Multivariate prediction of nitrogen concentration in a stream using regression models

Total Kjeldahl Nitrogen (TKN) is an important parameter in the analysis of water quality since its concentration level outside established ranges can lead to serious health problems and endanger aquatic ecosystems. Measuring TKN is a tedious and complicated task because it requires different procedures, specific equipment, and trained personnel to obtain the information. This work aims thus to estimate TKN from physicochemical parameters that can be easily measured in water. A correlation analysis between the parameters was performed to assess their associations and select the most relevant predictors to regression models. Three regression methods were used to estimate the nitrogen concentration from the data, namely multiple linear regression, regression trees, and support vector regression. Total alkalinity, chlorides, color, conductivity, total hardness, nitrates, dissolved oxygen, pH, total solids and temperature were the input variables to the models. The prediction was assessed using absolute root mean square error (RMSE), mean absolute error (MAE), and R-squared (R²). The best TKN prediction was achieved using regression trees (RMSE = 0.29, MAE = 0.13 and R² = 0.84). This result shows that it is possible to estimate such a difficult to measure but important parameter from parameters that can be measured more easily and with lower production of hazardous waste, which represents an advantage for water quality analysis in remote and hard-to-reach places.