The emergence of indoor agriculture as a driver of global energy demand

Abstract

Cultivating crops in mechanized greenhouses and windowless plant factories – a practice known as controlled environment agriculture (CEA) – is widely advocated as a way to enhance sustainability. Yet CEA provides less than 1% of the US food crops while consuming more energy than all open-field cultivation, potentially expanding tenfold to ~7% of national energy use. This meta-analysis of 116 studies spans 40 countries and Antarctica, and 23 crop types. Energy per harvest weight varies by five orders of magnitude, depending on facility type, crop, geography, and other factors. Improvements in energy efficiency are constrained and transitioning to solar energy would require three-times more land area than open-field cultivation, negating one of CEA’s prime intended benefits. Due primarily to energy using systems, CEA has significantly higher capital and operating costs than open-field cultivation, translating to higher commodity prices and substantial business risk. Particularly high energy intensities have rendered grains, root crops, and other important staple crops nonviable. Sustainable outcomes require increased scrutiny by planners and policymakers, informed by improved assessment rigor and more objective comparisons of open-field and CEA practices.

Introduction

For 10,000 years, agriculture was conducted exclusively outdoors, gradually becoming a major driver of global environmental change. Consequently, indoor agriculture was explored as a means to potentially reduce land and water use, provide protection from environmental hazards, extend the growing season, and garner other benefits. In the first century CE naturalist Pliny The Elder wrote that Roman Emperor Tiberius Caesar grew melons in greenhouses glazed with translucent stone¹. Greenhouses were first heated in Korea during the 1450 s to “force” crop growth². Electric lighting was introduced in the 1800s, coming into widespread use in Norway, Russia, and Sweden by the late 1960s and later in Holland and Japan along. Experimental multi-level glazed towers for food production appeared in Poland around this time^3,4,5. Interesting applications have been implemented in harsh environments, including skylights in the deck of Robert Scott’s 1901 tall-ship expedition⁶ to Antarctica that illuminated vegetables growing below, and artificially conditioned insulated shipping containers with are deployed today near both poles^7,8. In modern times, indoor-cultivation practices became increasingly industrialized thanks in part to research with an eye to supporting life under the sea or in space⁹ with subsequent adoption for commercial agriculture. However, considerable energy costs and greenhouse-gas emissions are now the most often-cited business constraints and negative environmental impacts of modern indoor agriculture.

Low-tech “open” greenhouses rely on outdoor environmental conditions and involve only minor energy uses such as mechanical ventilation and irrigation, while “closed” greenhouses and windowless structures (referred to hereafter as “plant factories”) are more tightly controlled, which requires energy for artificial lighting, heating, cooling, dehumidification, and other purposes. Facility sizes vary from small cabinets to tall warehouse-sized structures covering tens of thousands of square meters. Further expanding the cultivation area, most plant factories contain multiple upright or horizontal cultivation surfaces, a configuration commonly referred to as “vertical” farming. Cultivation media ranges from soils to synthetic materials, which can have considerable embodied energy. While definitions and terminology vary widely, the term “controlled environment agriculture (CEA)” is adopted here to broadly indicate the artificial maintenance of growing conditions across a range of facility types and degrees of mechanization¹⁰.

This article assembles a meta-analysis of available CEA energy-use data globally, together with discussion of prospects for energy efficiency and renewable energy, methodological issues, and other topics raised by this emergent literature. These studies adopt a wide variety of system boundaries (e.g., “on-farm”, “cradle-to-farmgate”, or “cradle-to-grave”) with corresponding life-cycle assessment (LCA) approaches. The focus here is on-farm energy use—typically the greatest contributor to life-cycle environmental impacts and operating costs—which has unique drivers, approaches to identifying energy hot spots, and methods for improving efficiency. The lack of consensus on whether CEA yields net environmental benefits compared to open-field practices makes improved understanding of on-farm energy use a key need¹¹.

