Redesigning the Urban-Rural Interface along the Mississippi River Watershed is a complex problem. Viable approaches require incorporating various stakeholder inputs as well as engaging in careful reasoning about the associated food, energy, water and built environment interactions. An endemic feature of such complex problems is the interaction of multiple phenomena, requiring a confluence of physics-based modeling, human factors modeling and data-driven approaches. We advocate for three key concepts in our quest to model and design resilient Food-Energy-Water (FEW) systems.
Principled collection and dissemination of data: The availability of persistent, cheap and ubiquitous sensing opens up the opportunity of detailed data-driven modeling of these interactions. Data is being collected at multiple scales (ground, aerial, satellite) which affords us the opportunity to think systematically about which data is ‘useful’ (i.e., data with significant information content). We advocate development of data-driven approaches that can help identify which, where and when data is needed. This will enable a rigorous and systematic approach to data collection and assimilation. Most importantly, we as a community must be open to data sharing to enable a community-based effort in model building. This calls for a nuanced discussion about policy changes that create incentives for data sharing.
Data-driven approaches for bridging sub-models: Modern data-driven (specifically machine learning and/or artificial intelligence, ML/AI) approaches have transformed a host of application areas that involve assimilating data streams to make useful predictions. It appears that the time is ripe to leverage these advances for design and analysis of complex systems. However, there is the very real possibility of us forgetting the forest in our quest to see the trees. This is because current ML/AI systems tend to let data entirely dictate the narrative. Such approaches preclude incorporation of contextual information and domain knowledge that the community has painstakingly produced over decades (if not centuries) of scientific exploration. Most current ML/AI approaches additionally suffer from lack of generalizability and unsatisfactory parsimony and explainability. This is especially damaging when the end-goal – identification of functional relationships in complex systems – requires generating insights into the modeled system. We advocate for development of physics-aware machine learning models that seamlessly integrate domain knowledge with available data.
Networked system-of-systems: We advocate framing the problem of modeling complex FEW systems through the lens of ‘(data-driven) analysis, prediction, and multi-scale control of a multi-scale, networked, system-of-systems.’ A major challenge is that such systems (-of-systems) exhibit diverse behaviors across multiple time and spatial scales (Figure 1). The recent emergence of a number of data-driven strategies for discovering nonlinear, networked, and multi-scale dynamical systems from spatio-temporal-series measurement data is giving rise to unprecedented scientific opportunities. This opens up principled approaches to reason about critical questions including “Where is the best location to inject policy, control or externalities?” and “How will changing features at one scale propogate across the system?”
Joseph C. and Elizabeth A. Anderlik Professor in Engineering
Department of Mechanical Engineering
Iowa State University
This event is supported by the National Science Foundation, Award #1929601. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.