Urban agriculture’s bounty: contributions to Phoenix’s sustainability goals

With over half of the world’s population living in cities, there is mounting evidence indicating that investments in urban sustainability can deliver high returns on socioeconomic and environmental fronts. Current scholarship on urban agriculture (UA) reports a wide range of benefits which have been shown to vary with the scale and type of benefit examined. Notably, most city-scale studies do not align benefits of UA with locally meaningful goals. We fill this gap by conducting a city-scale analysis for Phoenix, the fifth largest city in the USA by population, and evaluate these benefits based on their ability to contribute to select desired outcomes specified in Phoenix’s 2050 Sustainability Goals: the elimination of food deserts, provision of green open space, and energy and CO2 emissions savings from buildings. We consider three types of surfaces for UA deployment—undeveloped vacant lots, flat rooftops, and building façades—and find that the existing building stock provides 71% of available UA space in the study area. The estimated total food supply from UA is 183 000 tons per year, providing local produce in all existing food deserts of Phoenix, and meeting 90% of current annual consumption of fresh produce based on national per capita consumption patterns. UA would also add green open space and reduce by 60% the number of block groups underserved by public parks. Rooftop deployment of UA could reduce energy use in buildings and has the potential to displace more than 50 000 tons of CO2 per year. Our work highlights the importance of combining a data-driven framework with local information to address place-based sustainability goals and can be used as a template for city-scale evaluations of UA in alternate settings.


Introduction
Globally, the proportion of urban dwellers is expected to increase from 55% to 68% by 2050 (United Nations (UN) 2018). Cities occupy a small portion of the world's land area (1%-3%), but are responsible for most of the energy and resources used and emissions generated (Grimm et al 2008. Today, large-scale urbanization is a change process occurring simultaneously in many parts of the world . However, the nature of these urban transitions and patterns of urbanization varies from place to place, each with their particular challenges and consequences (Cohen 2006. Although the growth of cities in Asia and Africa involves loss of highly productive cropland with implications for global food production (Bren d'Amour et al 2017), developed cities in North America and Europe have completed their agrarian transition. There, retrofitting existing building stock is a sustainability priority that will lead to substantial reductions in energy use (Georgescu et al 2015. Therefore, it is important to identify the appropriate city-scale response for achieving sustainability, which then needs to be complemented by an understanding of feedbacks and interactions across scales. Urban agriculture (UA) is one tool in a portfolio of measures for achieving urban sustainability and making cities more resilient (de Zeeuw et al 2011, de Zeeuw andDrechsel 2015). It contributes to resilience by providing locally produced food and diversifying existing food supply, creating alternative earning opportunities for residents (Zezza andTasciotti 2010, de andZeeuw et al 2011). It serves as an opportunity to productively use increasingly scarce urban water resources by reusing greywater, and redirecting household organic waste (de Zeeuw et al 2011). Commodities supplied through UA are distributed via short marketing chains, potentially reducing transportation costs and associated emissions (Weber and Matthews 2008). UA landscapes can act as urban habitats for wildlife (Goddard et al 2010).
There are risks and drawbacks associated with UA that may reduce the magnitude of benefits, though these can be managed by testing, treatment, and appropriate management practices. Food production can be inefficient with respect to large-scale commercial agriculture (Desrochers and Shimizu 2012). Plants grown on contaminated soils (Beniston andLal 2012, Meharg 2016) or in areas of high air pollution (Bell et al 2011) reduce yields and pose health and safety concerns. Other risks associated with UA involve the application of herbicides and pesticides, and use of untreated manure or wastewater, and introduction of new pests and diseases (Smit et al 2001).
