Cities’ Role in Mitigating United States Food System Greenhouse Gas Emissions

Current trends of urbanization, population growth, and economic development have made cities a focal point for mitigating global greenhouse gas (GHG) emissions. The substantial contribution of food consumption to climate change necessitates urban action to reduce the carbon intensity of the food system. While food system GHG mitigation strategies often focus on production, we argue that urban influence dominates this sector’s emissions and that consumers in cities must be the primary drivers of mitigation. We quantify life cycle GHG emissions of the United States food system through data collected from literature and government sources producing an estimated total of 3800 kg CO2e/capita in 2010, with cities directly influencing approximately two-thirds of food sector GHG emissions. We then assess the potential for cities to reduce emissions through selected measures; examples include up-scaling urban agriculture and home delivery of grocery options, which each may achieve emissions reductions on the order of 0.4 and ∼1% of this total, respectively. Meanwhile, changes in waste management practices and reduction of postdistribution food waste by 50% reduce total food sector emissions by 5 and 11%, respectively. Consideration of the scale of benefits achievable through policy goals can enable cities to formulate strategies that will assist in achieving deep long-term GHG emissions targets.


S1 Explanation of Emissions Quantified in US Food System
Below are summaries of the quantification methods for emissions from various data sources, based on the components listed within Figure S.1. As mentioned, this study applies a combination of a metareview of LCA studies and emissions calculated based on government and literature data on energy consumption and fugitive gases (refrigerants, landfill gas, anaerobic digesters, composting).
A life cycle-based approach to estimating the greenhouse gas (GHG) emissions associated with the food system is quantified using the boundaries illustrated in Figure S.1. These include emissions from raw materials and farm-level production, as well as primary processing (i.e., milling grain to flour, canning of produce, milk pasteurization) quantified through a meta-review of life cycle assessment studies. Further processing, distribution, retailing, food service and household emissions associated with refrigeration and energy consumption is obtained using US government data. In addition, transportation data for distribution between further processing to households is also tabulated, with the exception of trips for food service sector meals. Transportation emissions related to disposal of food waste and packaging are calculated using the emissions factors from the USEPA WaRM model 1 .

Figure S1: Process diagram of food system life cycle stages quantified in this analysis
Data related to GHG emissions have been gather from a number of sources, which are then compiled to calculate the emissions total for the US food system. Life cycle emissions from farm to primary S4 processing are obtained for 99 commodities from Heller & Keoleian 2 based on a literature meta-review of emissions factors of various food types and the USDA's Loss-Adjusted Food Availability dataset 3 .