Only a small subset of studies addressing facility-level CEA energy treat the matter quantitatively. This review provides a substantially larger and more diversified compilation of published energy use estimates than prior efforts^{11,12,13,14,15,16,17}, revealing far broader variations in energy use intensities over a wider range of geographies, facility characteristics, and crop types. Among past assessments, an urban-centered review by Milestad et al.¹⁴ identified 83 studies, with energy use qualitatively addressed in 20 cases. Dorr et al.¹¹ examined 36 urban studies, evaluating combined on-farm and off-farm energy use data in 17 cases. Miserocchi and Franco¹⁷ identified 18 studies, stipulating a variety of values absent in the primary sources. Trudel-Brais et al.¹⁶ summarized quantitative metrics from only 3 of 47 studies reviewed, bemoaning the dearth of information. Vatistas et al.¹⁸ gathered energy estimates for isolated end uses (e.g., lighting) but not aggregate values. In many other cases, e.g., Roussis et al.¹⁹, carbon emissions are reported but not the underlying energy use, which removes the opportunity to isolate the influences of fuel mix, energy efficiency, and the widely varying carbon intensity of electricity production across diverse geographies. It is particularly problematic that no studies compare best-practice outdoor cultivation (yields or energy efficiency) to optimized CEA processes.

In quantitatively examining the energy outcomes of varying CEA applications, the current study helps address a critical lack of evaluation in the existing literature pointed out by others^11,15, and enables development of an illustrative national CEA energy demand estimate for the US. No previous estimates of national CEA energy use across all crops and facility types were identified for any country.

Results

Meta-analysis of a broad literature

While thousands of publications discuss indoor agriculture, only 748 identified through bibliographic search reference energy use, among which 116 provide adequate data for analysis and intercomparisons. The 154 evaluated locations span a broad range of climates and potential solar energy resources (latitudes −0°S to 71°N) in 40 countries and Antarctica, with few located in lower-income countries (Fig. 1) (Table S1). These studies emerge from diverse research communities, the earliest dating from 1993 followed by low publication levels throughout the ensuing 25 years, after which activity increased markedly.

Fig. 1: Studies providing usable quantitative assessments of energy use in controlled environment agriculture (CEA).

a Map marker size and shading density reflect the number of measured or modeled cases at each location (two sites in Antarctica are not shown). Data sources provided in the Supplementary Information. b Assignment of research domain corresponds to the disciplinary focus of the publication. Reasons for the increase in publication activity likely reflect rising concern about the costs of energy and its environmental impacts, as well as increasing focus on whether CEA can contribute to climate resilience.

In reviewing the literature, no limitations were placed on geography, cultivation scale, or type of crop. Some were research or demonstration facilities, but all were represented as relevant to commercial production. Inclusion of studies in the compilation required sufficient data to determine aggregate energy use and yields for an entire year (often involving multiple cultivation cycles) and to compute normalized energy per unit yield (MJ/kg) or energy per unit production area (MJ/m² growing surface). Cases were excluded where major fuels or uses were omitted, seasonal variations not accurately captured, energy needs reported only indirectly in terms of greenhouse-gas emissions, or commercial “on-farm” energy use combined with that of labor, solar gains, or off-farm uses. Some sources were excluded where authors confused energy use with peak electricity demands and/or results were expressed in terms of loads, i.e., not accounting for on-site conversion efficiencies.

The identified literature includes measured or modeled energy use for 486 cultivation trials and 23 crop types, with 57% of the estimates being model-based and 43% measurement-based, the latter cohort representing 1686 production sites with an aggregate cultivation area of 14.6 million m². Of all cases, 59% were in greenhouses and 41% in plant factories. Energy use values for open-field cultivation use were offered for comparison in only 12 of the studies.

Orders-of-magnitude variation in energy use

The compiled energy intensities exhibit extraordinary variability (Figs. 2a, b and 3a–h). Differences are evident among and within crop types, facility types, and geographies. Other potential causes of divergence include yields per unit of growing area, energy efficiency, and various lighting factors including type, contribution of daylight, illuminance levels (PPFD), and artificial lighting on-times (photoperiod).

