A defining characteristic of environmental science and engineering research is the complexity of the systems being investigated. This complexity and existence of many confounding variables changing concomitantly can make relationships between variables opaque and hinder mechanistic understanding of the system or system subprocess under investigation. Advances in sensor technologies, analytical instrumentation, and edge computing are providing the ability to collect and analyze massive and complex data sets comprised of many environmental system variables. This opens the door to improving our understanding of how environmental systems will respond to perturbations, how to design new materials, tools, and technologies for environmental management, or how to better understand the impact of complex environmental exposures (the exposome) on health. There are many environmentally relevant research areas where advances in data science, such as machine learning (ML) and artificial intelligence (AI), have been applied to large data sets to better decipher the complex relationships between system variables affecting system behaviors and provide new insights on understanding and solution development. The following are a few nonexclusive examples of applications where ML/AI approaches are likely to advance environmental engineering and science: analytical data sets, including 2D and 3D chromatography, nontarget mass spectrometry, remote sensing, or chemical data predicting toxicity pathways and adverse outcomes tracing sources and predicting environmental fates of pollutants analysis of “multi-omic” data sets to predict biological responses to perturbations, for example, microbiomes elucidating the “exposome” for a broad range of organisms and systems sorbent and catalyst development and optimization watershed-scale terrestrial and aquatic responses to perturbations such as deforestation and climate change influence of molecular scale materials selections on life cycle environmental impacts materials discovery and design using machine learning digital twins to represent the natural and built environments for prediction, optimization, and decision making text mining, image processing, video analysis of environmental literature, and data sets for information extraction and processing managing water and wastewater treatment and distribution, for example, minimizing disinfection byproduct (DBP) formation managing local and regional air quality based on desirable epidemiological outcomes integrated and systems-level analysis that blends environmental exposures with the corresponding internal biological responses predicting emissions of air pollutants and greenhouse gases using ML or AL predicting the impacts of climate change on environmental sustainability predicting the impacts of environmental policy alternatives new ML or AI methodologies developed for advancing atmospheric sciences or environmental chemistry There can be massive longitudinal data sets generated for each of these examples that ultimately can be analyzed using existing or novel machine learning methods to enable predictions of how a system will respond to future perturbations, predict the toxicity potential of a compound, or predict material properties that are most suitable for a selected function. Despite the “black box” nature of unsupervised machine learning approaches, careful examination of the algorithms can provide insight into controlling variables or processes, leading to more “biophysicochemically-informed” supervised machine learning approaches (or hybrid thereof) that leverage existing mechanistic underpinnings of the processes at play. Such approaches can ultimately be used to better understand complex environmental system behaviors, help find solutions to problems in environmental science and engineering that have been previously unsolvable, relate environmental exposures to effects on human health, and enable better policy decisions to protect the environment and its inhabitants. ES&T and ES&T Letters are focusing attention on this emerging and fast-moving area through an upcoming joint special issue on data-driven machine learning and artificial intelligence in environmental science and technology. Through a public call for papers, the journals are seeking contributions on ML and AI research studies that demonstrate the great potential of these approaches to improve our understanding of natural and engineered environmental systems, toward maintaining a healthy ecosystem or building a circular economy. Papers are desired that include either novel applications of data science/ML methodologies and approaches adapted for use in environmental data sets or that address knowledge gaps in an important environmental science and technology that were not approachable using standard analysis tools. Submissions to the special issue should demonstrate the “value added” of taking a ML or AI approach over existing approaches. Submissions should also ensure that the data sets are large and complex enough that ML approaches are necessary and robust, and researchers have to go beyond the “black box” of simple agnostic applications of existing algorithms to determine the “best one”. Ideally, these papers would glean insights into mechanistic underpinnings of the system being investigated. Lastly, we anticipate these papers will be model examples of ML and AI analyses on complex environmental data sets. Therefore, papers must facilitate reproducibility by adhering to FAIR data principles (https://publish.acs.org/publish/data_policy), demonstrate computational rigor (e.g., discuss model assumptions/limitations, data considerations, cross validation, model performance), and provide ML and AI models and data sets to readers through publicly available data repositories (https://publish.acs.org/publish/author_guidelines?coden=esthag#data_requirements). The deadline for submissions for both journals is 12th January 2023. To complement the planned special issue publication, the journals will also publish a joint virtual issue, showcasing a collection of already published content on this topic in both journals over recent years. ES&T and ES&T Letters look forward to receiving submissions on this important area of development. More information on article types and formatting of submissions are available on each journal’s author guideline pages available here: ES&T and ES&T Letters. This article has not yet been cited by other publications.