Global studies , Clinton et al 2018 emphasize the scale of the UA opportunity. In addition, quantifying the potential for co-benefits that extend beyond merely addressing food insecurity concerns (Clinton et al 2018) require place-based assessments to tailor how UA can be used to achieve sustainability and contribute to resiliency in specific urban contexts. In the process, it is essential to focus on solutions that are consistent with place-based desired outcomes-'a sustainability wish list'-of urban policymakers and residents. These can accelerate the transition towards urban resilience and reduce risks due to a lack of understanding of the socio-political infrastructure (Eakin et al 2017). This makes sustainability science relevant to city needs, consistent with its goal to be 'use-inspired' and 'action-oriented' (Turner et al 2007, William and Clark 2007. Here we quantify UA benefits for the City of Phoenix, AZ, through a 'desired outcomes' framework by connecting UA with Phoenix's locally defined sustainability goals. Phoenix is the fifth largest and second fastest growing city in the US by population (1.63 million people in 2017 with 1.5% increase over 2016-2017; US Census Bureau 2018a). The region's climate allows for year-round crop production with the aid of irrigation. Not unlike other Sunbelt cities, however, urban population growth and climate change exacerbate existing water-resource concerns for Phoenix (Gober andKirkwood 2010, Georgescu et al 2013). The overarching question we address here is to what extent can UA be a tool to advance sustainability goals in arid cities like Phoenix and what tradeoffs exist with use of scarce water resources?
We adopt the data-driven approach introduced by Clinton et al (2018), which performed the first global assessment of ecosystem benefits owing to UA deployment. Our work builds upon prior city-scale assessments of UA (e.g. Mendes et al 2008, Patel and MacRae 2012, McClintock et al 2013, Ackerman et al 2014, Haberman et al 2014, Orsini et al 2014, CoDyre et al 2015, Mack et al 2017, Saha and Eckelman 2017 by extending analysis beyond food production. We inquire whether a broader range of locally-defined sustainability outcomes, in addition to food production, could be attained, and if so to what extent, through UA adoption. The developed framework is translatable to other cities and can be used as a template for similar city-scale evaluations of UA as well as other urban sustainability solutions.

Study area
Phoenix is the seat of Maricopa County and capitol of the state of Arizona, in southwestern United States (US) (figure 1(A)). Our study area includes all census block groups with LiDAR building footprint data that intersect the administrative boundaries of Phoenix ( figure 1(B)).
Arizona is the sixth largest state in the US by land area. Urban areas constitute 2% of land in Arizona (compared to 3% across the US), with Phoenix being the largest urban area in the state. Situated in the Salt River Valley in the Sonoran Desert, Phoenix is spread over nearly 1344 square kilometers (519 square miles). It has a subtropical desert climate with abundant sunshine (85% of daytime hours) and receives about 203.2 mm of rainfall per year.
Agriculture has a long history in Phoenix. Evidence of irrigation canals and seasonal farming predate modern settlement by over a thousand years. Modern Phoenix also has roots as an agrarian community. The city originally consisted of a small urban core surrounded by farmland. Since the 1940s, it has transitioned into a commercial and industrial center, resulting in the current land use pattern observed: loss of farmland to development and vacant lots in the urban core and along the periphery.

Desired outcomes and metrics
In 2016, the City of Phoenix adopted the 2050 Environmental Sustainability Goals towards becoming a 'Sustainable Desert City' as envisioned in the city's General Plan (City of Phoenix 2017a). The city's current goals were developed with widespread input from the community and encompass buildings and energy use, clean air, water use, parks and open spaces, The administrative boundaries of the city (blue); existing LiDAR buildings data coverage in metro Phoenix by block group (red); and the census block groups constituting the study area (green). Table 1. How UA can contribute to Phoenix's sustainability goals. Connects the City of Phoenix's sustainability goals, associated desired outcomes through deployment of UA by identifying the specific delivery mechanism and metrics. ( * ) City of Phoenix (2017a). local food systems, transportation, and waste management. Along with each goal are associated desired long-term outcomes.
We focus on three of these sustainability goals for which UA can help deliver the associated desired outcomes. These are (1) local food systems, (2) open space provision, and (3) building and energy use. We identify the delivery mechanism and associated metrics to evaluate the effectiveness of UA in meeting these desired outcomes (table 1). We then quantify these metrics on an aggregate city (study area) scale and analyze and report them at the block group level to pinpoint areas of the city that could benefit from prioritized deployment of UA. This method considers the variations in available area for UA as well as resource and income disparities in different parts of the city.