Fig. 2: Ranges of annual energy use intensity per yield and per unit growing area by facility type, energy efficiency features, modeled vs measured data, and crop type.

a Eenergy use per unit yield indicates cultivation energy intensity, while (b) energy use per cultivation area indicates facility energy intensity. Boxplot threshold percentiles: 5th, 25th, 75th, 95th with “X” indicating medians. Specific crops included in the “Leafy greens” category: arugula/rocket, chicory, spinach, miscellaneous brassicas. Specific crops included in the “Other and mixed” category: eggplant, potato, and unspecified mixtures of crops included in a single trial. The number of cases underlying each range bar are noted in the x-axis labels. Data sources are provided in the Supplementary Information.

Fig. 3: Annual energy use intensity per unit yield and per unit of growing area, by yield density.

Results by crop type are shown in (a, b), by facility type in (c, d), by presence or absence of energy efficiency measures in (e, f), and by modeled versus measured analysis in (g, h). Each symbol indicates a cultivation trial in the meta-analysis for which the corresponding metric can be calculated. Yields reflect total annual amounts harvested per growing surface (kg/m²-y) which in most cases occur multiple times each year and often on more than one growing level or surface, particularly in plant factories. Cases lacking yield density are not plotted. Corresponding values for open-field crops provided in some of the CEA studies are shown in (c, d). The tendency towards lower yields and lower energy intensities for open-field cultivation is evident. Sample sizes are indicated in the legends. Crops included in some of the aggregated categories are defined in the caption to Fig. 2 and the data and sources are provided in the Supplementary Information.

Lettuce and tomatoes exhibit loosely overlapping energy intensities, while herbs and leafy greens tend to be somewhat more energy intensive, cucumbers the least energy-intensive, and cannabis the most energy-intensive crop by orders of magnitude (Fig. 4a, b). Approaching the lower bound of cannabis are soybeans and wheat (gray and black markers) which are both found to be uneconomic given the high required energy inputs–despite use of LED lighting, high-efficiency heating and cooling systems, and a well-insulated facility envelope in the case of wheat. Microgreens are also among the more energy-intensive crops.

Fig. 4: End-use allocation of total energy consumption for a variety of controlled environment agriculture (CEA) operations.

Each bar represents one site and cultivation trial. HVAC (heating, ventilating, and air-conditioning) is noted where these three categories are combined. “On-site CO₂” pertains to production and/or distribution of CO₂ in the cultivation environment to increase growth. The “Other” category can be large when significant uses are uncategorized in the source publication. Data sources are provided in the Supplementary Information.

When grouped by facility type (Fig. 4c), greenhouses are the least energy-intensive (median 27 MJ/kg), followed by plant factories (127 MJ/kg) which includes 23,300 MJ/kg for cannabis and 78 MJ/kg for non-cannabis crops. While not always identifiable in the source literature, less-mechanized “open” greenhouses appear to operate between 1.5 and 5 MJ/kg. Yield intensities (kg/m²-y) in greenhouses fully overlap with those of plant factories, with far less energy input, an observation that has been made by others²⁰. For comparison, the median value for open-field crops is only ~1 MJ/kg.

Depending on metric, median energy intensities of facilities without identified energy-efficiency measures are only 1% (MJ/kg) and 3% (MJ/m²-y) higher than those with one or more efficiency features (Fig. 3a–f) although the variation is expectedly greater for those not identifying the presence of energy efficiency features (Fig. 2a, b). While individual side-by-side studies clearly show energy savings from efficiency measures, many other factors appear to have more influence on the overall sample outcomes. Energy intensities are uncorrelated with facility size for both metrics (Fig. S1).

No substantial differences between energy use for modeled and measured results are evident (Fig. 3g, h). Detailed validations of models against measured data for a given facility are uncommon. Model results appear to assume systematically higher annual yields — which translates to lower energy intensities — perhaps reflecting idealized conditions and neglecting real-world factors such as crop losses.

Energy consumption at the end-use level is disaggregated in 191 cases (Fig. 4). This information is valuable for understanding drivers of energy use and identifying opportunities to improve energy efficiency. For example, dehumidification represents more than half of total energy use for cannabis cultivation in hot and humid locations while heating is negligible compared to cold climates where dehumidification is minimal and heating represents about two-thirds of total energy use. Wide variations in end-use shares are notable even for a given crop or facility type.

