“Open knowledge on U.S. food systems to support decision-making”

Dr. Karakoç presented a data-driven effort to better understand how food moves through the United States—and why that matters for resilience, efficiency, and food security. Thirty-three participants attended the webinar, hosted by CIAS, campus partners, and the NSF AI Institute ICICLE. 

At its core, her work addresses a simple but consequential gap: while the U.S. food system is highly complex and deeply interconnected, the data we use to understand it are relatively coarse, infrequent, and not well suited to local or infrastructure-level decision-making. The research responds by building a more detailed, spatially explicit picture of food flows and using that to identify vulnerabilities and opportunities for improvement.

The starting point is the structure of the agri-food system itself. Food moves through three broad stages—production, processing, and consumption—linked by transportation networks such as highways, railways, and waterways. While this structure is widely understood, the actual pathways food takes between these stages are much less visible. Existing data from the US Department of Transportation’s Freight Analysis Framework (FAF) provide valuable information on commodity flows by mode and value, but they are aggregated into large regions and updated only every five years. This limits their usefulness for understanding local dynamics or responding to disruptions in real time.

To address this limitation, the research develops a modeling framework that estimates food flows at a much finer spatial resolution—down to the level of individual counties. Methodologically, this involves combining network inference with econometric modeling. A logistic regression is used to estimate whether a connection between two regions is likely to exist, followed by a gamma regression to estimate the magnitude of flows along those connections. These modeled flows are then constrained using practical assumptions about how food systems operate, including supply–demand balance and the tendency to prioritize shorter-distance transport for perishable goods. The result is a semi-linear optimization framework that produces a plausible, high-resolution representation of food movement across more than 3,000 U.S. counties.

When these reconstructed networks are analyzed, several key patterns emerge. First, the spatial structure of U.S. food distribution is highly stable over time. Across multiple years, the same regions consistently appear as central hubs. The Midwest, particularly along the Mississippi River corridor, plays a dominant role in moving food, reflecting its position as a major production region. In contrast, large metropolitan areas function primarily as consumption centers. This stability suggests a well-established system, but it also implies structural dependence: disruptions in these central regions could have far-reaching national impacts.

This dependence becomes especially clear when examining system responses to shocks. For example, during the 2012 Corn Belt drought, the model captures a decline in grain movement in the Midwest and a corresponding increase in flows from regions such as Texas and California. While this indicates a degree of adaptability in the system, it also highlights the concentration of production and the importance of a limited set of regions in maintaining overall supply.

Building on this foundation, the research introduces a framework for identifying “choke points” within the food system—locations that are especially critical for maintaining connectivity, efficiency, and resilience. These are defined using multiple network metrics, including degree of connectivity, centrality, and the impact of node removal on overall system performance. The analysis shows that choke points are concentrated in major logistics and economic hubs, often overlapping with regions of high population and strong transportation infrastructure. In practical terms, these are places where disruptions—whether due to climate events, infrastructure failure, or policy shocks—could cascade across the entire system.

Importantly, the spatial distribution of these critical nodes varies by commodity type. For unprocessed agricultural goods such as cereal grains, choke points tend to align with production regions like the Corn Belt. For more processed foods, however, the network becomes more geographically dispersed, with critical nodes concentrated in urban areas and regions with significant food manufacturing activity. This distinction underscores the role of processing in reshaping food distribution patterns and suggests that vulnerabilities differ depending on where a commodity sits in the value chain.

To translate these insights into more actionable terms, the research maps estimated food flows onto actual transportation infrastructure using a GIS-based framework. By linking flows to specific highways, rail lines, and waterways, the analysis identifies which physical routes carry the greatest volumes of food. This reveals the backbone of the U.S. food distribution system: highways dominate overall, particularly for domestic flows, while rail and inland waterways play a crucial role in transporting bulk commodities such as grains. The Mississippi River emerges as a particularly important corridor, especially for large-scale domestic and export movements.

