Regulatory bodies that monitor community water systems (CWS) can be strained for time and resources, making it difficult to take science-informed preventative action before contamination affects local populations. This research aims to develop an easy to deploy, data-driven method to identify CWS that are vulnerable to drinking water quality degradation from groundwater contamination. We focus on a case study of Los Angeles County (LA County) in California to explore the utility of machine learning methods, specifically random forest models (RF) and artificial neural networks (ANN), as quickly deployable decision support tools based on publicly available data. We also aim to provide insight into which factors contribute to vulnerability to groundwater contamination, which may be useful in understanding how vulnerability to contamination emerges. The results of this analysis, which are based entirely on publicly available data, can help policymakers and planners target specific systems, as well as better tailor the type of support needed. We find that both RF and ANN methods can produce relatively low prediction errors but differ in what they predict and how they weigh the relative importance of input variables. The results also suggest that model results can provide stakeholders with a starting point for prioritizing at-risk service areas, but it is important to remember that the model results are most useful in combination with expert opinion.

监测社区供水系统 (CWS) 的监管机构可能会因时间和资源紧张而难以在污染影响当地人口之前采取科学预防措施。本研究旨在开发一种易于部署的数据驱动方法，以识别易受地下水污染导致饮用水质量下降的水煤浆。我们专注于加利福尼亚州洛杉矶县 (LA County) 的案例研究，以探索机器学习方法的效用，特别是随机森林模型 (RF) 和人工神经网络 (ANN)，作为基于公开可用的快速部署决策支持工具数据。我们还旨在深入了解哪些因素会导致地下水污染的脆弱性，这可能有助于理解污染脆弱性是如何出现的。这种完全基于公开数据的分析结果可以帮助政策制定者和规划者针对特定系统，以及更好地定制所需的支持类型。我们发现 RF 和 ANN 方法都可以产生相对较低的预测误差，但它们的预测内容不同，以及它们如何权衡输入变量的相对重要性。结果还表明，模型结果可以为利益相关者提供一个起点来确定有风险的服务领域的优先级，但重要的是要记住，模型结果与专家意见相结合最有用。我们发现 RF 和 ANN 方法都可以产生相对较低的预测误差，但它们的预测内容不同，以及它们如何权衡输入变量的相对重要性。结果还表明，模型结果可以为利益相关者提供一个起点来确定有风险的服务领域的优先级，但重要的是要记住，模型结果与专家意见相结合最有用。我们发现 RF 和 ANN 方法都可以产生相对较低的预测误差，但它们的预测内容不同，以及它们如何权衡输入变量的相对重要性。结果还表明，模型结果可以为利益相关者提供一个起点来确定有风险的服务领域的优先级，但重要的是要记住，模型结果与专家意见相结合最有用。