S1.1 Meta-Review of LCA Studies
The meta-review approach taken here uses published LCA data to arrive at representative carbon footprint values for the diversity of food commodities consumed in the US. Results are drawn from a variety of sources (as described in Heller & Keoleian 2 ), compiled by food type, and averaged across comparable commodity type. U.S.-based data is limited, and thus this meta-review includes data from other developed countries. These sources include studies representing a variety of countries of origin, climatic conditions, transportation distances, and production methods and therefore are intended to provide a reasonable range of expected values rather than a definitive result for each food type. As explained in the main manuscript, implications for potential double-counting are mitigated through the additional scrutiny of studies that may include post-primary processing GHG emissions for commodities that either have a substantial contribution to total food system emissions or substantial variation across studies. A full list of commodities and associated emissions factors is provided in the supporting information for Heller and Keoleian 2 and is replicated below in Table S1. The revised list of emissions factors for modified commodities are given in Table S2.
All emissions associated with further processing, packaging, transportation, distribution, retail, household preparation, and waste disposal were calculated separately (see Figure S1). Estimates of energy demand, packaging material consumption, waste disposal are made for the year 2010 for each stage of the food supply system. These are described below.
The meta-review from Heller and Keoleian 2 was refined for the top commodity contributors (defined as those that contributed more than 50 kg CO 2 e/cap) or those that had a substantial variation across studies (defined as greater than 100 kg CO 2 e/cap), presented in Table S2. 2 This was to ensure that error introduced from double counting was reduced for these. The commodities included in this refinement were beef, chicken, pork, cheese, fluid milk, eggs and added sugars and sweeteners. For these foods, a further review of LCA emission factors was conducted, with assurance that boundary conditions only extend to the primary processing stage. Added sugars and sweeteners is represented by a consumptionweighted average of honey, white sugar, and high fructose corn syrup. Emissions are calculated by multiplying emissions factors by per capita demand in 2010 for each stage as presented in the lossadjusted food availability data presented in Table S3. This column indicates the number of studies that have been included in calculating the average for an individual food. When no direct match has been found, the column indicates the proxy group that has been chosen. Categorical proxies (e.g., "stone fruits") are an average of all foods that fit the category. Cases where a unique minimum and maximum exist with only one study reported in the average are from single studies that report a range of values.  25 3.29 1.42 7 Figure S2: Uncertainty analysis of dietary commodities with high greenhouse gas emission intensities inedible portions of meats, poultry, fish and nuts are accounted for prior to the retail level; i.e., these foods are presented as boneless-equivalent, edible weight, or shelled basis There is considerable variation across literature reported emissions with important meat and dairy commodities. This reflects, in part, real variation in production practices for these foods, as well as uncertainty introduced by data quality and methodological choices. As the underlying per capita food S10 consumption data does not differentiate production practices, a more refined assignment is not possible in this study. US-specific data are not available for all commodities, and reliance on single or limited USspecific studies may introduce biases that do not represent the true variability that is likely to be seen across production practices.
Food loss and related emissions occurring between farm and primary processing are assumed to be captured in individual LCA studies, with any losses between primary processing and retail assumed to be negligible.
Some data were not available for the year 2010 (discussed below) and therefore were assumed to be correlated with population or economic activity, and estimated accordingly. Per capita emissions calculated from data sources use intercensal population on July 1st (309,326,295); as a result, an assumption of correlation with population, as well as zero-growth with any other independent variables is inherent.

S1.2 Packaging Materials
GHG emission from packaging materials are calculated using the following equation where m i is the total weight of packaging of material i disposed of in a given year, FP i is the fraction that is food packaging, EI i,j is the emissions intensity of material i from source j (either recycled or virgin). Packaging used for food products is derived from the US EPA's 2010 Municipal Solid Waste factsheet (Table 2), using the provided disaggregation of material types used in packaging and containers. 4 Fractions of these total that are used for food packaging are taken from reports and personal communication from the industry groups (Table S5). Emissions factors associated with these, along with recycled content taken from US EPA WARM factsheets for aluminum, glass, steel, HDPE. 1  S12 minimum value Gitlitz and maximum value from US EPA. 8,9 All other values are taken from US EPA (2015) with default values being the average of range of estimates from the meta-review, if provided. 11 Emissions intensity of virgin and recycled materials are taken fromUS EPA. 11 It should be noted that changes made to electricity grid emissions intensity in the tool will not be reflected in the packaging material emissions intensities used here (i.e., grid intensity assumed by US EPA applies) 11 .

S1.3 Secondary Processing
A tabulation of emissions related to secondary processing (i.e., food manufacturing related to further modifying primary products such as flour, processed produce, or pasteurized milk) is based on emissions associated with refrigerant leaks and energy consumption of this sector.
where F r are the fugitive emissions associated with refrigerant r, GWP r is the global warming potential of refrigerant r, EC i,j is the consumption of energy source j from food processing sector i, and EI j is the emissions intensity of energy source j.
Refrigerant leakage is obtained from 2010 data for industrial process refrigerant from the 2014 submission of the US National Submission to the United Nations Framework Convention on Climate Change (CRF Table 2(II).FS1). 10 . The share of industrial process refrigeration is assumed to be the same as the food share of cold storage (88%; Jones Lang LaSalle IP) . 11 GHG emissions associated with refrigerants are taken from US EPA. 12 Energy consumption by fuel type of the secondary processing sector is calculated from the values for the entire food manufacturing sector in the 2010 Manufacturing Energy Consumption Survey 13 . The quantities are scaled to represent secondary processing by the using cost of fuel and quantity of electricity purchased by these processor, using the 2012 Economic Census 14 . Subsectors that are identified to be secondary processing are shaded in Table S9. Note that not all secondary processing sectors are shaded below (e.g., cheese manufacturing, peanut butter, produce canning, ice cream), as these were captured in the meta review of the LCA studies 2 and quantified in the loss-adjusted food availability data. 3