At this stage, the analysis turns to a central policy and systems question: how do different transportation modes perform across key supply chain objectives? The research evaluates trade-offs among cost efficiency, environmental sustainability (measured through carbon emissions), and resilience. The results highlight a fundamental tension. Waterways are the most cost-effective and lowest-emission option but are highly vulnerable due to limited redundancy—if a key route like the Mississippi River is disrupted, alternatives are scarce. Highways, in contrast, offer the greatest flexibility and resilience, with multiple alternative routes, but they are more expensive and carbon-intensive. Rail occupies an intermediate position, offering efficiency for bulk transport but less flexibility than trucking.

A complementary measure of resilience—flow diversity—reinforces this pattern. Even where multiple routes exist, flows are often concentrated along a small number of dominant pathways, particularly in water-based transport. This concentration increases systemic vulnerability, as disruptions to a single route can have disproportionate effects.

An optimization exercise focused on cereal grains illustrates the implications of these trade-offs in a concrete way. When the system is optimized for cost or emissions, it shifts toward greater use of rail and waterways. When optimized for resilience, it relies more heavily on highways. However, observed real-world patterns do not align fully with any of these optimal scenarios. Instead, the system exhibits a strong reliance on trucking, even for long-distance bulk transport where other modes might be more efficient. This divergence suggests that current logistics decisions are shaped not only by technical efficiency but also by institutional, infrastructural, and operational constraints.

The broader implication is that there may be significant opportunities to improve system performance through targeted investments or policy interventions—for example, by expanding rail capacity or improving the reliability of waterway transport—while also recognizing the need to maintain flexibility and redundancy.

The talk concludes by outlining several ongoing research directions that extend this framework. These include linking spatial patterns of food availability to food insecurity and waste, examining how different types of shocks (such as COVID-19, trade disruptions, and climate events) affect food flows across regions and commodities, and expanding the analysis to global supply chains. In the global context, particular attention is given to the vulnerability of low-income countries to disruptions in cereal grain trade and the potential for diversification strategies to enhance resilience.

The discussion following the presentation highlights both the promise and the challenges of this work. Participants raise questions about the sustainability of government data sources and the potential role of commercial, real-time data in improving temporal resolution. Others probe the assumptions underlying the model, including how it handles on-farm consumption or commodity transformations (such as corn used for ethanol). There is also strong interest in the accessibility of the tools and data, with the researcher emphasizing a commitment to open-source platforms and the eventual release of code and datasets.

Taken together, the talk bridges technical rigor and practical relevance. It demonstrates how advances in data integration, network modeling, and spatial analysis can make the food system more visible and measurable, while also highlighting the structural dependencies and trade-offs that shape its performance. For researchers, it offers a methodological contribution to the study of complex supply chains. For practitioners and policymakers, it provides a clearer basis for identifying critical infrastructure, anticipating disruptions, and designing interventions to improve the resilience, efficiency, and sustainability of the agri-food system.

Biography: Deniz Berfin Karakoc is an Assistant Professor in the School of Computing and Augmented Intelligence at Arizona State University. She holds a Civil and Environmental Engineering Ph.D. from the University of Illinois at Urbana-Champaign with a specialization area in Sustainable and Resilient Infrastructure Systems. Her research focuses on developing interdisciplinary frameworks to model agri-food systems across spatial scales and enhance their resilience, sustainability, and equity characteristics. During her studies, she was awarded the 2019 Best Thesis Award from the University of Oklahoma and the 2023-2024 Dissertation Completion Fellowship from the University of Illinois at Urbana-Champaign. She was also part of the 2023 CEE Rising Stars cohort at MIT. Her recent publications have been featured in Nature Food, Environmental Research Letters, and Environmental Science & Technology.

Credits: This webinar series is made possible by the ICICLE – Intelligent CI with Computational Learning in the Environment – U.S. National Science Foundation AI Center under OAC-2112606 and organized by the University of Wisconsin-College of Agricultural and Life Sciences Center for Integrated Agricultural Systems, with support from several UW partners: the Grainger Center for Supply Chain Management, the Kaufman Lab, Center for Sustainability and the Global Environment (SAGE), Organic Collaborative,  Center for Cooperatives, and the Food Studies Network.