Available area estimation for UA
We focus on underutilized exterior spaces suitable for UA. Following Clinton et al (2018), we consider three types of urban space available for UA: undeveloped (unpaved) vacant lots, flat rooftops, and building façades (vertical growing spaces). Rooftop UA infrastructure involves a substrate made up of specialized membranes and drainage barriers for growing vegetation on top of buildings. The façade format grows vegetation alongside exterior walls of buildings, using infrastructure like trellises or cages to support plants. Both of these formats have been employed in UA applications (Thomaier et al 2015). We exclude other forms of vertical UA within buildings such as controlled environment agriculture (CEA; Goodman and Minner 2019) or building-integrated agriculture (BIA; Gould and Caplow 2012).
Available area from vacant lots is based on a detailed inventory of all vacant property-built, paved lots, or undeveloped lots-in Phoenix metro area compiled for 2010 and 2017 following Smith et al (2017). The available area from rooftops and vertical surfaces are derived using the 2014 (most recently available) LiDAR building footprint data for Phoenix. We do not consider alternative land use patterns (beyond what current data reflects) for assessing food supply. (See SI-1-1 and SI-5 for details available online at stacks.iop.org/ERL/14/105001/mmedia).
We derive total potential area (A t ) available for UA by aggregating areas for the three types of spaces (i) at the block group level and then summing across block groups (1). where j denote block groups (N=910) in our study area and i denote types of spaces considered for UA with 1=vacant lots, 2=rooftops, and 3=building façades.
2.4. Local fresh produce supply from UA We identify 34 suitable crops consisting mostly of vegetables and fruits grown in Maricopa County (Census of Agriculture 2012). Field crops (e.g. alfalfa, cotton) or tree crops (e.g. citrus) are excluded because of potential space and time (crop growth) limitations under UA. We rank the selected crops using three criteria on an equal-weighted basis: estimated water use (mm/m 2 ), historical yields (ton/ha), and supermarket retail prices from 2016 ($/ton).
Crops with higher yields, lower water use, and higher prices are preferred, and therefore ranked higher. The combined preference for high yields, low water use (highest cost input) with high retail prices in ranking crops is our proxy for profitability.
We next sort the 34 crops and form terciles of suitability: good, average, and poor. Planting area is allocated proportional to suitability: e.g. crops in the 'good' group are allocated more area than those in the 'poor' group. All UA sites get the same crop-area allocation, except for façades, which are limited to a subset of these crops (22 out of 34) since some crops are less suitable for vertical cultivation. (See SI-1-2 for details.) Total annual food supply from UA is the sum of crop production at block group level (2): where A ij is the area in m 2 by type of UA space i (e.g. vacant lots) as described previously. Each crop's weight (weight c ) is proportional to the total rank of all crops suitable for UA (3)

Open green space provision
We consider any open green space provision by UA to be associated with the use of undeveloped ( We compute vacant lot area that will be counted towards open green space provision in two ways. First, we consider the impact of deploying all available vacant lots for UA, regardless of lot size. Second, we place a lower limit on vacant lot size (5 000 m 2 ) and prioritize UA deployment on fewer but larger lots that could more easily incorporate park-like features. We refer to these larger vacant lots (either on an individual or aggregated basis) as 'vacant sites'.
Using this sub-sample of vacant sites, we further estimate how open green space provided by UA can improve walkability using buffer analysis (see SI-1-3 for details).

Energy savings and avoided emissions from buildings
Rooftop applications of UA can reduce the energy usage of and emissions from buildings via improved insulation. The UA substrate added on rooftops increases thermal resistance (R-value) of roofs and reduces building energy use. We follow Clinton et al (2018), modifying it with locally available data for Phoenix.
The difference between total kWh electricity use before UA and after UA provides an estimate of avoided building energy use. Using CO 2 emission rates from power generation in Arizona, we also derive corresponding estimates for avoided CO 2 due to UA. Further methodological details are provided in SI-1-4. Additional local/regional data used in the analysis and interpretation of results are described in SI-4.

Results and discussion
3.1. Potential area available for UA In our study area, using 2017 data, we estimate a total of 7230 hectares is available for UA, representing 7.1% of the land area, and 5% of total surface area (including surface area of buildings). More than half (53%) of this total area is provided by rooftops, 29% by vacant lots, and the remaining 18% from building facades. Thus, 71% of the total available area for UA is supplied by existing building stock (table 2).
