Sustainability Plan enables the federal government to purchase 100 percent clean electricity and 100 percent zero-emissions light-duty vehicles over the decade.

Samaras noted that the Long-Term Strategy of the United States includes multiple pathways under uncertainty to achieve net-zero emissions by 2050.¹ These pathways span the electricity, transportation, industrial, building, land, and agriculture sectors, which he said should be addressed both individually and as a system. Optimum pathways can be established, he continued, but creating achievable pathways is more practical. Two key actions from the Long-Term Strategy of the United States include a net-zero electricity sector by 2035 and electrification of light-duty transportation, heating, and some industry. He explained that, currently, primary energy sources are converted to secondary energy, which is moved, stored, and used for vehicles, buildings, and industry. All of these components have to work together all the time, in real time, but climate affects each of these systems and their physical infrastructure. Therefore, as the nation electrifies, it also has to decarbonize and optimize for resilience so that the electricity system is designed for the climate of today and of the future.

Samaras pointed out that because the massive cyberphysical infrastructure layered on top of the energy system creates both opportunities and challenges, larger systems analysis of the entire sector is necessary, and optimization, systems integration, and coordination would be beneficial. Furthermore, although the energy community is often stovepiped, it is essential to ensure that all components of the 21st century energy system are smart, resilient, affordable, and equitable. He emphasized that climate change makes real-time delivery more difficult, especially as new uses are being added to the system for decarbonization. Optimization tools have improved, but further development would be beneficial, as electricity will likely remain the nation’s primary source of energy for more than half of end uses. He stressed that the research community plays a key role in understanding what an optimized energy system looks like. Part of this optimization involves deploying cyberphysical controls and managing smart devices while maintaining cybersecurity: remote sensing, artificial intelligence, and forecasting are critical.

Samaras highlighted three critical and interconnected challenges related to electrification and decarbonization: ensuring a net-zero power grid, confirming that electrified end uses work together, and enhancing climate-induced reliability and resilience. The transition from petroleum products to clean fuels is critical, and a significant opportunity exists to

___________________

¹ Department of State and Executive Office of the President, 2021, The Long-Term Strategy of the United States: Pathways to Net-Zero Greenhouse Gas Emissions by 2050, https://www.whitehouse.gov/wp-content/uploads/2021/10/US-Long-Term-Strategy.pdf.

electrify commercial and residential building heat with heat pumps. As opportunities to electrify end uses continue to be discovered, he underscored that the economy should be electrified, not just one sector or another.

Samaras indicated that OSTP is developing a national electrification innovation strategy, the goal for which is to accelerate electrification to four times its historic pace. Innovation in equipment, infrastructure, and electricity grid use are essential to achieve this goal. Potential benefits to consumers include increased efficiency and lower energy costs, better air quality, increased energy equity, and reduced air pollution in historically overburdened communities (e.g., Los Angeles). However, electrification is not without challenges—for example, household peak power demand could double with two electric vehicles charging in a garage. Thus, as neighborhoods increase electrification, local power infrastructure capacity upgrades will be required, and systems optimization will be critical. Demand optimization with smart devices and smart infrastructure could reduce the total capacity needed at peak under electrification by nearly one-fourth. As regions electrify, he continued, regional power generation and transmission capacity upgrades will also be required.

Samaras observed that the current grid is mostly one-way power flow, but “Grid 3.0” will be more interactive, flexible, responsive, integrated, and coordinated. He urged the optimization and analysis community to focus on multiple criteria to ensure a successful energy transition, including reducing household energy burdens, increasing energy security, addressing historical equity and justice concerns, reducing GHG emissions, building secure supply chains for critical materials for the 21st century economy, and creating new well-paying jobs and industries. He stressed that innovation in 21st century cyberphysical systems enables decarbonization, reliability, and resilience in an increasingly complex energy system.