中文翻译：

---

推进环境科学、工程和技术的数据科学：ES&T 和 ES&T 快报中即将出现的特殊和虚拟问题

环境科学和工程研究的一个决定性特征是所研究系统的复杂性。许多混杂变量的复杂性和同时变化的存在会使变量之间的关系变得不透明，并阻碍对正在研究的系统或系统子过程的机械理解。传感器技术、分析仪器和边缘计算的进步提供了收集和分析由许多环境系统变量组成的大量复杂数据集的能力。这为我们更好地理解环境系统如何应对扰动、如何设计新材料、工具和环境管理技术，或者如何更好地理解复杂环境暴露的影响打开了大门。暴露体) 关于健康。在许多环境相关的研究领域，数据科学的进步，例如机器学习 (ML) 和人工智能 (AI)，已被应用于大型数据集，以更好地破译影响系统行为的系统变量之间的复杂关系并提供新的见解关于理解和解决方案的开发。以下是 ML/AI 方法可能推动环境工程和科学发展的一些非排他性应用示例：分析数据集，包括 2D 和 3D 色谱法、非目标质谱法、遥感或预测毒性途径和不良结果追踪的化学数据来源和预测污染物的环境命运 分析“多组学”数据集以预测生物对扰动的反应，例如，预测系统将如何应对未来的扰动，预测化合物的潜在毒性，或预测最适合选定功能的材料属性。尽管无监督机器学习方法具有“黑匣子”性质，但仔细检查算法可以深入了解控制变量或过程，从而产生更多“生物物理化学信息”的监督机器学习方法（或其混合），利用现有的机械基础发挥作用的过程。这些方法最终可用于更好地理解复杂的环境系统行为，帮助找到以前无法解决的环境科学和工程问题的解决方案，将环境暴露与对人类健康的影响联系起来，并做出更好的政策决策来保护环境及其居民。ES &T和ES&T 字母通过即将出版的关于环境科学技术中数据驱动机器学习和人工智能的联合特刊，将注意力集中在这个新兴和快速发展的领域。通过公开征集论文，这些期刊正在寻求对 ML 和 AI 研究的贡献，以证明这些方法在提高我们对自然和工程环境系统的理解、维护健康的生态系统或建立循环经济方面的巨大潜力。论文需要包括数据科学/机器学习方法的新应用和适用于环境数据集的方法，或者解决重要环境科学和技术中使用标准分析工具无法解决的知识差距。向特刊提交的内容应展示采用 ML 或 AI 方法优于现有方法的“附加值”。提交的内容还应确保数据集足够大且足够复杂，以至于 ML 方法是必要且稳健的，研究人员必须超越现有算法的简单不可知应用程序的“黑匣子”，以确定“最佳算法”。理想情况下，这些论文将收集对正在研究的系统的机械基础的见解。最后，我们预计这些论文将成为对复杂环境数据集进行 ML 和 AI 分析的模型示例。因此，论文必须通过遵守 FAIR 数据原则 (https://publish.acs.org/publish/data_policy) 来促进可重复性，证明计算的严谨性（例如，讨论模型假设/限制、数据考虑、交叉验证、模型性能），并通过公开可用的数据存储库 (https://publish.acs.org/publish/author_guidelines?coden=esthag#data_requirements) 向读者提供 ML 和 AI 模型和数据集。两种期刊的投稿截止日期为 2023 年 1 月 12 日。为了补充计划中的特刊出版，这些期刊还将发布一个联合虚拟期刊，展示近年来在这两种期刊上已发表的关于该主题的内容的集合。ES &T和ES &T Letters期待收到有关这一重要发展领域的投稿。有关文章类型和提交格式的更多信息，请访问每个期刊的作者指南页面：ES&T 和 ES&T Letters。这篇文章尚未被其他出版物引用。