A machine learning model for real-time total nitrogen monitoring in Tampa Bay

Total Nitrogen (TN) concentration is a critical water quality indicator for the health of Tampa Bay, as excessive nitrogen loading is a primary driver of eutrophication, harmful algal blooms, and the degradation of seagrass habitats. TN has been monitored through monthly water sampling since the 1970s along with other water quality variables. While the traditional monitoring provides high-quality data, it often misses rapid environmental changes due to the low sampling frequency. Conversely, autonomous sensors on the Tampa Bay Observation Network (TBON) provide high-temporal-resolution data but lack the capability to measure TN directly. To bridge this gap, we developed an artificial neural network (ANN) machine learning model using historical data from 2000 to 2020. We then applied it to high-temporal-resolution, sensor-measurable water quality variables (i.e., temperature, salinity, and chlorophyll-a) for TN estimation. The model performance was evaluated in two ways: (1) a random split of historical data was kept independent from model development and used strictly for testing, yielding a Pearson correlation coefficient (R) of 0.82 and a root mean square error (RMSE) of 0.07 mgN/L (number of data points n = 95); and (2) model results were compared with independent field TN observations from 2023 and 2024, yielding an RMSE of 0.05 mgN/L (n = 7). Importantly, the model successfully captured seasonal patterns and significant TN spikes during hurricane events. This modeling framework, combined with the real-time TBON system, provides a viable solution for high-temporal-resolution nitrogen monitoring, offering a useful tool for water quality management.