4.2 FLEXIBLE PLANNING AND DESIGN: A WIN-WIN APPROACH TO DECISION MAKING UNDER UNCERTAINTY

Richard de Neufville, Massachusetts Institute of Technology, described an uncertain future—the climate, the economy, energy demands, and technology have changed over the past 20 years, and the predictions made about these issues 20 years ago were almost all incorrect. Reality often differs significantly from predictions in timing, level, and direction. He asserted that the standard “requirements” defined for future engineered systems are also often incorrect, and standard models of uncertainty are inadequate. Therefore, flexibility and adaptability in decision making could better accommodate the reality of uncertainty. He suggested seizing new opportunities and avoiding technology “lock-ins” that could result

in poor outcomes, and emphasized the value of “real options”—that is, the right to adjust plans easily, with modular designs and provisions for technology change. This more flexible approach could increase the economic and social value of designs. For example, a recent flexibility analysis revealed that a flexible plan generated a ~20 percent increase in cost effectiveness and performance of a system.²

De Neufville stressed that the decision-making process includes a sequence of choices made over time in the context of changing situations, where each choice is part of a strategy to win and to avoid loss. Effective decision making demands careful consideration of possible developments over time, consequences of possible decisions, and evaluation of the overall results from a combination of possible developments and decisions. He noted that discrete-event simulation (which is, in effect, an extension of a spreadsheet analysis) provides information for decision making: it accommodates multiple patterns of uncertainty at the same time, enables discounted cash flow economic analysis, and simulates patterns of decision making according to possible future developments. Furthermore, this simulation generates a range of results quickly for average and extreme outcomes, with consideration of trade-offs important to stakeholders.

De Neufville explained that because different, incommensurable performance metrics are standard features of decision making, the use of optimization is limited. “Best” cannot be defined: relative weights are subjective and variable, and no good way to create a meaningful objective function exists. Furthermore, the level of risk that is acceptable for some benefit is not clear for any individual or group. He encouraged decision makers to think about diverse stakeholder values, especially for urban infrastructure: value conflicts; ethnic, minority, commercial, and other community interests; and suburban versus city center interests. Because what is best for one group is not always best for another, he underscored that the concept of “urban optimization is dubious.”

4.3 UNINTENDED CONSEQUENCES OF DATA COLLECTION AND USE

Safiya U. Noble, University of California, Los Angeles, explained that social scientists and humanists have become deeply invested in data and their impact on society, with the recognition of both intended and unintended consequences of large-scale computing projects and platforms. She encouraged the development of a broader framework for thinking about this work, especially because “data-fied” projects have political

___________________

² J. DiPietro, 2022, “Flexible Design Approach to Fleet Management,” MS Thesis, MIT Systems Design and Management.

ramifications. Echoing de Neufville, she noted that predictive analytics are often incorrect, and the resulting consequences are different for different people in different parts of the world. For example, narrow artificial intelligence projects may be deployed without a set of strategies for accountability, which reflects a significant disconnect between projects that are intellectually interesting for researchers and the reality of what happens when these projects involve and harm people.

Noble observed that data themselves often unfairly misrepresent people, especially those in vulnerable communities, and intervening on those data is difficult. For example, several smart cities projects are now under scrutiny. Although governments aim to deploy millions of sensors in cities to better understand weather and climate change, cities are growing concerned about privacy issues owing to the information that is collected about individuals. Reflecting on Latanya Sweeney’s work,³ Noble explained that it is no longer difficult to reconstitute data and identify these individuals. Another emerging issue is the use of last-mile public transportation infrastructure for redlining—for example, neighborhoods that experience surge pricing from or are avoided by mobility service providers. She maintained that as the government considers solutions to such problems, relationships between data scientists and social scientists are crucial to understanding possible disparate impacts.

Noble mentioned that the use of data-intensive analytics is also becoming problematic in law enforcement when scrutiny and policing are directed toward the most vulnerable people but not within affluent neighborhoods. She urged cities and governments to pay attention to these issues and noted that some have already banned technologies that violate civil rights. For example, an intervention has begun and lawsuits have been filed in response to the “untenable” situation in Pasco County, Florida, where software that alleges to use predictive data from school records to determine which students would become criminals has been deployed.⁴

Noble reiterated that the consequential dimensions of data projects should be considered instead of thought of only in the abstract. She encouraged local governments to implement better procurement practices and audits of systems; explainable systems are important, and better explanations of training data used for modeling these systems are critical. Centers of expertise are emerging around the United States where scholars who are concerned about the negative impact of training models

___________________

³ For more information about Professor Sweeney’s work, see https://www.hks.harvard.edu/faculty/latanya-sweeney, accessed August 28, 2022.

