Comparison of various estimation techniques to predict nitrate load in Maumee River

Water quality issues related to hypoxia are prominent in water bodies in the United States (US) and all over the world. Nutrient loads in surface water can be estimated by monitoring their concentrations in the water bodies. The challenge is to determine the optimal data sampling frequency and the best method for the estimation of nutrient concentrations that minimizes the uncertainty caused by data and modelling errors in the estimated data. For that purpose, researchers have used various regression models such as LOADEST, WRTDS, linear regression and error minimization techniques over the years. However, there is no consensus in the scientific community on which method or sampling frequency is best suited to different scenarios of water quality data. This analysis seeks to establish a scientific agreement by comparing the performances of different estimation techniques on a complete dataset. This study compared the accuracy of LOADEST, WRTDS, and linear interpolation methods to predict daily nitrate concentration at four different sampling frequencies using Maumee River (Ohio) dataset collected for over 30. The results show that a higher data sampling frequency gives better results for linear interpolation, but not necessarily for LOADEST and WRTDS. The bias for WRTDS method is significantly lower than for linear interpolation and LOADEST. However, the RMSE and R² values for LOADEST and linear interpolation models are comparable, and significantly better than for WRTDS.