Optimal Allocation of Tomato Supply to Minimize Greenhouse Gas Emissions in Major U.S. Metropolitan Markets


 The United States food system requires energy, water, and land in significant proportions, releases large quantities of greenhouse gases, and contributes to other environmental concerns. Meeting future demand for fresh food will be especially challenging, requiring the adoption of holistic, systems-level thinking to maximize production and supply while limiting consequences to the climate and natural resources. We have developed a cradle-to-market life-cycle environmental model to assess the greenhouse gas footprint of fresh tomatoes supplied to ten of the largest metropolitan statistical areas in the United States. A linear optimization algorithm is applied to determine the optimal tomato distribution scheme that will minimize tomato-related greenhouse gas emissions across all ten areas. Monte Carlo simulation was performed to assess the uncertainties in the data. Results indicate that the current tomato distribution scheme is suboptimal; re-allocating the fresh tomato supply across these ten areas has the potential to decrease transportation-related emissions by 34% and overall tomato-related greenhouse gas emissions by 13%—from 277,000 MTCO2e to 242,000 MTCO2e. The substantial variability of the optimized scenario raises questions about its practical implementation. Ultimately, however, production practices and geographic conditions (such as soil and climate) are more significant with respect to environmental impact than the supply allocation or the seasonality of supply. Our analysis found a roughly six-fold difference between Philadelphia tomatoes sourced from open-field Virginian production (0.38 kgCO2e/kg) compared with controlled-environment Mexican production (2.3 kgCO2e/kg).

Notes: Italicised rows indicate metropolitan statistical areas that were excluded from the analysis due to lack of data. The total population for all ten areas included in the analysis comes to 84 million, representing roughly one quarter of the U.S. population in 2019.
We characterize the carbon footprint of fresh tomatoes for each of these metropolitan statistical areas during each week of the year. Next, we implement a linear optimization algorithm with 4,680 decision variables to compute the optimal tomato distribution scheme for the ten metropolitan statistical areas that minimizes the total environmental impact across all ten areas. Last, we comment on whether the presence of an omnipresent national-level agricultural "social planner" could potentially mitigate food-related GHG emissions, or whether the current scheme-whereby each city acts in its own particular self-interest-is preferable.
Tomatoes were chosen as the focus of this study for a number of reasons. First, tomatoes are one of the most popular specialty commodities in the United States. Roughly 9 kilograms (21 pounds) of fresh tomatoes and 30 kilograms (66 pounds) of processed tomatoes are consumed annually per person in the United States 8 .
Second, tomatoes are grown using a variety of production methods, including indoor. In 2012, greenhouse tomatoes were a $400 million industry with over 1000 acres of greenhouse tomatoes in production 9 . Tomatoes account for more than half of all greenhouse production by area and nearly two-thirds of all greenhouse production by economic value 9 . Although indoor tomato production often requires more energy relative to conventional production, transportation distances to the consumer are typically shorter. Finally, tomato production in the United States is diffuse; in 2019, ten states reported over 1000 acres harvested 9 .
Life-cycle assessments of tomatoes are numerous in the literature. Table 2 presents 47 cradle-to-farm gate life-cycle carbon footprints collected from 29 published journal articles. The values represent a variety of growing practices and geographic regions. The data presented in Table 1 re ect only the tomato production stage; processing, transportation, storage, and other stages beyond the farm gate are not included. In some cases, estimates were made in order to subtract transportation-related GHG emissions from the original value presented in the journal article. If the methodology of a journal article was insu ciently transparent to isolate the cradle-to-farm gate portion of the life-cycle carbon footprint, that article was excluded from Table 2. Although the cradle-to-farm gate carbon footprint of tomatoes has been studied extensively, a much smaller number of studies estimate the cradle-to-market or cradle-to-consumer environmental impact. Even fewer consider the impacts of seasonality and logistics. Roos and Karlsson 17 found that the carbon footprint of Swedish tomato consumption was strongly impacted by seasonality since out-of-season tomatoes are likely to be traveling greater distances or produced in heated greenhouses. Kulak et al. 39 , estimated the environmental impact of fresh produce sourced from an urban community farm and versus those obtained through conventional means. They used linear optimization to determine the optimal community farm design to maximize environmental savings.
This study is unique in its application of a holistic cradle-to-market environmental model to estimate the carbon footprint of tomatoes from a variety of production regions and production practices. It is also the rst study of its kind to apply linear optimization to compute the optimal supply portfolio of an agricultural commodity at a national level. The model was applied to ten of the largest metropolitan statistical areas in the United States to investigate the potential reduction of environmental impacts from food production and distribution.

