Fostering Convergence: Strategies for Designing a Graduate Training Program at the Intersection of Environmental Engineering and Computational Sciences

Environmental challenges are increasingly complex. From climate change to nutrient pollution, biodiversity loss, and inequity of environmental quality, pressing issues are moving from local to global and from specific to systemic.¹ Often, such complex challenges can be addressed only by using a truly interdisciplinary approach, such as those adopting a framework of convergence. To foster a culture of convergence, the National Science Foundation has emphasized the training and education of the next generation of researchers as a crucial characteristic of successful convergence research projects. Convergence requires a new model of graduate education, in which concepts and tools from a primary domain, e.g., environmental engineering, as well as from other disciplines, are deeply integrated. However, how best to train a new generation of students to become leaders in this new scientific landscape remains an open question.

Complex environmental problems often involve underlying relationships in which linear cause-to-effect models are not available, requiring the application of data science tools in analyzing and solving these problems.^2,3 Even though there is a clear need for incorporating artificial intelligence (AI) and machine learning (ML) tools from research and practice perspectives, the lack of adequate training in data science has been identified by the National Academies as a significant challenge in environmental engineering,⁴ highlighting the urgent need for strategies to develop a workforce that is equipped with data science skills in addition to environmental domain expertise. Furthermore, because modern AI/ML algorithms evolve rapidly, issues such as their responsible uses and quality assurance may emerge before regulations or best practices are in place; we contend that the best way to facilitate the integration of AI/ML tools into new solutions addressing complex environmental challenges while preventing unintended consequences is to grow a future generation of researchers who are equipped with the skills and perspectives to identify and address such issues. We envision that the next generation of convergence researchers will be able to build new AI and computational tools that meet the needs and address the complexity of environmental challenges.

In this Viewpoint, we propose specific strategies for incorporating AI/ML into environmental engineering graduate education in a convergence framework. These strategies are incorporated in the National Research Traineeship Program “AI Advancements and Convergence in Computational, Environmental and Social Sciences (AI-ACCESS)” recently funded at Washington University in St. Louis. The AI-ACCESS program aims to build a cohort of new investigators, trained at the intersection of computational science (specifically, AI/ML), environmental science and engineering, and social sciences, with the skills to capitalize on the enormous synergistic potential presented by AI/ML technologies. While certain aspects of the AI-ACCESS program were drawn as examples, the strategies outlined in this Viewpoint are not limited to a specific institution. We intend our viewpoints to serve as a road map and inspiration for others who seek to design and implement transdisciplinary graduate education programs.

Adopting a Team Strategy That Models Cross-Disciplinary Collaborations

Graduate education in a convergence framework requires a team strategy. In AI-ACCESS, we have engaged members affiliated with academic programs spanning environmental science and engineering, earth science, social sciences, and computer science (Figure 1). The team has grown out of collaborative activities among faculty at the institution and a shared motivation to bring AI methodologies to environmental and social challenges. As importantly, Ph.D. students will also be recruited from these academic programs. As contributors to the “environmental science” track in AI-ACCESS (Figure 1), environmental engineering Ph.D. students will receive core training in AI as well as their domain expertise.

Figure 1.

Disciplinary mapping and training strategies in the AI-ACCESS model.

Incorporating Curated AI/ML Coursework Complementary to the Environmental Engineering Ph.D. Curriculum

To help students develop competence and self-confidence in AI/ML techniques, we recommend multiple coordinated courses as part of the formal training. Multiple courses can be particularly helpful to build student confidence. In fact, a previous report showed that the initial anxiety in the first graduate statistics course can be eased following completion of the course, and transition into excitement as well as plans toward more quantitative courses.⁵ Specifically, in the AI-ACCESS program, the doctoral students in environmental engineering will take courses covering inferential statistics (including causal reasoning), the acquisition and management of data, and ethical AI practices (Table 1). They will take these core cohort-building AI/ML courses in the first year of their Ph.D. studies. These cohort-building courses will be supplemented by gap-filling courses for students who had not taken the prerequisites in their undergraduate careers.

Table 1. Overview of AI-ACCESS Courses Taken by Trainees in the Environmental Science Track.

(A) Overview of the AI-ACCESS Curriculum
gap-filling courses	core cohort-building courses	domain depth courses
Fundamentals of Computer Science	Introduction to AI-ACCESS Graduate Research	Environmental Science and Engineering Domain Courses (Year 2)
Introduction to Machine Learning	Explorations in AI + Environmental and Social Sciences
	Data Wrangling
	Environmental Data Science

(B) Overview of Learning Objectives and Topics Covered in Core Cohort-Building Courses
course	overview of learning objectives and topics covered
Introduction to Graduate Research, DCDS 501 (three credits)	This course will introduce general data science concepts, e.g., ethics, the nature of research, robustness and reproducibility of research, and research opportunities.
Explorations in Computational and Data Sciences, DCDS 500 (three credits)	This course aims to lay the foundation for conducting transdisciplinary research involving AI and, more broadly, computational science with environmental and social sciences. Students will interact with real-world data sets and work in teams to apply methods to case studies.
Data Wrangling, DCDS 510 (three credits)	This course will provide an introduction to conducting research using data science methods. Students will examine several sources of data. They will gain skill to ingest data, perform analyses, and document their findings using an electronic notebook such as Jupyter. This course will train students to make reproducible analysis pipelines.
Environmental Data Science, EECE 535 (three credits)	This course will provide training on visualization and analysis of environmental science data sets. Students will gain knowledge and skills in statistical inferences, such as null hypothesis significance testing and calculating statistical powers. An introduction of predictive machine learning models will also be provided.