S1.4 Distribution
Greenhouse gas emissions that are quantified from the distribution of food include refrigeration leakage from cold storage and energy demand, as well as trucking associated with shipping processed food.
where EC e is the electricity demand of cold storage from food, EI e is the emissions intensity of the grid, D j is the transport distance of a food product, and FE is the average fuel economy of transportation. Transportation emissions are calculated using mileage taken from US DoT 15 . Categories included that are assumed to capture emissions beyond the meta-review are listed in Table S10, providing an estimate of post-primary processing for food road freight of 10.4% of total 2002 US trucking miles. Mileage associated with transport from farm gate to primary processing industries are assumed to be accounted for in the meta-review of LCA studies and are excluded from the mileage total for distribution. Estimates of 2010 trucking data and fuel economies for single-unit and combination trucks from 16 are used to calculate diesel consumption. GHG emissions are the product of this diesel consumption estimate, the 10.4% fraction of trucking miles calculated using the data in Table S10, and the emission intensity of diesel, providing an upper boundary estimate of 182 kg CO 2 e/cap. As an alternate calculation, 2010 GHG emissions from trucking are taken from US Department of Transportation 15 , and again scaled by the post-primary food fraction of mileage to provide a lower boundary estimate of 146 kg CO 2 e/cap. The primary estimate used is the average of these two. 15 It is assumed that transportation of prepared foods (i.e., post-primary processing) by other modes of transportation are negligible. There is some potential for double counting with these transportation emissions, as sectors are not demarcated using NAICS classifications. Refrigerant leakage from cold storage and refrigerated transport are taken from US EPA (CRF, Table  2(II).FS1). 10 The fraction of commercial refrigeration that is cold storage is assumed to be 10-40% (with retail refrigerant leakage contributing the remaining 60-90%), based on an estimate from a US EPA official responsible for quantification for the US National Inventory Report (Goodwin, 2015; personal communication). An average of 25% was applied for the cold storage share for the primary estimate. The fraction of refrigerated transport emissions (totally 15.8 Mt CO 2 e in 2010) that is used for food is assumed to be the same as the fraction of cold storage that is used for food (88%). 11 S16

S1.5 Retailing
Energy consumption and refrigerant leakage are considered for retail food GHG emissions, considering supermarket and warehouse club retailers.
In this case, i is the type of retail outlet (supermarket or warehouse club). Energy consumption is estimated based on 2011 estimates of US supermarkets and warehouse clubs/supercentres, which contributed to 80% of food-at-home sales in 2010 19,20 . Specific energy demand (natural gas and electricity) for supermarkets are taken from ICF International, 21 while a range of specific electricity demand for warehouse clubs are taken from the average of a UK data sample of 150 UK retail outlets ranging in size from 5,000 to 10,000 m 2 . 21,22 Average floor area for these types of retail outlets are presented in Table S11, based on the revenue-weighted average of Costco and Sam's Club stores, which comprised 87% of sectoral revenue in 2015. 23

S1.7 Grocery Trips
Driving trips between households and food retail have been identified as a potentially significant source of GHG emissions 29 . A number of jurisdictions have begun to promote grocery delivery as a means by which GHG mitigation can be achieved 30 . While there is potential for multi-purpose trips adding uncertainty the total annual mileage that can be allocated to these grocery trips, the assumption made here is that these are single-purpose trips.
Annual GHG emissions from driving are calculated using literature values as the product of annual grocery driving trips, average trip distance, fuel economy, and fuel emissions factors.