Results
Under the current (i.e., baseline) scenario, supplying the ten metropolitan statistical areas with fresh tomatoes releases roughly 277,000 metric tons of CO 2 e per year. Figure 1 was created by summing the environmental impact of fresh tomatoes across all ten destination cities. The optimization scenario saves roughly 35,000 MTCO 2 e per year-a 13% improvement. Since our model assumes xed supply and demand, the only opportunity for improvement is in reducing transportation-related emissions by varying the supply portfolios of the ten destination cities. By our calculations, transportation represents 33% of the total environmental impact of fresh tomatoes delivered to these ten cities. This fact limits the potential for improvement. However, our optimization reduced transportation-related emissions by 34%. Figure 2 plots the environmental impact of fresh tomatoes delivered to market in all ten cities under the current system. The environmental impact varies relatively little throughout the year and from city to cityroughly 40% between the best city (Dallas) and worst city (Boston). The results can be roughly grouped by geography; the northeastern cities of Boston, New York City, Philadelphia, and Washington DC all share similar characteristics. The same can be said of the southeastern cities (Atlanta, Miami) and the western cities (Los Angeles, San Francisco). Chicago appears to be in a category of its own, although it shares many characteristics with the northeastern grouping. Dallas is similarly in its own category and exhibits the lowest overall carbon footprint, primarily due to its proximity to Mexico. The top panel of Fig. 5 shows Philadelphia's current (i.e., baseline) tomato supply on a weekly basis. As illustrated by the gure, Philadelphia currently receives tomato shipments from seven out of the nine major production origins. Under an optimized scenario (bottom panel), Philadelphia's tomato supply would shift to less diverse production regions that the supply of one region dominates the total supply each week. In addition, the optimized scenario suggests that Philadelphia should receive a larger proportion of its tomatoes from nearby regions (e.g., Florida, South Carolina, Virginia) and a lesser proportion from distant regions (e.g., California, Mexico). These general conclusions are consistent across all ten destination cities.
The top panel of Fig. 6 illustrates the current cradle-to-market carbon footprint of Philadelphia tomatoes on a weekly basis. Considering the temporal variation in tomato supply shown in the top panel of Fig. 2, the carbon footprint of Philadelphia tomatoes is surprisingly consistent, remaining around 0.73 kgCO 2 e per kg throughout the year. Under the optimized scenario (bottom panel), the carbon footprint drops to roughly 0.60 kgCO 2 e per kg for the majority of the year. However, the carbon footprint under the optimized scenario experiences three distinct spikes in July, September, and November. These spikes can be attributed to an increase in shipments of Mexican tomatoes during these time periods. Once again, these conclusions are consistent across all ten destination cities. In general, the carbon footprint of tomatoes is lower under the optimized scenario but is prone to signi cant uctuations. This fact raises some concerns for practical implementation, as will be discussed in the Discussion and Conclusions section.