⁴ K. McGrory, N. Bedi, and D.R. Clifford, 2021, “Targeted,” Tampa Bay Times, https://projects.tampabay.com/projects/2020/investigations/police-pasco-sheriff-targeted/.

on the public are beginning to collaborate and provide resources (e.g., Distributed Artificial Intelligence Research Institute⁵ and Weapons of Math Destruction⁶). Additionally, the House Committee on Energy and Commerce is reviewing the American Data Privacy and Protection Act, which would put greater limits on data collection from the public. Europe has thus far been much more aggressive in passing such legislation, but she anticipated more public policy in the United States that will protect people from excessive data quantification, data collection, and datafication.

4.4 OPTIMIZATION IN PROBLEM SOLVING

David Shmoys, Cornell University, explained that optimization methods provide a means to frame models, conduct analyses, gain insight, and help understand the overall landscape in which decisions will be made. Optimization minimizes a given function over a set of feasible points, and convex optimization and linear optimization models are solvable and widely used. However, the world is uncertain and the future is unpredictable, particularly with respect to human behavior; trying to make predictions in such an uncertain world calls for more robust models (e.g., stochastic optimization models, where the objective value depends on an underlying probability distribution). He noted that as the world continues to progress, it should be understood as an evolving system governed by probabilistic dynamics. Furthermore, many applications will be in the domain of discrete (non-convex) optimization.

Shmoys illustrated a well-known discrete optimization problem, the “traveling salesman problem” (a difficult problem for which finding optimal solutions had been intractable). He presented a circuit board application that displayed the real-time run of an integer programming formulation that can find the optimal solution to this problem. This demonstrates the growth in the power of optimization models, which have expanded their scope over the past decade with better computation and increased algorithmic speed.

Shmoys discussed a range of optimization problems, including single-agent and multi-agent optimization, and noted that multiple agents are competing for resources in our complex world. An emerging area of optimization is non-convex, non-smooth optimization, which is driven by the need to solve models relevant in training deep neural nets and in reinforcement learning. Thinking about an optimization model in terms of

___________________

⁵ For more information about the Distributed Artificial Intelligence Research Institute, see https://www.dair-institute.org/about, accessed August 28, 2022.

⁶ C. O’Neil, 2016, Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy, New York: Crown.

either real-time response or long-term design shapes decisions about what tools could be used, and these tools provide power for the multi-objective exploration of trade-offs that our complex world requires.

Shmoys described the multistep (iterative) process to optimize investments to enhance urban sustainability infrastructure: develop a suite of metrics that capture system performance, optimize operations for a given system design with respect to these metrics, optimize the system design repeatedly, and integrate the impact of human interaction. He shared a case study of using optimization to guide infrastructure design for bike-sharing. Launched in 2013, Citi Bike has more than 20,000 bikes, has more than 1,500 stations, has conducted more than 28 million rides in the past 12 months, and continues to expand. He described the Citi Bike system imbalance in New York City in 2016, noting a low-activity station in Brooklyn in comparison to a much more active station in midtown Manhattan. His team at Cornell built a model with data that could be continuously updated to understand trade-offs and determine how many bikes to stock at each station, because each station in the system has its own dynamics. This gave rise to a straightforward discrete optimization problem that made it possible to identify a target value for each station in the system as well as the number of bikes needed at different times of the day. Such models provide insight into optimal stocking levels, even though the system might not be able to achieve them each day. In addition to this opportunity to manage the operations of the system, Shmoys and his team noted that it was also possible to redesign the system by reallocating capacity and moving docks—moving docks once instead of moving bikes daily is an optimized system design.

Shmoys emphasized that optimization provides guidance, not answers, and he cautioned about the potential impacts of “optimized solutions.” For example, aircraft and crew scheduling were combined to optimize the airline industry at scale in the 1990s; however, the unoptimized system was more robust because robustness was not included in the optimization model. Another example is Waze, which offers personalized driving instructions to drivers but does not consider impacts on backroad neighborhoods. He urged researchers to consider carefully what they are modeling and to stress-test continuously to understand potential impacts. He underscored that partnerships among academia, government, nongovernmental organizations, industry, and community are critical (e.g., Urban Tech Hub at Cornell Tech⁷) and championed the use of available tools to guide decision making.