= × × ×
Where T is the number of annual trips per person, D t is the trip distance, FE is the 2010 average fuel economy of cars and light trucks, and EI is the emissions intensity of gasoline. The Food Marketing Institute suggests 2.1 household grocery trips made are made per week (with a +/-50% uncertainty range applied), which is consistent with a Puget Sound study 24,31 . Trips per household is divided by Census data on average household size (2.6) to convert this to per capita data 32 .  Table S12.
The main article refers to an extreme scenario where trip emissions could be as high as 370 kg CO 2 e/cap; this would employ an average trip length of 7.0 miles ( 38 , for "errands"), 87.3% of trips made by automobile 31 , and 3.15 trips per week (50% increase on Food Marketing Institute estimate 24 ).

S1.8 Household Storage and Preparation
Emissions from household storage of foods are taken from 2014 39 estimates of electricity consumption of freezers and refrigerators adjusted to 2010 estimates using national population data 40 , along with annual refrigerant loss. 10 where i is the energy end use (refrigerators, freezers, cooking appliance), and j in this case is the domestic energy source (natural gas or electricity).
Cooking emissions from electricity and natural gas are calculated using US EIA estimates. 39

S1.9 Food Waste Disposition
Food waste is assumed to be disposed using the methods presented in Table S13. Non-edible waste components (e.g., bone) are assumed to be disposed of per the weighted average (by mass) of the other three waste streams. Donated food is assumed completely consumed. Uses for direct land application, animal feed, and biofuel production are assumed to be carbon neutral, as are emissions from incineration. Energy production from food waste incineration is not included.

S1.9.1 Landfill Emissions
Landfill gas (LFG) emissions from food waste disposal is calculated using the methane commitment approach from the Global Protocol for Greenhouse Gas Emissions Inventories for cities. 41 Methane generation potential (L o in t CH 4 /t MSW landfilled ) is calculated as where MCF is the methane correction factor, DOC is the degradable organic carbon content of food waste (or dry sludge), DOC f is the fraction of DOC that is degraded, F is the fraction methane in landfill gas, 16/12 is the stoichiometric ratio between methane and carbon.
where MSW x is the total food waste (t) deposited in landfill in a given year, f rec is the fraction of total methane that is collected through landfill gas collection systems, OX is the fraction of that is oxidized through landfill covers, and GWP CH4 is the global warming potential of methane including feedbacks (34;Myhre et al, Table 8.7). 42 Food waste generated is based on estimates from Heller and Keoleian which are, in turn, from data from loss-adjusted food availability data series for 2010. 3 The share of waste produced that is sent to landfill is taken from various sources (presented in Table S14). Off-site emissions are assumed to be avoided through the generation of electricity from LFG-to-energy operations (4% of all LFG generated in 2010). 48 The 2010 US national average grid emissions factor of 0.56 kg CO 2 e/kWh 49 is applied to determine emissions offset by generation from LFG. The lower heating value energy content of methane 50.03 MJ/kg and electric generator efficiency of 36% are applied to calculate total grid electricity that would be offset.

S1.9.2.1 Composting
Emissions avoided through the reuse of food waste has been calculated to provide clarity on the net climate impact of waste management approaches. Emissions are assumed to be avoided through the replacement of organic fertilisers and improved carbon storage. The 2010 values of food waste diverted from landfill was 24%, with 14% of the total being sent to compost and 1.5% anaerobically digested. Emissions from composting are calculated using a modified form from as where M is mass of organic waste treated, EF i is the emissions factor for a greenhouse gas i (methane or nitrous oxide resulting from a given treatment option), and R is the amount of gas recovered (applicable to methane). Using IPCC (2006), methane and nitrous oxide emissions factors of 0.004 kg CH 4 /kg wet waste and 0.0003 kgNO 2 /kg wet waste, respectively, are applied; it is assumed that no methane is recovered.
In this study, 100% of compost produced is assumed to be applied to land. Offsets achieved through the composting of food waste include carbon sink from land application and substitution of inorganic fertilizers (nitrogen, phosphorous, and potassium) by compost. Fertilizer offsets are calculated as where W c is mass food waste composted, i represents a nutrient, RV i is the inorganic fertilizer replacement value in weight of the nutrient per unit of compost (kg/kg ww) and EF i is the emissions factor of production of the nutrient. These values are presented in Table S15. Emissions offsets are achieved through displaced inorganic fertilizers from land application of digestate and the generation of electricity from methane; these are calculated as in Equation S5 and the method used for calculating LFG electricity generation, respectively (see Table S16 for relevant data). Offsets from carbon storage from land application are assumed to be 0.041kg CO 2 e/kg food waste digested ( 53 ; p.805).  It should be noted that in the context of a low-carbon electricity grid, the net benefits of electricity generation from anaerobic digestion and landfill gas will be reduced as the magnitude of the carbon offset is relatively low.