Discussion
Out of ten major metropolitan statistical areas in the United States, Dallas proves to have the lowest-impact tomatoes-0.61 kgCO 2 e per kg on average-due to its relatively close proximity to Mexican production. Boston has the highest impact at 0.87 kgCO 2 e per kg, an increase of roughly 40%. More signi cant is the tomato production origin; open-eld tomatoes supplied to Philadelphia from Virginia were found to have a carbon footprint of 0.38 kgCO 2 e per kg, whereas controlled-environment tomatoes supplied to Philadelphia from Mexico had a carbon footprint of 2.3 kgCO 2 e per kg. This discrepancy represents a nearly six-fold increase.
The impact of seasonality was minimal; winter, spring, summer, and fall tomatoes for the Philadelphia market were found to have environmental impacts of 0.72, 0.72, 0.75, and 0.77 kgCO 2 e per kg, respectively.
Our analysis indicates that the current national tomato distribution scheme is suboptimal. Under the current system, urban markets source tomatoes from a wide variety of production regions, some of which are located at great distances. Under an optimal scenario, each city would source tomatoes from a select subset of production origins, giving preference to local production. Such a scheme could reduce transportation-related life-cycle GHG emissions by 34% and overall cradle-to-market life-cycle GHG emissions by 13%. The potential bene ts of the optimization are limited by the fact that transportation accounts for only 33% of the total environmental impact of fresh tomatoes delivered to these ten cities. This is consistent with Weber and Matthews' conclusion that 28% of the carbon footprint of fruits and vegetables is attributable to transportation. Based on these results, it is likely that transportation mode and growing practices have a more signi cant impact on the carbon footprint of fresh tomatoes than the supply portfolio.
The tomato distribution systems generated by our model might be also optimized for cost. Comparing with the cost data of tomato shipments from the USDA AMS database 40 , the environmental impacts share similar trends with the costs of tomatoes: (1) The environmental impacts and costs of tomatoes from protected environment production are higher than those from open-eld cultivation. (2) The GHG emissions and costs of tomatoes with longer transportation distances are higher than those with shorter distances (e.g., the cost of California tomatoes is higher if shipped to Boston than to San Francisco).
Before implementing such an optimal allocation scenario in practice, we must consider other factors besides GHG emissions. First, optimizing based on annual GHG emissions may prove economically undesirable. One characteristic of the optimal scenario is that it increases the week-to-week variability in the average environmental impact of tomatoes relative to the baseline. In the case of Philadelphia, this variability is as much as a factor of two. The linear optimization algorithm does not impose any penalty to discourage variability. It is therefore conceivable that the optimal scenario could produce signi cant and undesirable uctuations in the weekly market price of fresh tomatoes. Perhaps a slightly higher environmental impact is the penalty that we pay for market stability. Second, this analysis assumes that all tomatoes are capable of serving the same purpose, regardless of the production method or geographic region (e.g., an open-eld tomato is just as avorful as a greenhouse-grown tomato). Greenhouse-grown tomatoes are typically costlier and may occupy a different niche than tomatoes produced outdoors. In practice, it may not be realistic to assume, for example, that Philadelphia can make do without any greenhouse-grown or controlled-environment tomatoes.
The uncertainty analysis showed that the environmental impacts of protected-environment systems have larger variation than the open-eld cultivation system due to geographic conditions and production techniques. As demonstrated by the literature review in Table 1, there is signi cant variability within these subclassi cations of protected cultivation. The "greenhouse" category is particularly nebulous; the de nition of a greenhouse is far from consistent in the literature and can refer to a wide range of production practices and technologies. Another suggestion from the uncertainty analysis is to improve the results by high-resolution transportation data. Since the USDA movement reports used in the model only include data on the origin-but not the destination-of agricultural shipments, city-level supply matrices had to be estimated by adjusting national-level movement data based on city-level terminal market reports.
This study presented a comprehensive cradle-to-market environmental model estimating the life-cycle GHG emissions footprint of fresh tomatoes for ten of the largest metropolitan statistical areas in the United States.
Our analysis demonstrated that the current fresh tomato distribution scheme is suboptimal. Simply reallocating tomato supplies could decrease the overall environmental impact of tomatoes-and likely other fresh fruits and vegetables-in the United States. However, the results also suggest that geography and production practices may play a more signi cant role in mitigating the environmental cost of fresh fruits and vegetables than the allocation portfolio or the seasonality. The accuracy of these results, as well as the applicability of this systems-level approach to other commodities and regions, could be greatly improved by the adoption of a universal framework for agricultural data collection and reporting. Such a framework would allow for the development of regionally-and temporally-speci c carbon footprint of agricultural commodities, and would lay the groundwork for optimal decision-making at the nexus of food, energy, and water.

