Basin-wide tracking of nitrate cycling in Yangtze River through dual isotope and machine learning

Fazhi Xie; Gege Cai; Guolian Li; Haibin Li; Xing Chen; Yun Liu; Wei Zhang; Jiamei Zhang; Xiaoli Zhao; Zhi Tang

doi:10.1016/j.scitotenv.2023.169656

Basin-wide tracking of nitrate cycling in Yangtze River through dual isotope and machine learning

Sci Total Environ. 2024 Feb 20:912:169656. doi: 10.1016/j.scitotenv.2023.169656. Epub 2023 Dec 27.

Authors

Fazhi Xie¹, Gege Cai², Guolian Li¹, Haibin Li², Xing Chen¹, Yun Liu¹, Wei Zhang³, Jiamei Zhang⁴, Xiaoli Zhao⁵, Zhi Tang⁵

Affiliations

¹ School of Environmental and Energy Engineering, Anhui Jianzhu University, Hefei 230031, Anhui, China.
² School of Materials and Chemical Engineering, Anhui Jianzhu University, Hefei 230031, Anhui, China.
³ Institute of Intelligent Machines, Chinese Academy of Sciences, Hefei 230031, Anhui, China.
⁴ School of Environmental and Energy Engineering, Anhui Jianzhu University, Hefei 230031, Anhui, China. Electronic address: zjm1126@ustc.edu.cn.
⁵ Chinese Research Academy of Environmental Sciences, Beijing 100000, China.

PMID: 38157890
DOI: 10.1016/j.scitotenv.2023.169656

Abstract

The nitrate (NO₃^-) input has adversely affected the water quality and ecological function in the whole basin of the Yangtze River. The protection of water sources and implementation of "great protection of Yangtze River" policy require large-scale information on water contamination. In this study, dual isotope and Bayesian mixing model were used to research the transformation and sources of nitrate. Chemical fertilizers contribute 76 % of the nitrate sources in the upstream, while chemical fertilizers were also dominant in the midstream (39 %) and downstream (39 %) of Yangtze River. In addition, nitrification process occurred in the whole basin. Four machine learning models were used to relate nitrate concentrations to explanatory variables describing influence factors to predict nitrate concentrations in the whole basin of Yangtze River. The anthropogenic and natural factors, such as rainfall, GDP and population were chosen to take as predictor variables. The eXtreme Gradient Boosting (XGBoost) model for nitrate has a better predictive performance with an R² of 0.74. The predictive models of nitrate concentrations will help identify the nitrate distribution and transport in the whole Yangtze River basin. Overall, this study represents the first basin-wide data-driven assessment of the nitrate cycling in the Yangtze River basin.

Keywords: Dual isotope; Machine learning; Nitrate; SIAR calculation; Yangtze River.