S1.9.2.3 Nutrient Recovery from Sludge
Emissions from US sludge production in 2010 is taken from the US Environmental Protection Agency (2015a; Table 7.7), taken as the sum of CH 4 emissions from domestic and food industry wastewater treatment, as well as N 2 O from domestic wastewater treatment. 48 Landfill GHG emissions from sludge disposal in landfill are calculated as above. Nitrogen removed with sludge are also taken from US Environmental Protection Agency (2015a), with phosphorous : nitrogen in sludge assumed to be 1:10 56 .
Offsets from the land application of sludge from displaced nitrogen and phosphorous inorganic fertilizer are 8.9 and 1.8 kg CO 2 e/kg nutrient, respectively (Boldrin et al; Table 5). 55 Carbon storage from this approach is estimated at 0.25 kg CO 2 e/kg biosolids (dry). 57 Emissions (N 2 O; CH 4 is assumed to be negligible) from land application of sludge are taken as 0.91 kg CO 2 e/kg N applied from Brown et al. 57 Share of biosolids that are land applied in 2010 is assumed as 55%. 58

S1.10 Emissions Factors Applied
The following is a list of emissions intensities that applied for various energy sources used in the calculations above.

S2.1 Urban Agriculture
Urban agriculture is assumed to mitigate emissions relative to conventionally-delivered produce through a reduction in food loss, transportation, distribution, and reduced landfill gas production (due to avoided waste in the food supply chain). Urban agriculture in this instance is defined as open-field production of fruits and vegetables that requires only transportation from retail to household. A reduction in food production required through the conventional food systems that would have otherwise been lost provides an estimate for emissions savings from food loss. To state this another way, emissions associated with conventional production still occur, but the quantity of emissions from food produced and then lost between farm to retail is reduced by the amount produced through urban agriculture.
A reduction in cold storage requirement is assumed through urban agriculture; the fraction of cold storage that is produce is derived from the LAFA, assumed to be total fruits and vegetables produced that require cold storage (fresh fruits, vegetables and fruit juices) divided all perishable food commodities (milk, dairy, meat, seafood, butter, margarine). This gives a value of 45% of all food that is kept in cold storage. It is likely that this is an overestimate, as the LAFA quantities do not include perishable goods resulting from secondary processing.
Emissions associated with transportation of fruits and vegetables are estimated to be 11% of farm to retail emissions by Weber and Matthews. 59 In this case, the emissions intensity of the urban agriculture share of production is assumed to be the average of the literature values obtained from the metareview for fresh and canned produce. Finally, landfill gas emission mitigation associated with food waste reduction are calculated using the methane commitment methodology for landfill gas emissions. 41 Annual yields of vegetables are taken from the average of literatures sources, whose yield data are themselves averaged from the various operations described within their research 60,61 , providing default, high and low values of 2.3, 3.9 and 0.7 kg/m 2 , respectively. Fruit yield is assumed to be 0.56 kg/m 2 , based on 62 . The maximum productive land of urban areas is based on an estimate of percentage of total urban land (275,000 km 2 ) 63 that is vacant of the total urban land area (15% vacant on average, from 70 US cities in 2000). 64 It is assumed in this measure that 50% of this land is used and is suitable for production (which makes this a generous estimate). Fractions of land used for vegetable production are calculated based on current levels of consumer-side consumption and yield data provided above. This is estimated with the relationship where f is the fraction of production of a commodity (denoted with subscript f for fruit or v for vegetable), T is the total consumer demand of commodity (taken from the USDA Loss Adjusted Food Availability data) 3 ; f v is the quotient of T v and T, and y is the estimate of yield for that commodity. This is developed from the relationships where A x is the land area required for commodity x. Seasonality of crops, site quality (i.e., related soil, water, solar resources), temporal mismatch of production/consumption, and property ownership are also not considered, making the estimated available yields optimistic. 65