Grouped by type of space available for UA, our study area contains 10 708 individual undeveloped vacant lots, over 20 000 suitable rooftops with roof areas >464.5 m 2 (5000 ft 2 ), which represents 35% of all flat rooftops by area, and more than 432 000 individual buildings. Not all block groups have all types of space available for UA: of the 910 block groups in our study area, we document vacant lots in 565, suitable rooftops in 843, and buildings in 907, and only one block group has no available UA area of any type.
The area of block groups ranges from 85.5 to over 6500 hectares in Phoenix and affects the total available area for UA. Controlling for size, the block groups with largest available area from vacant lots are concentrated in southwestern Phoenix (figure 2(A)). Block groups with the greatest rooftop area are found along the main commercial-industrial corridors of the city ( figure 2(B)). The block-group distribution of UA area from façades (figure 2(C)) is correlated with the number of individual buildings, and therefore favors more densely settled residential districts with smaller lot sizes.
Compared to Clinton et al (2018), which completed a similar area assessment for UA at the global scale, the contribution to total UA area from vacant lots is substantially lower (29% in Phoenix versus 80% globally), meaning that area from rooftops and façades is higher (71% in Phoenix versus 20% globally). This is partly driven by the fine-scale inputs used in area estimation, highlighting the need for city-scale assessment.
The total area for undeveloped vacant lots in Phoenix decreased from 2010 to 2017 by 23% following the economic recovery. Since we apply the same crop mix to all vacant lots available for UA, this means the corresponding food supply would have been 23% higher in 2010. Notably, the reduction in vacant lot area over this period is not uniformly observed across the city. While the majority of block groups (N=345) experienced a reduction in undeveloped vacant lot area, others increased (N=161) (see SI-5 for details).

Local fresh produce supply from UA
The total estimated food supply from UA by primary weight (raw weight inclusive of the weight of peel, skin, pit, seed, and stem, depending on the item) is 182 983 tons yr −1 , 89% of which comes from vegetables and the remainder from fruits. Food supply varies widely on a block group basis, ranging from under 1 ton to more than 6654 tons. The amount of produce correlates with block group area, where larger block groups have, in general, more available area for UA and therefore supply more output (figure 3(A)). Normalizing food supply by the available area for UA produces a more compact distribution, ranging from 21.8 to 26.1 tons ha −1 of available area (figure 3(B)), with a block group average of 24.4 ton ha −1 . Higher and lower output block groups are clustered both in terms of absolute tons and on a tons/ha of available area basis (with Moran's I z-score=13.1 p<0.00 and z-score=12.0 and p<0.00, respectively). Higher output block groups are clustered in south Phoenix, and lower output ones tend to be in northern parts of the city.
The average output estimated in our analysis (24.4 ton ha −1 ) is consistent with recent field data from urban gardens with broadly comparable food baskets On a per capita basis, the food supply obtainable from UA is 125 kg/person. The median supply at block group level is 68 kg/person, with block group means ranging from 0.25 kg/person/year to over 1000 kg/person/year. In densely populated residential neighborhoods there is generally less area available for UA, which generates smaller per capita estimates. Conversely, some of the larger block groups are less densely populated, and thus have more area available. Not every block group can produce meaningful amounts of fruits and vegetables from UA per person, and a few block groups produce very large amounts ( figure 3(C)).
Nationally, the per capita intake of non-citrus fresh fruits and fresh vegetables is 138.7 kg/person/ year (2016 figures on primary weight basis from USDA 2018a). The estimated output from UA can meet 90% of this demand for fresh produce.
More than 85% of Arizonans do not consume the recommended amount of fruits and vegetables (Lee-Kwan et al 2017). Therefore, UA can be an important pathway for provision of fresh produce in areas where there is limited local supply (e.g. food deserts), and could supplement consumption in other areas. UA can also satisfy the increased consumer interest in local foods, based on economic evidence (willingness-topay that falls with the transport distance for food; Grebitus et al 2013). This is noteworthy because over half (51%) of the study area population lives in the 439 block groups which overlap with known food deserts (see SI-4-2). We estimate food supply from UA in all 439 of these block groups, constituting 55% (101 008 tons) of the total UA food supply ( figure 3(B)). Notably, the per hectare food supply is also higher in these block groups by 0.4 ton (significant based on median test with Pearson's X 2 statistic=18, P<0.00). Average food supply from UA is higher on a per capita basis, but this is driven by a few large and sparsely populated block groups. Statistically, the group medians are not different (66 kg/person/year in food deserts versus 69 kg/person/year in non-food desert block groups). As figure 3(C) shows, some of the outlier block groups are food deserts.
