties/costs associated with data collection during storm events (major contributors to P exports), which prevents a traditional machine learning 'big data' approach.