This large and diversified sample reveals findings not captured by less-comprehensive studies, and distinguishes energy intensities by crop and facility type. It also provides a counterpoint to claims based on far smaller samples that energy use is largely insensitive to climate^{13,21,22,23,24}. End-use analysis also helps scrutinize rules-of-thumb, e.g., that all plant factories have the same end-use breakdowns²³ or the assumption that a multiplier of 1.25 can be readily applied to lighting energy to obtain total energy use^25,26,27. In practice, per Fig. 4, artificial lighting can constitute between ~0% and ~95% total energy use.

Extreme CEA: Arctic and Antarctic applications are particularly energy-intensive

CEA facilities are operated by five countries in Antarctica²⁸ to produce food and conduct research for space-based applications. Interestingly, despite minimum temperatures of −43.5 °C, greenhouses with ultra-high-efficiency envelopes are employed in some cases to take advantage of daylight and minimize fuel use or costly solar installations. Measured energy intensities of ~17,000 MJ/kg for mixed vegetables⁷ are among the highest in the meta-analysis, and more than 200-times greater than the median for food production.

Energy intensities of ~6000 MJ/kg occur for cucumbers in container farms in the Canadian arctic⁸, where fresh produce must otherwise be flown into Inuit villages. Despite the waste heat from lighting and dehumidification meeting nearly all heating needs, nearly 50 kW of on-site renewable energy and battery storage, plus diesel generator backup, is needed to support this modest 6.7 m² growing area.

In Icelandic greenhouses, heat is required year-round, with supplemental lighting much of the year, along with summertime cooling²⁹. Significant additional energy use is associated with melting snow from greenhouse glazing. Energy intensities are 2600–5800 MJ/kg for low-calorie crops such as lettuce and tomatoes.

The cannabis conundrum: the most energy-intensive crop

Cannabis emerges as the most energy-intensive crop (Figs. 2 and 3), and represents approximately 95% of present-day US CEA crop value^30,31. About 63% of this cannabis (legal and illicit) is produced in CEA facilities, representing 17% of the country’s overall CEA facility footprint. Little data on cannabis cultivation is available for other countries.

Likely drivers for the relative energy inefficiency include the yield being measured in dry- rather than wet-weight, high crop losses from interdiction and other factors, high proportions of unused and woody biomass (low harvest index), long cultivation cycles (~70 days) and photoperiods (12–18 h/d), higher lighting intensities (PPFDs up to 1323 micromoles/m²/s) creating corresponding elevated cooling needs, lower tolerable humidity levels (50–60%), unsuitability for multi-layer vertical farming, higher air-circulation rates, and higher levels of CO₂ enrichment than most food crops (1000–1200 ppm are common).

In the U.S., cannabis has become the largest cash crop, while releasing CO₂ emissions at the rate of ten million cars with an energy expenditure of $11 billion each year³¹. While there is general debate about the environmental merits of urban agriculture¹¹, the expansion of factory-farm cannabis cultivation into cities has raised distinct secondary energy issues, including strain on power grids, significant power outages, and environmental justice concerns³¹.

Notably, the broader literature on CEA crop production is virtually devoid of reference to cannabis, while that on cannabis rarely provides comparisons to other crops. LCA analyses for cannabis are rare^31,32,33.

Energy is a constraint to producing essential staple crops

Multiple studies find that because fruits, cereals, legumes, seeds, root crops, and other high-protein/high-calorie staple crops are significantly more energy-intensive than crops such as lettuce and herbs they cannot be cost-competitively produced under CEA^{15,20,29,34,35}. For example, even at scale, strawberries from CEA facilities are priced at ten-times that of their field-grown equivalents⁴.

In another study, soybeans had among the highest energy intensity²⁶ of all food crops in this meta-analysis. While yields were up to six-times greater than open-field soybeans, energy intensity was 1000- to 2000-times greater, even with energy-efficient LED lighting.

Asseng et al.³⁴ model a hypothetical 18,225 m², 10-story wheat-production facility with high energy efficiency (insulation, heat pumps, LED lights, and heat-recovery ventilation). To maximize yields (63-times greater than field-grown wheat), very high illuminance levels (5- to 10-times that of most other studies in this compilation) are assumed to be delivered 24 h per day, with natural background levels of atmospheric CO₂ tripled (to 1200 ppm) to accelerate plant growth. Estimated energy costs for the facility are $16 M/y based on an inexplicably low assumed price of $0.02/kWh – versus an average US industrial price of $0.09/kWh in summer 2024³⁶ – representing 77% of total operational costs, or $13/kg yield, greatly exceeding the value of the commodity.