The food desert block groups represent a segment of Phoenicians with lower median household incomes and more reliance on SNAP benefits compared to non-food deserts. Residents also tend to have lower rates of health insurance coverage, and experience higher rates unemployment. Median home values are also lower. (All differences significant per median tests with Pearson's X 2 statistics of P<0.00.) Prioritizing UA deployment in these block groups can therefore be an important strategy to improve sustainability not just from a local food supply but also from a community resilience standpoint. However, this requires a clearer understanding of socioeconomic processes that might be at work. For example, the fact that we observe larger vacant lots and lower home values in food deserts suggests an underlying dynamic where vacancies might be related to lower home values (i.e. low returns for developers), while lower home values might be related to other factors (such as higher crime rates), which in turn might be influenced by high vacancy rates.
UA and sale of UA produce in farmers' markets may increase fruit and vegetable intake (McCormack et al 2010). However, the direct health and other beneficial outcomes (such as lower cost of purchased food) from eliminating food deserts are less clear because these are areas where access and affordability of healthy food options both play a role. Low income residents in food deserts either travel longer to supermarkets, or pay more for food at nearby convenience stores (Ver Ploeg et al 2009). For UA to be impactful, the cost of fresh fruits and vegetables supplied by UA in food desert areas needs to be cheaper than traveling to the nearest supermarket. Allowing UA-based food supply and UA-vendors to be SNAPeligible could increase the opportunities for lowincome residents in food deserts to benefit from UA.
Two sensitivities to this baseline analysis are also considered. One ranks our list of suitable crops based on productivity alone, and the second with respect to national consumption patterns. While food supply increases (by 16% and 6% respectively, for the high productivity and national consumption pattern scenarios), so does water use (SI-2). Thus, our baseline approach is more sustainable, albeit more conservative from a supply standpoint.
We also convert our weight-based analysis of food supply into calories. Under UA, less of the highly consumed, calorie-dense crops like potatoes are produced, compared to national consumption patterns. As a result, UA food supply provides 46% fewer calories from fruits and vegetables than the US average (42 kcal/person/day versus 92 kcal/person/day). However, on a cup-equivalent basis, which is the metric for determining recommended daily intake levels, UA food supply (1.40 cups d −1 ) surpasses the national average (1.38 cups d −1 ) (US Department of Health and Human Services and US Department of Agriculture 2015; also see SI-3).

Open green space provision
There are 154 existing public parks in Phoenix, covering about 12% of our study area (12 467 ha). Most are local community parks (N=144) and a few are large regional parks (e.g. the South Mountain Preserve; N=10). The regional parks constitute the bulk of the public park area in the city (90%), but are not within daily reach of most residents. The block groups adjacent to these regional parks represent only 6% of the city population.
Most block groups in our study area (78%, N=713) do not have any public parks that residents can access to on a day-to-day basis and are thus underserved. If all suitable vacant lots in our inventory are developed as urban gardens, the number of block groups without any prior green open space can be reduced by 433, decreasing the proportion of underserved block groups to 31% ( figure 4(A)), which will also expand total green open space to 14% of the study area.
Vacant sites (N=764), consisting of either individual or aggregated adjacent vacant parcels with an area greater than 5000 m 2 , constitute 85% of vacant lot area available in Phoenix. If the deployment of UA is strategically prioritized for these larger vacant sites, the reduction in number of underserved block groups is lower, though still substantial (N=216, figure 4(B)). UA applications in these larger vacant sites can emphasize features like pathways that would encourage interaction with residents. Note that if UA deployment is restricted to these vacant sites alone, this would lower food supply from UA by 50%, concentrate benefits to fewer block groups, and reduce UA's coverage of food deserts (from 439 to 122 block groups).