Methods
The objective of the linear optimization is to develop a mathematical model to minimize the total annual environmental cost of meeting the fresh tomato demand of major U.S. metropolitan areas. The model assumes that supply and demand are both xed; production cannot be increased beyond the current capacity of each production origin and per-capita tomato consumption cannot change from the status quo of each destination city. Since we nd no support for a difference between the quality of tomatoes from open-eld and protected cultivation, we assume that tomatoes grown under eld and protected conditions are interchangeable in the market. The problem formulation is as follows: The United States primarily relies on 9 production pathways to supply the majority of our fresh tomatoes.
California, Florida, Mexico, South Carolina, and Virginia are home to signi cant open eld tomato production.
In addition, California, Florida, and Mexico have protected production. Mexico's protected tomato production can be further subdivided into adapted environment and controlled environment. Table 3 summarizes the various classi cations of protected agriculture used in this analysis.

Greenhouse (GH)
A framed or in ated structure, covered by a transparent or translucent material that permits the optimum light transmission for plant production and protects against adverse climatic conditions. May include mechanical equipment for heating and cooling 41 Controlled environment (CE) Grown in a fully-enclosed permanent aluminum or xed steel structure clad in glass, impermeable plastic, or polycarbonate using automated irrigation and climate control, including heating and ventilation capabilities, in an arti cial medium using hydroponic methods 42 The environmental cost matrix consisting of 90 origin/destination pairs was computed (Table 4). Following the method of Bell and Horvath (2020), the environmental cost matrix includes GHG emissions associated with the production, post-harvest processing, packaging, and transportation stages. The emissions from the production stage include the life-cycle emissions associated with the uses of electricity, direct fuel, fertilize, various materials, pesticides, and water. The processing stage includes electricity use for short-term cold storage. The packaging stage covers the emissions from the manufacturing of cardboard for packaging tomatoes. The emissions from the transportation stage are the life-cycle emissions from shipping tomatoes by truck. The transportation distances were determined by Google Maps. The detailed method and data sources can be found in the Supporting Data (S1-S4). Each value in the cost matrix (c ij ) represents the cradleto-market life-cycle carbon footprint between the production origin and the destination city (i.e., the environmental cost of supplying one unit of tomatoes from the production origin to the destination city, measured in kgCO 2 e emitted per kg of tomatoes delivered to market). The available supply for each production origin in each week was assumed to be the current tomato production, as determined from USDA Agricultural Marketing Service (AMS) specialty crop movement reports 40 . These national-level data were scaled down proportionally to account for the fact that the ten metropolitan statistical areas comprise only one quarter of the U.S. population. This analysis does not consider the possibility of increasing regional tomato production. The fresh tomato demand for each city in each week was calculated from the national-average per-capita fresh tomato availability, scaled up based on the population of each metropolitan statistical area in 2019 7,8 .
Uncertainty assessment Monte Carlo simulation was performed to assess the uncertainties in the data. The sources of uncertainty included electricity use for storage, material use for packaging, transportation distance, and emission factors of production practices, electricy, fuels, packaging materials, and transportation. Most of the ranges of the parameters were based on existing literature. The probability distribution functions of the parameters are provided in the Supplementary Data (S5). We ran 10,000 iterations for each city, and the error bars show 90% uncertainty intervals of simulated results.
Declarations Figure 1 Total environmental impact of fresh tomatoes for all ten U.S. cities (baseline vs. optimized scenario).

Figure 2
Page 18/20 Environmental impact of fresh tomatoes delivered to market in ten U.S. cities (baseline).

Figure 4
Cradle-to-market life-cycle GHG emissions for Philadelphia's fresh tomato supply. Errors bars represent 90% uncertainty ranges obtained from Monte Carlo simulations. Key: AE = adapted environment, CE = controlled environment, GH = greenhouse