Prospects for greater adoption of energy efficient technologies are constrained

In a global survey of greenhouse and plant-factory operators³⁷, two thirds report not employing any energy efficiency strategies. Providing insight as to why uptake is low, another industry survey identified multiple barriers, in order of decreasing severity: upfront costs (compounded by narrow profit margins), access to capital and financing, lack of knowledge, skepticism and lack of trust in the performance of efficient technologies, lack of executive support, and insufficient training³⁸. Cultivators of even the most energy-intensive crops demand, on average, a rapid 4-year payback time on their energy efficiency investments³⁹. While some qualitative assessments are optimistic⁹, model studies document a limited cost-effective energy savings potential of <10% for most measures in cannabis farms⁴⁰, which should be among the most promising applications given the high energy use and expenditures.

The benefits of LED lighting have been found to be highly variable. In field trials, Leichleiter et al.⁴¹ found energy savings ranging from negligible to substantial depending on cultivar. Measurement-based studies for tomato production found LED lighting required augmentation with less-efficient high-pressure sodium (HPS) lighting⁴². The efficacy of LED products appears to also be an issue, given the 2.2-fold variance in yields of microgreens per unit of energy depending on LED manufacturer⁴³.

Facility energy efficiencies among the sites included in the meta-analysis vary widely, with some having single-glazing, combustion heating systems, minimal insulation, low-efficiency lighting, and deficiencies in system operation⁴⁴. However, compared to the aforementioned industry surveys, a disproportionate number of studies in the meta-analysis involved one or more energy-efficiency measures (Fig. 3e, f). Such measures were reported in 50% of the studies and 70% of the cases (individual measurement trials or model runs), including 60% with LEDs, 24% with heat pumps, 15% with insulation, 3% with evaporative cooling, 2% with economizer cooling, and 1% with multiple glazing. The potential for energy efficiency – within a given CEA facility type – is thus already partly captured, while facility operators perceive many barriers to further adoption.

For indoor agriculture (and all other industries) the primary path to full decarbonization requires conversion of all energy uses to electricity, and then producing that electricity with renewable energy. Electrification also offers unique opportunities for energy efficiency (e.g., heat pumps instead of combustion heating). Just under 60% of the facilities reviewed here are all-electric. If not provided for at the time of construction, capital requirements for such conversions are substantial.

Process change can reduce or increase energy efficiency

The choice of CEA facility type (greenhouse instead of plant factory) is perhaps the single-most impactful determinant of energy efficiency. A few studies document other promising forms of process change. These include measures such as reducing photoperiod and light intensity in the case of Stevia⁴⁵, growing strawberry runners outdoors before bringing them into the controlled environment for flowering and fruiting³⁵, and varying fertigation techniques, light levels, lighting location, and greenhouse operation⁴⁶. Light deprivation is commonly used in plant-factory and greenhouse operations to compress cultivation times for cannabis and certain other crops, thereby reducing energy requirements. Pennisi et al.⁴⁷ varied the spectrum of LED lighting, finding variations in energy use for lettuce (1.5x) and chicory (1.2x). Field tests in Australia⁴⁸ found that optimizing planting dates can yield energy savings in cool winters. Increasing yield was found to improve efficiency for tomatoes bred to also thrive at slightly lower temperatures⁴⁹. One particularly effective case study utilized root-zone cooling for cucumbers in the extreme United Arab Emirates climate, lowering energy use by 20-fold while maintaining yields⁵⁰. While AI may play a role in attaining operational benefits²⁴, it also requires substantial energy inputs.

Cultivar choice can also impact energy use, yet only one-quarter of cases in this review identify the exact cultivar, and few of these explore the energy implications. Almeida et al.⁵¹ attribute some of their observed differences in energy intensity to the choice of tomato cultivars. Leichleiter et al.⁴¹ found a 6-fold difference in energy intensity of different types of cannabis cultivated in otherwise identical plant factories. It is proposed that genetic engineering can develop cultivars that grow better under energy-efficient LED lighting⁵².