Additionally, the walkability (to nearby parks or green open spaces) benefits from UA will accrue mainly from the 764 vacant sites identified for strategic deployment (even if UA is deployed fully throughout the study area). Based on existing parks, we determine 30% of our study area is either green open space or within 5 min walking distance of such spaces using a 402 m (0.25 mile) buffer zone. Under the assumption that the larger vacant sites are converted into UA, the proportion of such walkability zones increase by 25%-55% (figure 5). In our study area, existing parks overlap with block groups with a combined population of 306 773 (20% of the total population). The added green open space under prioritized deployment increases the number of residents served by 391 153, raising it to 48% of the total population. Not surprisingly, large vacant sites utilized for prioritized deployment tend to be in block groups with lower population density and fewer households (compared to the Phoenix median). To improve walkability in denser block groups, additional strategies to UA may need to be considered.

Energy savings and avoided emissions from buildings
In our analysis, the avoided energy (electricity) use and associated emissions reductions are sensitive to and vary nonlinearly with the existing thermal insulation on rooftops (R o values), since the insulative benefits from added UA substrate (R a =0.25) are assumed to be constant. We consider three levels of insulation for existing roofs: well-insulated (high R o =3.66); average insulation, which is our baseline estimate (using average R o =1.84), and poorly insulated (low R o =0.25).
Typical R o values likely vary across buildings and neighborhoods. To establish a range of values, we first consider all rooftops suitable for UA to have the same starting thermal insulative properties ( figure 6(A)). Under these three scenarios, the annual avoided electricity consumption due to rooftop UA ranges from 33 712 MWh (using high R o ) to 3859 475 MWh (low R o ), with a baseline value of 125 451 MWh (average R o ) ( figure 6(A)). The lower and upper bounds represent 0.1%-13% of electricity sales to commercial endusers in Arizona, and about 0.2%-23% of total estimated electricity consumption in our study area. Corresponding CO 2 emissions avoided range from 13 778 to 1577 315 tons, with a baseline estimate of 51 270 tons. The baseline represents 1.7% of commercial building GHG emissions in Phoenix in CO 2 -equivalent terms (SI-5-1). At building-level, the baseline electricity savings we estimate are 6237 kWh yr −1 . This is about 3% of the annual site consumption of an average US commercial building (US EIA 2016).
Poorly-insulated buildings are a driver of high energy burdens (share of energy expenditure in total spending). The median Phoenix household has a higher energy burden (4.2%) than average among a sample of 48 large US metro areas (3.6%) (Drehobl and Ross 2016). On the other hand, Phoenix had the nation's highest share of EnergyStar certified new single-family homes in 2018 (58%; EnergyStar 2019). This suggests Phoenix likely has a mix of buildings with different insulative properties. Therefore, treating the savings from UA based on the average R o as baseline is realistic.
Alternatively, we assign the high, average, low R o values to block groups based on median year housing structures built in that block group ( figure 6(B)). The underlying assumption is that older buildings will have poorer insulation. This yields total avoided electricity use of 1104 329 MWh and avoided emissions more than 451 324 tons (SI-7). Accuracy of the energy and emissions savings estimation would improve with higher spatio-temporal resolution weather data: e.g. by computing hourly energy flux at a finer spatial scale (<1 km).

Meeting Phoenix's desired outcomes
We summarize the contribution of UA to Phoenix's sustainability goals by revisiting table 1 and adding our findings (table 3, Results).
UA can supply nearly 183 000 tons of fresh produce per year in Phoenix. This amount meets 90% of the current per capita fresh vegetable and fruit consumption in our study area. There is potential for locally-produced food from UA in all block groups that overlap with food deserts, constituting 55% of total UA food supply. Rooftops provide the largest area for UA in Phoenix. As we do not know the proportion of selected rooftops that are structurally suited to carry the incremental UA loads (substrate plus applied water), there is some uncertainty about the UA food supply from rooftops. Also, specific design and substrate choices could influence the magnitude of food supply and environmental benefits Through the use of vacant lots for UA, the number of underserved block groups is reduced by 433, and green open space area is increased by 17%. Using larger vacant sites to develop park-like UA spaces, walkability zones are expanded by 25% to cover 55% of the study area. The incremental green open space access zones provided by UA more than doubles the number of residents with access (391 153 in addition to 306 773 existing), representing 48% of population in our study area. For UA to contribute to improved walkability, access rights to private property need to be negotiated between landowners and UA operators. Neither the UA operators, nor the landowners might be willing to seek or grant unlimited access to the general public. Landowners would be concerned with establishment of public rights over their land where there were none, and UA operators might be concerned with issues like food contamination and safety.