___________________

⁷ For more information about the Urban Tech Hub, see https://urban.tech.cornell.edu/, accessed August 28, 2022.

in a decision support tool, and three different disaster types and related damage were considered. Outcomes useful for decision makers emerged, including the development of resilience indifference curves to understand which hazards present the greatest challenges.

Finally, using the Port of Singapore as an example, Miller-Hooks described the use of a digital port twin to replace complex operational constraints; resilience was measured as a function of berth-on-arrival, which was useful to determine post-event recovery actions. She highlighted the importance of port “co-opetition,” in which disruptions cascade in an intermodal network of connected ports. Thus, she emphasized that resilience quantification not only should involve individual components but also would benefit from considering system interactions. She noted that ports operate within a larger maritime network. Considering these system-level interactions may even suggest that a port can gain market share by investing in a competing port serving vessels along common routes.

Serving as session moderator, Robert Lempert, RAND Corporation, posed a question about specific preparedness actions. Miller-Hooks explained that models vary by application; preparedness actions could include training, memoranda of understanding, available equipment, equipment-sharing contracts, or available police presence. She said that sophisticated mathematical models can be difficult to use in real operations; however, these models are useful for developing protocols and benchmarks.

4.6 COOPERATIVE AND ADAPTIVE INFRASTRUCTURE INVESTMENT PATHWAYS FOR URBAN WATER SUPPLIES

Patrick Reed, Cornell University, described challenges to urban water supply planning, including a growing population and a changing climate. The American Society of Civil Engineers estimates that the U.S. drinking water infrastructure requires more than $400 billion in investments by 2029. Utilities confront supply reliability and financial stability trade-offs, and droughts and drought mitigation can destabilize utility budgets. Beyond droughts, financial shocks can also significantly impede utilities’ ability to manage and invest in their infrastructure. For example, during the 2008–2009 financial crisis, 20 percent of U.S. water utilities had unacceptably high financial risk.

Reed emphasized that regional cooperation improves utility efficiency by coordinating drought management, sharing water through treated transfers, and co-investing in new infrastructure when necessary. However, this level of cooperation across jurisdictions can be difficult owing to deeply uncertain future conditions. Deep uncertainty relates to conditions

when various parties to a decision do not know or cannot agree on the system, its boundaries, and measures of success; the outcomes of interest and their relative importance; and the prior probability distributions for uncertain inputs to the system. He noted that stochastic and Monte Carlo simulation are critical for mapping the transition from the state of the world, to the action, to the realized outcome. Concerns about climate, demand, construction and permitting, policy effectiveness, and financial conditions that influence the investment horizon are difficult to define in the near term and even more so in the uncertain long term.

Reed and his team have been working with utilities and regional partnerships in the Research Triangle of North Carolina for more than a decade to form a cooperative between large utilities/cities and small utilities/cities, with connections in treated water, geography, and jurisdictional constraints about what resources can be shared in order to increase capacity and efficiency. He described the overall approach as cooperative regionalization in water portfolio management and infrastructure investment pathways. The pathways dynamically and adaptively mix cooperating utilities’ short-term actions and long-term infrastructure investments. Mathematically, state-aware and contextually appropriate actions are identified using model-free policy approximation reinforcement learning in a simulation and optimization framework that exploits heuristic, multi-objective search tools. The focus is to discover state-aware adaptive action policies rather than specific “optimal” decisions. As part of this approach, the regional system’s risk of failure rules will trigger a portfolio of actions when capacity to meet demand reaches a threshold critical condition where an unacceptable risk of failure emerges. Actions could be short term (on a weekly basis) or longer term, with consideration of the individual and collective supply capacity through architectural design and prioritization. Using this model of cooperative, adaptive infrastructure investment pathways, it might be possible to wait 25 years before adding specific infrastructure based on expected future conditions; if the future is more challenging, investment might be needed 5 years earlier, and if the future is mild, no investments would be needed.