S2.2 Cultured Meat
Cultured meats are reported to reduce GHG emissions relative to conventional production 66 , with an emissions intensity of between 1.9 -2.2 kg CO 2 e/ kg of cultured meat. A direct substitution of conventional meat products is assumed, with cultured meat replacing each of these in proportion to their current levels of consumption (i.e., not favoring the substitution of any animal source). The emissions factors of conventional meats are taken from the meta-review described above (presented in Table S18).

S2.3 Food Waste Avoidance
Assumes that food waste avoided fully offsets need for production of and related inputs of current balance of commodities, as well as other upstream inputs that result in emissions. For example, for each unit of mass of food waste avoided, the upstream production requirements are also assumed to be avoided. Greater benefits would be realized by avoiding the waste of animal products, but food products are expected to be preserved according to the current distribution of consumption and no weighting towards higher value or more carbon intensive foods is applied.

S2.4 Improved Recycling of Food Packaging
Under the improved recycling rate scenario, the diversion rates in Table S19 were applied in place of those presented in Table S6.

S2.5 Divert Food Waste from Landfill to Anaerobic Digesters
In assuming increased anaerobic digestion of food, the amount of food waste is assumed unchanged, however, diversion is assumed to be as shown in Table S13 apart from landfilled waste from household and retail are reduced to 47% and 8%, respectively. This then allows for 50% of these waste streams to be directed to anaerobic digestion. Emissions are avoided through a reduction in landfill gas generation, offsets of electricity generation/fertilizer production, and increased soil carbon storage.

S2.6 Apply 90% of Biosolids to Agricultural Land
An increase of biosolids application to agricultural land from 50% to 90% reduces emissions of landfill gas from its diversion from landfill, as well as offsetting energy demands from inorganic phosphorous and nitrogen production. Carbon storage is also considered, as discussed in the section "Nutrient Recovery from Sludge".

S2.7 Reduce Grocery Trips by 50%
Emission reductions from grocery trips used the analysis of Wygonik and Goodchild, where the impact of grocery delivery was examined for three Washington state counties. 30 Their analysis considered randomly allocated destinations (consumers selecting delivery times), as well as optimized delivery to reduce distance travelled (provider selecting delivery times). Their results estimated emissions reductions of 45 and 86%, respectively for these two delivery approaches. These reductions are assumed to be consistently achieved in other cities, and are applied as the lower and upper limits achievable, and the average of the two is applied to a scenario in which a 50% reduction in grocery trips is achieved through deliveries.

S2.8 Electricity Grid Decarbonisation
Electricity consumption for the food system is disaggregated for all segments in Figure S1 that are quantified in this study (those that have a solid outline). A decarbonization scenario involves reducing 2010 grid emissions factor of 0.563 kg CO 2 e/kWh 67 to zero.

S2.9 Meatless Mondays
This scenario considers annual average per capita meat production level emissions across all types of meat and seafood produces are reduced by 1/7 = 14.3%, with this being replaced by legumes (and associated production emissions). Emissions associated changes in cold chain and differences in storage and preparation are not considered.

S2.10 Replacing 25% of Beef Consumption with Chicken
Similar to the "Meatless Monday" scenario, beef production emissions are reduced by 25% and the weight of this annual consumption is substituted directly for chicken. As above, emissions associated changes in cold chain and differences in storage and preparation are not considered.

S3 Baseline Emissions
Based on the emissions described above, the following is a summary of the emissions from each sector. Uncertainty estimates are provided where a range could be identified from the literature sources.