UA can also deliver energy savings and associated reductions in emissions from buildings. The desired outcome is an 80%-90% reduction in emissions from transport, buildings and waste (from 2012 levels). The baseline savings through UA represents 0.4% of commercial sector electricity consumption for Arizona, and avoided emissions are 1.7% of GHG emissions from commercial buildings in Phoenix (City of Phoenix 2017b). Our baseline savings in energy use are within range of similar building-scale estimates (e.g. Wong et al 2003). The reductions are highly sensitive to the existing rooftop insulation, with substantial benefits for poorly insulated buildings. UA operations integrated with existing building systems have the potential to deliver larger savings in energy and emissions (Specht et al 2014). Other rooftop sustainability interventions such as cool roofs can deliver higher reductions in cooling energy demand (e.g. up to 14% city-wide when all roofs are completely light-colored as in Salamanca et al 2016, and ranging from 14% to 30% at the building-scale based on Synnefa et al 2007, Xu et al 2012. Yet, net benefits of cool roofs over the year tend to be lower because of increased heating needs during winter (Karl et al 2009, Georgescu et al 2014. To some extent, savings are also limited by our assumptions. We consider flat rooftops greater than 464.5 m 2 (5000 ft 2 ) that would be most suitable for commercial-scale UA, and not all flat rooftops. Likewise, possible variation in micro-scale temperatures is not captured in our analysis, because of a lack of high spatio-temporal resolution weather data. We also do not account for additional building energy use from rooftop UA applications, which may reduce net benefits, thereby yielding lower desired outcomes than those specified here.
Even if the estimated benefit appears small in proportion to the desired outcome, it grounds expectations from the intervention considered (here, the scale of energy and emission savings achievable through UA), and underscores the need for alternative measures to achieve that goal.

Economic and policy considerations
Deploying UA as a sustainability strategy also requires an understanding of costs and trade-offs. For most urban areas, the availability and cost of suitable spaces for UA is the primary consideration and a binding constraint. Land tenure affects the level of investment made by a farmer. In a survey, urban farmers in the US reported a preference for land ownership, but due to high land values, they often lease or acquire temporary user permits (Pressman et al 2016).
Residential land prices can be used as a lower bound for the value of undeveloped or underutilized spaces allocated to UA. In our study area, they range from $217 000/ha to over $3.02 million/ha ($21.7/m 2 to $302/m 2 ) by census tract (Davis et al 2019). The average price of residential land in the Phoenix metro area ($720 419/ha) is lower than the average for the largest urbanized areas (population500 000) in the US ($1270 065/ha) (US Census Bureau 2018b, Davis et al 2019; also see SI-4-4). Combining our block group-level UA area estimates with the census tractlevel land prices, the weighted average cost of 'space' for UA in our study area is $536 760/ha ($53.7/m 2 ). This can be treated as a one-time land acquisition cost, or converted into an annual rent of about $27 000/ha ($2.7/m 2 ), assuming a 5% capitalization rate (see SI-4-4).
In Phoenix, water is the critical non-land input for UA. We estimate that the cost of municipal water (at 0.64 $/m 2 ) would be about seven times the cost of fertilizer per square-meter of cultivated area (SI-6-2). Water use in Phoenix also has a high social cost owing to limited fresh water resources available. The full deployment of UA in our study area would require 35 630 953 m 3 of water per year, adding about 7.2% to municipal water consumption in Phoenix. Yet, this is still less than 0.6% of total state-level water use, 74% of which is consumed by irrigated agriculture (SI-6-1). Nonetheless, opportunities to recycle and reuse water for UA (i.e. incorporating UA into green infrastructure and vice versa) could limit further stress on scarce water resources. Changing residential water use patterns, such as eliminating turf grass and backyard pools, can improve water sustainability in Phoenix (Gober and Kirkwood 2010). Our estimate of UA water use is comparable to the amount needed to maintain mesic landscapes and backyard pools in Phoenix (SI-6-3).
