Reimagining the Kendall plot: using <i>δ</i> ¹⁵ N and <i>δ</i> ¹⁸ O of nitrate and advanced machine learning to improve N pollutant source classification

Nitrate (<mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:msubsup><mml:mrow><mml:mi>NO</mml:mi></mml:mrow><mml:mn>3</mml:mn><mml:mo>-</mml:mo></mml:msubsup></mml:math>) pollution is a serious water quality issue in many countries due to contamination of lakes, rivers, and aquifers by intensive agriculture practices and inadequate wastewater management. Nitrate pollution and associated cultural eutrophication are anticipated to increase worldwide, highlighting the need to control and reduce nitrogen pollution. The stable isotope ratios of nitrate (<i>δ</i>¹⁵N, <i>δ</i>¹⁸O) are widely used as tracers of nitrogen pollution sources. The primary technique for identifying nitrate sources has been the longstanding Kendall boxplot, despite improved methods using Bayes' theorem and the R language for estimating source fractions using hydrogeochemical context, N source data and expert assessment. This article improves the classification of aqueous nitrate sources using comprehensive published stable isotope data for nitrate from four known pollutant types and applying machine learning algorithms. AI modelling results reveal improved source depictions and offer a robust statistical framework for identifying N pollution sources. This is essential given the increased published data sources and the need for better-informed water quality management strategies.