Reed reiterated that uncertainty arises in part from the difficulty in defining problems. He presented four categories of regional utility performance: (1) minimum expected investment, (2) expected drought performance and financial stability, (3) drought crisis robustness, and (4) drought crisis and long-term financial stability robustness. Example objectives for these categories relate to reliability, restriction frequency, peak financial cost, infrastructure net present cost, worst-case cost, and unit cost of infrastructure investment. Reed and his team specialize in finding and communicating trade-offs across these regional objectives (see Figure 4-1). The sheer frequency and scale of these problems are often

frontloads essential services such as debt, policing, firefighting, garbage collection, and education; the next tier includes infrastructure maintenance, and the remaining resources might be allocated toward fulfilling elected officials’ campaign promises. Establishing a “rainy-day fund” is neither feasible nor practical, especially in smaller cities, he noted, because money could always be “put to work.”

Banks suggested that decision makers not plan too far ahead, remain flexible in framing problems, and maintain back-up plans and exit ramps. Decision making under distributed responsibility is inefficient, he continued, and when elected officials, lawyers, accountants, and city managers with diverse interests try to coordinate, challenges arise that could result in suboptimal compromises. He observed a dearth of techniques to help city planners structure their decision making. Many vendors sell software to aid in budgeting, zoning, and other functions, but none offer an integrated approach, and no comparative assessment of their strengths and weaknesses yet exists. Furthermore, these tools specialize only in components of city management instead of providing unified support. Because a small rural town faces different challenges and has different resources than a large urban city, planners cannot implement a one-size-fits-all solution.

Banks described agent-based modeling as a useful way for city planners to explore what-if scenarios for infrastructure investment. He proposed the implementation of standard practices for city governance that are explicit and for which cities are held accountable. He also observed that the software support available for city managers is not very sophisticated, created by for-profit companies without accountability. Instead, systems that scale would allow city managers to take advantage of more sophisticated tools and analyses for better optimization.

Banks shared four key takeaways: (1) urban planning and management is important, but optimization theory is rarely used; (2) consistency and flexibility are both critical but difficult to achieve in a democracy; (3) guidance on best practice does not seem to account adequately for specific situations; and (4) this workshop creates a path for follow-up actions to advance the field. For example, Banks suggested the use of free individual spreadsheets based on town size that explain respective decision theories and priorities.

Lempert asked for an example of an institution that has used this analytical decision-making process, and Banks said that the Department of Transportation has allocated money more thoughtfully to railroads, airports, and roadways by using optimization techniques introduced in the 2000s.

market: suppliers offer waste, consumers request derived products, transportation and technology providers offer services, optimization determines prices for remunerating stakeholders, and the solution is a new market design where investments come from the stakeholders participating in the systems. A benefit of this approach is the ability to more systematically include and monetize environmental impacts by capturing the environment as a stakeholder in the market. He stressed that this is not a new concept—for example, electricity markets are coordinated markets because the electric power grid coordinates a vast, complex infrastructure. Benefits of these coordinated markets include that they are scalable, ensure fair allocations of wealth and resources, ensure reliability, and accelerate innovation and investment.

In conclusion, Zavala underscored that urban sustainability requires scalable solutions owing to complex interdependencies and complex space–time behavior. Stakeholder coordination is essential and allows for economic and policy consistency and standardization of products, technologies, prices, and impacts. Opportunities for coordination exist in transportation, decarbonization, food waste, and water pricing.

Lempert asked how to address issues of market power among stakeholders, and Zavala reflected on lessons learned from the power industry, which designed rules to control market power—markets designed with well-established rules can achieve fair outcomes. Lempert also inquired about extended producer reliability for materials recycling, and Zavala replied that the producers of materials are facing much pressure and are interested in solving the problem, but no mechanism currently exists to do so.

4.9 A PICTURE IS WORTH A THOUSAND WORDS: BRIDGING THE GAP BETWEEN OPTIMIZATION AND ACTION THROUGH VISUALIZATION

Lauren Davis, North Carolina Agricultural and Technical (NC A&T) State University, discussed the Emergency Food Assistance Provider System, which includes national and regional programs and resources for those in need of food assistance. Those who are identified as “food insecure” have inadequate access to food and/or to a healthy lifestyle. The level of food insecurity varies across the United States; in North Carolina, for example, some rural areas have food insecurity rates that are higher than the national average. Barriers to equitable food access include transportation, poverty, and physical locations of food distributors (i.e., the food landscape).

