Huang Zipeng, He Baonan, Chu Yanjia, Song Yuanbo, Shen Zheng
Key Laboratory of Groundwater Conservation of MWR, China University of Geosciences, Beijing, P. R. China.
Institute of New Rural Development, School of Electronics and Information Engineering, Tongji University, Shanghai, China.
Water Environ Res. 2025 Feb;97(2):e70033. doi: 10.1002/wer.70033.
Groundwater, a pivotal water resource in numerous regions worldwide, confronts formidable challenges posed by severe nitrate pollution. Traditional research methodologies aimed at addressing groundwater nitrate contamination frequently struggle to accurately depict the intricate conditions of the groundwater environment, particularly when dealing with high variability and nonlinear data. However, the advent of machine learning (ML) has heralded an innovative approach to simulating groundwater dynamics. In this study, six ML algorithms were deployed to model the concentrations of shallow groundwater nitrates in the Shaying River Basin. The efficacy of each model was assessed through comprehensive metrics including the coefficient of determination (R), mean absolute error (MAE), and root mean square error (RMSE), gauging the alignment between observed and predicted groundwater nitrate levels. Subsequently, to discern the principal environmental factors influencing NO-N concentrations, the most proficient model was selected. Among the array of models, the XGB algorithm, renowned for its capacity to handle extreme values, demonstrated superior performance (R = 0.773, MAE = 7.625, RMSE = 11.92). Through an in-depth analysis of groundwater NO-N across major urban centers, Fuyang city was identified as the most heavily contaminated locale, attributing the phenomenon to potential sources such as domestic sewage and agricultural activities (feature importance of Cl = 78.64%). Conversely, Zhengzhou city emerged as the least polluted city, with notable influences from K and NO (feature importance = 52.06% and 18.41%), indicative of a prevailing reducing environment compared to other cities. In summation, this study explores a methodology for amalgamating diverse environmental variables in the investigation of groundwater contamination. Such insights hold profound implications for the effective management and mitigation of nitrate contamination in the Shaying River Basin, offering a demonstration for similar endeavors in analogous regions. PRACTITIONER POINTS: Six machine learning models were utilized to simulate the nitrate contamination. XGB model for groundwater nitrate pollution prediction outperformed other models. Relative importance of environmental variables was identified using the XGB model. Impact of main environmental variables on groundwater nitrate was discussed.
地下水是全球众多地区的关键水资源，面临着严重硝酸盐污染带来的巨大挑战。旨在解决地下水硝酸盐污染问题的传统研究方法，在准确描绘地下水环境的复杂状况时常常困难重重，尤其是在处理高变异性和非线性数据时。然而，机器学习（ML）的出现开创了一种模拟地下水动态的创新方法。在本研究中，部署了六种机器学习算法来模拟沙颍河流域浅层地下水硝酸盐的浓度。通过包括决定系数（R）、平均绝对误差（MAE）和均方根误差（RMSE）在内的综合指标评估每个模型的有效性，衡量观测到的和预测的地下水硝酸盐水平之间的一致性。随后，为了识别影响NO-N浓度的主要环境因素，选择了最有效的模型。在一系列模型中，以处理极值能力著称的XGB算法表现出卓越性能（R = 0.773，MAE = 7.625，RMSE = 11.92）。通过对主要城市中心的地下水NO-N进行深入分析，阜阳市被确定为污染最严重的地区，将该现象归因于生活污水和农业活动等潜在来源（Cl的特征重要性 = 78.64%）。相反，郑州市成为污染最轻的城市，受K和NO 影响显著（特征重要性分别为52.06%和18.41%），表明与其他城市相比存在普遍的还原环境。总之，本研究探索了一种在地下水污染调查中整合各种环境变量的方法。这些见解对沙颍河流域硝酸盐污染的有效管理和缓解具有深远意义，为类似地区的类似努力提供了示范。从业者要点：利用六种机器学习模型模拟硝酸盐污染。用于预测地下水硝酸盐污染的XGB模型优于其他模型。使用XGB模型确定了环境变量的相对重要性。讨论了主要环境变量对地下水硝酸盐的影响。
