A nonlinear autoregressive exogenous (NARX) model to predict nitrate concentration in rivers

Area of Science:

• Environmental science
• Water resource management
• Computational hydrology

Background:

• Nitrate contamination in rivers poses risks to ecosystems and drinking water supplies.
• Accurate forecasting of nitrate levels is vital for effective water management and treatment strategies.

Purpose of the Study:

• To evaluate the efficacy of nonlinear autoregressive with exogenous inputs (NARX) neural networks for predicting nitrate concentrations in rivers.
• To assess the impact of various hydrological and water quality parameters as exogenous inputs on prediction accuracy.
• To determine the optimal time series length and input variable selection for accurate nitrate forecasting.

Main Methods:

• Utilized NARX neural network models to forecast nitrate plus nitrite concentrations.
• Selected the Susquehanna River and Raccoon River in the USA as case study locations.
• Incorporated water discharge, temperature, dissolved oxygen, and specific conductance as exogenous variables.

Main Results:

• NARX models demonstrated high accuracy in predicting nitrate concentrations for both studied rivers.
• For Kreutz Creek, prediction accuracy (R²=0.77) improved with the inclusion of all exogenous variables.
• For the Raccoon River, high accuracy (R²=0.94) was achieved using only water discharge as an input.

Conclusions:

• NARX neural networks provide a robust and accurate method for forecasting riverine nitrate concentrations.
• The selection and number of exogenous input variables significantly influence prediction performance.
• The developed models are suitable for both short-term and long-term nitrate forecasting, supporting water quality management.