A support vector regression model to predict nitrate-nitrogen isotopic composition using hydro-chemical variables.

Area of Science:

• Environmental Science
• Water Chemistry
• Geochemistry

Background:

• Nitrate is a major water pollutant globally.
• Isotopic analysis of nitrate aids in source tracing and transformation studies.
• Current isotopic analysis methods are time-consuming and expensive.

Purpose of the Study:

• To develop and assess a Support Vector Regression (SVR) model for predicting nitrogen isotopic composition of nitrate (δ¹⁵N-NO₃^-).
• To explore an alternative, cost-effective method for estimating nitrate isotopes in aquatic systems.
• To enhance water quality management through indirect prediction of environmental isotope values.

Main Methods:

• Utilized 16 hydro-chemical variables from a rural-urban river system.
• Employed Principal Component Analysis (PCA) for variable reduction.
• Optimized SVR model parameters using grid search and V-fold cross-validation.

Main Results:

• The SVR model achieved high prediction accuracy (R²=0.88, NS=0.87) for δ¹⁵N-NO₃^-.
• The SVR model significantly outperformed multivariate linear regression and neural network models.
• Accurate predictions were obtained even with reduced datasets and PCA-processed variables.

Conclusions:

• The SVR model offers a feasible and accurate indirect method for predicting nitrate's nitrogen isotopic composition.
• This approach can complement direct isotopic measurements, reducing time and cost.
• The SVR model shows potential for effective water quality management and pollution source identification.