Installation and operating costs for UA can vary highly with the scale and design of the actual UA operation and crop selection. Representative cost estimates come from detailed crop budgets for fruits and vegetables (e.g. https://coststudies.ucdavis.edu/en/ current/) for the US and surveys of rooftop garden applications. The one-time installation costs (excluding land) can range from under $10 (on vacant lots) to $30-$200 per square meter (for rooftop operations) and annual operating costs can range from $1/m 2 to $14/m 2 (Rosenzweig et al 2006, Carter and Keeler 2008, Conner and Rangarajan 2009, Blackhurst et al 2010. Use of permanent or semi-permanent structures (a hoop-house, fixed planters, fencing, cooler, or storage sheds) would lead to higher installation costs. Similarly, site preparation can raise upfront costs if any contamination remediation is needed, or the site needs re-zoning. Automated water and nutrient delivery systems would also increase installation costs, but might reduce operating (labor and material) costs.
Local governments and residents also need to consider the opportunity costs of city-scale UA with regard to tax tradeoffs, provision of local food versus affordable housing, rooftop solar energy, cool roofs and other development strategies.
Removing barriers to UA could include strategies such as clarifying zoning for intra-city farming activities that are conducted for profit versus not-for-profit (subsistence farming, educational farms) and expanding building codes to codify structural requirements for adding live substrate on rooftops.
Some jurisdictions also consider policies that create explicit incentives for commercial UA. These may target property taxes. For example, California's AB551 (2014) legislation allows landowners in metropolitan areas to receive tax incentives for placing their land in agricultural use. In Utah, SB122 (2012) allows qualified parcels in Salt Lake County to be assessed at lower property rates provided the land is used as for-profit farming. Purchase requirements that support local farmers may also benefit UA. For example, in Illinois HB3990 (2009) calls for 20% of all food products purchased by state agencies and state-owned facilities to be local farm or food products by 2020.

Conclusions
We combine fine scale data on available urban surfaces (undeveloped vacant lots, façades and rooftops) with Phoenix-or Arizona-specific inputs to determine the contribution of UA towards meeting Phoenix's desired sustainability outcomes. Our findings show that the total UA food supply in the fifth largest city in the US meets the annual consumption of fresh produce based on national per capita consumption patterns. UA can also supply local produce in all food deserts of Phoenix, and more than half of all output would be produced in the half-mile LILA census tracts identified by the USDA's widely-used national assessment of food deserts. This assessment likely overestimates food access in poor neighborhoods because of spatial aggregation (Bao and Tong 2017), meaning locally produced food from UA might prove more impactful than we estimate here.
UA would also add green open space for Phoenix residents. Full deployment of all available vacant lots can reduce number of underserved block groups by 60%. Larger vacant sites converted into UA could include park-like features, which would bring 55% of the study area within 5 min walking distance to a green open space (up from 30%). Rooftop deployment of UA could reduce electricity use and associated CO 2 emissions from buildings and displace more than 50 000 tons of CO 2 based on average roof insulation.
An advantage of UA as a sustainability solution is its potential to improve and encourage multi-purpose utilization of existing capital stock (infrastructure, buildings) within an urban area. In addition, there are several co-benefits from UA that are not explicitly evaluated here. These include public health benefits (possible reduction in obesity rates and diet-related disease) to the extent such potential health and nutritional benefits are a function of improved food access; benefits from shortened supply chains such as lower emissions from transport of produce (i.e. food miles); and local dust mitigation and runoff control from cultivation of otherwise underutilized vacant lots.
Today, the fastest growing cities in the world are those with fewer than 1 million residents, rather than megacities (>10 million residents). By 2030, more urban residents will live in small to mid-size urban areas (500 thousand to 5 million residents), similar to Phoenix. In the US alone, the number of cities with 1-5 million residents will grow 25% by 2030 (United Nations (UN) 2018). It is therefore important to examine sustainability solutions in small and mid-size city settings.
Yet, UA cannot be a sustainability panacea for all cities, and we are not advocating that urban areas should be used for growing crops. Our goal is to demonstrate the value of a data-driven approach that involves using fine-scale, local data to evaluate UA. Ultimately a database of comparative city-scale assessments is needed to show why UA can be a sustainability solution in one context and not others (Webster 2018).