Davis emphasized that the gap between optimization and action in this case is caused by a lack of both data and effective partnerships. She

defined the data landscape as complex, with a mix of public and private data, issues with manual data collection, and varied data storage strategies. All of these data could be brought together meaningfully and translated to inform policy; creating visualizations serves as an entry point into model complexity by transforming results into a more understandable form.

Davis described a project that visualized the food landscape of Durham, North Carolina—a collaboration among Duke’s World Food Policy Center, the National Science Foundation, and NC A&T State University—by creating a dashboard that showed food access points in the county. The goal of the project was to increase food access in an equitable way and ensure policy support. This project combined siloed data from the North Carolina Department of Commerce, the Census Bureau, and the Durham Neighborhood Compass and considered food vendor names and locations; community health, transportation, and housing; and census tract boundaries to understand food access. For example, the visualization of the food landscape by business category showed fewer grocery stores in more heavily populated areas of the county; more grocery and convenience stores and restaurants were located in the center of downtown Durham (see Figure 4-2). Key observations from this project include that

Pathway, and operations and maintenance risks are tackled with continual improvement through an environmental and asset management system. Emerging issues include the need for sustainable and resilient infrastructure standards via contractor and designer engagement and performance expectations; the importance of infrastructure, organizational, and fiscal resilience as well as related benefits for the community; and the value of standardized metrics and reporting on investments and investment instruments. Liban underscored that although novel ideas are emerging from academia, a significant gap remains in achieving key outcomes for local infrastructure operations.

4.11 BRIDGING THE GAP BETWEEN OPTIMIZATION AND ACTION

Marie Lynn Miranda, University of Notre Dame, pointed out that data, information, and knowledge should translate to wisdom; however, not all knowledge is used. She suggested spending more time determining how to bridge the gap between knowledge and wisdom to create action.

Miranda described a project on childhood lead exposure in partnership with a Durham, North Carolina, community. Combining several data sources with the use of geographic information systems and statistical software, a spatial data architecture was created to determine the weights of particular influences—age of housing, owning versus renting, assessed tax values, median housing income, and race—on lead exposure and reactions to lead. This information was leveraged to develop an intervention model that could be used to protect children (see Figure 4-3). Miranda explained that the team was prepared to deploy this model on laptops in community organizations so that people would know when their children or homes should be assessed for lead. Issues arose however, with the use of the laptops, and the community instead requested paper maps that could be posted throughout neighborhoods. Her team reallocated the laptop funding for a plotter and printed maps, and community members preferred these maps because they could better connect to and understand the problem through a visual representation. A 600 percent increase in the capture rate of children with elevated blood lead levels resulted.

Miranda emphasized that “implementation science” should become an integral part of research and training programs. Instead of simply downloading data, she encouraged researchers to upload results in ways that are accessible and actionable for the community, which bridges the gap from knowledge to wisdom.

Liban pointed out that many access points for “fresh food” in South Los Angeles are actually liquor stores, not grocery stores.

McNeil solicited advice for modelers and decision makers working in disparate and heterogeneous communities: at what level should models operate to capture the right information? Liban observed a lack of communication between people working in cities and researchers in academia, which prevents the practical achievement of academia’s suggested outcomes. He proposed increased partnerships between the two communities. Miranda said that too much time is spent operating at the county level, where much variation exists. Operating at the census tract level offers more refinement, but operating at the tax parcel level is best. She also encouraged decision makers to think carefully about spatially varying coefficients. Davis explained that she collects data at the lowest possible spatial level that the stakeholder needs for decision making and incorporates those into the model. To build better relationships, she works to understand the process that agencies use so it can be integrated into the model for decision making.

McNeil asked what mathematical models cannot accomplish. Liban highlighted the human factor in decision making: “the model is the model and the data are the data, but the science is an art.” Miranda directed workshop participants to study work by Kathy Ensor, Mercedes Bravo, and Dan Kowal on spatial analysis.⁸ Miranda and Davis championed building relationships with affected communities who will use the models to make decisions; the mathematics cannot substitute for understanding people and the full scope of their problems.

___________________

⁸ For more information about current work on spatial analysis, see http://ensor.rice.edu/, https://globalhealth.duke.edu/people/bravo-mercedes, and http://www.danielrkowal.com, accessed August 28, 2022.