Providing Cohort-Building Opportunities to Foster Convergence

Enrichments through cohort-building activities have been reported as crucial components of successful interdisciplinary educational efforts.^6,7 We believe that experiential requirements such as research rotations, which have been widely adopted in the biomedical field, can be useful activities for fostering convergence in environmental engineering. As an illustration, students in the AI-ACCESS program will be exposed to research in different areas through research rotations, starting in the fall of their admission year, after which they are expected to identify the specific track in which they plan to do research and pursue their Ph.D. To reflect the interdisciplinary nature of the program, students will also need to identify two AI-ACCESS faculty from different tracks and organize a multidisciplinary thesis committee. It is worth noting that AI-ACCESS fellows will receive specific training on communication, teamwork, and ethics in addition to their technical coursework and research. This training will be implemented through seminars and reading groups, as well as courses on communication, teamwork, and leadership.

Utilizing and Developing AI/ML Tools to Address Environmental Science and Engineering Challenges

Mentored research remains a crucial part of graduate education. We see integrating prediction, policy recommendation, and model explainability as promising areas in which AI integration with environmental science and engineering can produce benefits. The special issue of Environmental Science & Technology on data science provided rich examples of ML methods being applied in environmental and chemical data sets for predictions.³ Additional examples of resource and tool development can be found at the intersection of environmental engineering and bioinformatics. For example, to better characterize and predict the spread of antimicrobial resistance (AR), Brown and colleagues developed a curated database, mobileOG-db. This new resource enables more accurate analyses of mobile gene elements, which are involved in predicting AR spread and elucidating fundamental mechanisms in AR, which can be broadly applied across multiple research fields.⁸ Another example is microbiome-census, a new algorithm for human population estimation motivated by the need to derive more accurate insights from wastewater surveillance into population-level disease transmission. It is worth noting that microbiome-census was faster and more easily explained than off-the-shelf ML tools.⁹ In addition to the utility of AI/ML tools in addressing environmental challenges, considerations should be given to the environmental costs of employing these computational tools, such as greenhouse gas emission from training large language models.¹⁰

Conclusions and a Path Forward

The National Academies report (2019) laid out various specific challenges that environmental engineering researchers can help address but noted that the ultimate challenge in meeting all of them was “preparing the field to address a new future”.⁴ In particular, the report highlighted the imperative of strengthening foundational knowledge in complex system dynamics and the social and behavioral dimensions of environmental challenges. Thus, efforts to integrate AI/ML techniques and social sciences into environmental engineering graduate education, including the AI-ACCESS model, are steps toward a carefully considered vision for the field rather than a detour. Through developing the AI-ACCESS model, we find that environmental engineering research and education are particularly well-positioned to encourage cross-pollination of next-generation researchers. As importantly, current topics in environmental science and engineering provide specific challenges in which a convergence approach to problem solving can be applied. Thus, environmental engineering Ph.D. programs are well suited for the deep integration of AI/ML skills into training programs.

The strategies presented here are by no means the only model for integrated graduate education. In fact, the National Academies’ report (2014) on convergence¹¹ has noted that a “one-size-fits-all” approach is not possible when developing an environment that fosters convergence. A notable example of a graduate education model in a convergence framework is the NRT program Combating Antimicrobial Resistance (CIP-CAR) at Virginia Tech, which takes an approach to addressing antimicrobial resistance challenges through policy, data science, and engineering. Students will complete core coursework through a policy-focused certificate program and participate in cohort-building activities such as boot camps, field-based research, and interdisciplinary thesis chapters. More broadly, it has been proposed that the breadth of expertise needed to provide the ideal skill set for tackling certain aspects of sustainability may even exceed what a single institution can offer;¹² therefore, cross-institutional collaborations in graduate education may be the next question to ponder for researcher educators.

There are limitations to the framework presented here. First, we want to alert researchers and educators to the limitations associated with using AI/ML tools, including potential biases stemming from inadequate data sets used for model training.¹³ Attention should be paid to responsible use of AI.¹⁴ We envision that the transdisciplinary training model discussed here, with specific coursework focused on ethics, robustness, and reproducible research, can contribute to the growth of the next generation of researchers by fostering awareness of these limitations early in their training. Second, although convergence stands as the primary framework through which transdisciplinary research and education is organized in the United States, aligning with the consensus of the National Academies,¹⁵ internationally, the approaches for transdisciplinary work may exhibit variations in both cultures and structures. However, utilizing methodologies across disciplinary boundaries to address pressing environmental issues represents a paradigm shift that is taking place globally, which requires researchers and educators worldwide to innovate the approaches to support graduate students’ learning.^16,17 The AI-ACCESS model is one step toward adjusting the graduate curriculum to better prepare students to work on interdisciplinary projects. We invite continued discussion of strategies to innovate graduate education that will train the next generations of leaders in tackling complex environmental challenges.

Biography

Fangqiong Ling is an Assistant Professor at the Department of Energy, Environmental and Chemical Engineering at Washington University in St. Louis. She earned her Ph.D. (2016) from the Environmental Engineering and Science Program at the University of Illinois at Urbana-Champaign and conducted postdoctoral research at the Massachusetts Institute of Technology. The research group she leads is interested in studying microorganisms relevant for sustainability, public health, and biodiversity conservation. Alongside their research, the group strives to cultivate a training environment conducive to cross-disciplinary learning.

The authors declare no competing financial interest.