The purpose of this study is to develop a deep learning algorithm for conducting water quality monitoring of chemical monitoring sites in Baltimore City and Baltimore County watersheds. In Maryland, jurisdictions including Baltimore City and Baltimore are required to monitor and compile an annual report of local water quality and stormwater management efforts as part of the Municipal Separate Storm Sewer System (MS4) permit under the National Pollutant Discharge Elimination System (NPDES). Ultimately the goal is to improve the health of the Chesapeake Bay as a part of the Chesapeake Bay Cleanup Plan. Two elements of the Cleanup Plan are adhering to strict water quality standards and implementing best management practices to reduce nutrient pollution. Comprehensive monitoring data can be used to determine the efficacy of best management practices and the health of local water bodies. However, high-frequency monitoring of these sites is prohibitively expensive. In these cases, machine learning and deep learning techniques can be used to substitute for a number of in-person site visits and develop data models directly from available surrogate data. In water resources engineering, simulations, models, and evaluations derived from machine learning can be applied to provide solutions to surface water pollution control, identification of major sources of pollution, water quality improvement, and watershed management among some of the uses. In this study, deep learning algorithms were used to predict post-storm streamflow, total nitrogen, total phosphorous, total suspended solids, total bacterial species, and heavy metals in the Powder Mill Subwatershed of the Gwynns Falls Watershed.
