Utilizing Remote Sensing and Deep Neural Networks to Predict Biochemical Oxygen Demand in the Chesapeake Bay

Deteriorating water quality and the rising demand for clean water necessitate more efficient water quality monitoring. One key water quality metric is five-day biochemical oxygen demand (BOD5). Current methods of BOD5 measurement are expensive and time consuming, requiring a more cost-effective method of BOD5 measurement. In our study, we investigated machine learning models commonly used for water quality prediction and created a novel Deep Neural Network (DNN) capable of predicting BOD5 based on water quality metrics measured from satellites. Benchmarking our models on their Proportional Root Mean Square Error or RMSEP, we found that the DNN exceeded expectations by attaining a low RMSEP of 0.022 and outperforming commonly used machine learning models. In addition, the model showed no geographical bias after being tested with zones of varying salinity levels, suggesting that DNNs have significant potential for predicting BOD5 with data obtained solely from remote satellites.