道路网络是现代社会的重要组成部分,通过使货物和人员平稳流动,促进快速运输和推动经济活动。然而,道路系统的扩展具有重要的环境因素,特别是它对地下水质量的影响。因此,了解地下水水质与道路交通系统之间的复杂关系至关重要。本文旨在使用数据驱动的方法来确定道路运输系统对地下水质量的影响。具体来说,德克萨斯州的道路网和地下水化学数据是从一个开放的数据门户获得的。这项研究分两个阶段进行:可解释的人工智能(XAI)建模阶段和多变量分析阶段。在XAI建模阶段,使用极限梯度提升(XGB)开发了预测模型,以地下水化学参数为输出特征,公路运输属性为输入特征,即,高程,年平均每日交通量,距离,车道英里,速度限制和井深。此外,使用特征重要性和累积局部效应(ALE)检查了地下水化学参数与公路运输属性之间的关系。在多元阶段,Piper图和主成分分析(PCA)用于从XAI模型中确定选定的地下水化学参数的来源。预测模型的结果表明,5个地下水化学参数受到公路运输系统的显著影响,平均绝对百分比误差低于0.20,包括,pH值,温度,铝(Al),碳酸氢盐(HCO3-),和碱度。此外,开发了XAI模型,以了解五个选定参数上道路运输属性之间的关系。研究结果共同表明,德克萨斯州的地下水质量受到50米距离和100米井深范围内的道路运输系统的极大影响。这项研究为使用XAI技术监测地下水污染的点源提供了新的贡献。
Road networks constitute a vital component of modern society, facilitating rapid transportation and driving economic activities by enabling the smooth movement of goods and people. However, the expansion of road systems carries significant environmental considerations, particularly regarding its impact on groundwater quality. Thus, it is crucial to understand the complex relationship between groundwater quality and the road traffic system. This paper aims to identify the impact of road transport systems on groundwater quality using a data-driven approach. Specifically, road network and groundwater chemistry data in Texas were obtained from an open data portal. This study was carried out in two phases: the explainable artificial intelligence (XAI) modeling phase and the multivariate analysis phase. In the XAI modeling phase, a prediction model was developed using eXtreme Gradient Boosting (XGB), with groundwater chemistry parameters as output features and road transport attributes as input features, i.e., elevation, annual average daily traffic, distance, lane-miles, speed limit and well depth. Furthermore, the relationships between groundwater chemistry parameters and road transport attributes were examined using feature importance and accumulated local effect (ALE). In the multivariate phase, Piper diagrams and principal component analysis (PCA) were utilized to identify the source of the selected groundwater chemistry parameters from the XAI models. The results of the prediction model showed that five groundwater chemistry parameters were significantly impacted by road transport systems with below a mean absolute percentage error of 0.20, including, pH, temperature, aluminum (Al), bicarbonate (HCO3-), and alkalinity. Additionally, XAI models were developed to understand the relationship between the road transport attributes on five selected parameters. The findings collectively indicated that the Texas groundwater qualities are greatly impacted by road transport systems within a distance of 50-meters and a well depth of 100-meters. This study provides a novel contribution to monitoring point sources of groundwater pollution using XAI techniques.