Site selection is also a key variable. One study⁵³ found that the most energy-efficient approach to cucumber greenhouse cultivation in Norway was summertime cultivation in the southern part of the country, with moveable glazing insulation and no artificial lighting. Replicating that same method in the north increased energy intensity by 51%, with 11% reduced yields.

Conversely, efficiency losses can arise from efforts to increase yields. Early efforts to increase yields in the Netherlands included increased CO₂ enrichment, which was produced by burning fossil fuels on site⁵⁴. Elevating CO₂ levels requires increased light levels to support amplified photosynthesis, which, in turn, boosts transpiration and dehumidification energy needs. In another example, reducing temperatures for tomato greenhouses in a hot desert climate resulted in an 18% increase in energy use for a 0.35% increase in yield⁵⁵. Some studies found that yields level off or even decline appreciably with increasing light inputs, and, conversely, in some cases greenhouses without artificial light achieve less energy input per unit of yield without sacrificing product quality^{20,42,46,56,57}. Microgreen yields declined by 23% in Chile as light application was increased by 144% while energy intensity increased 72%⁵⁸. Qualitatively similar results were observed in Russia⁵⁹. In southern Norway, while intensified cucumber production with artificial lighting, cooling, and year-round cultivation tripled yields, it also increased energy intensity by 88%⁵³.

In a further complication, certain efforts to reduce non-energy environmental impacts can increase energy use. For example, water recovery from exhaust air is energy intensive²⁰ and steam-cleaning soils for re-use rather than disposal⁶⁰ can double on-farm energy use.

Shifting to renewable energy can negate the land-use benefits of intensive indoor cultivation

The elevated yields of CEA facilities are often credited with creating “freed” or “spared” land that can ostensibly be returned to a wild state⁴. In response to concerns about energy use, it is also commonly stated that CEA facilities can become more sustainable if fully powered with renewable energy to achieve net-zero emissions – yet these propositions rarely quantify the associated land requirements.

Individual case studies^{8,34,61,62,63,64,65,66,67} report that on-site solar electricity would require land areas ranging from 2- to 100-times the facility’s floor area – often more than that needed for open-field cultivation. A techno-economic optimization for US-based greenhouses (which are significantly less energy-intensive than plant factories) found the approach not to be cost-effective except in the mildest climates⁶⁸. Moreover, diversion of renewable energy for discretionary indoor cultivation is problematic given the current need to dramatically reduce global emissions in other sectors in order to manage rising climate risk, while the practical constraints on renewable energy expansion are many, ranging from ecological to cultural conflicts^17,20,31.

Figure 5 shows the standardized area required to meet all on-farm energy needs with solar photovoltaic (PV) power for cases in the meta-analysis. In all cases, the array area exceeds the building footprint. The average ground-mounted PV area (panels in excess of what will fit on the roof) required for food-producing greenhouses is 4-times their building footprint, 48-times for food-producing plant factories, and 122-times for cannabis-producing plant factories. In the best cases, highly energy-efficient greenhouses (envelope, lighting, and equipment) in an ideal solar climate (Turkey) required an array area equal to building footprint, but this outcome is rare. The largest solar arrays are required for energy-intensive crops and vertical-farming facilities for food crops with multiple cultivation layers and/or multiple-stories. The extreme case, a skyscraper lettuce-production greenhouse in a sunny Los Angeles location would require a PV array ~600-times its footprint to meet target lighting levels beyond what daylight provides. Inconveniently, multi-layer vertical farming ostensibly spares the most land (albeit before counting the solar requirements)⁴.

Fig. 5: Solar photovoltaic (PV) system ground-mounted area needed to achieve 100% on-site renewable energy production as a multiple of controlled environment agriculture (CEA) facility footprint.

Each data point indicates a case in the meta-analysis for which the metric can be calculated (N = 330). A value of 1 indicates that the required solar array would effectively double the operation’s land use. Sites achieving low values tend to be greenhouses using only daylight, with energy use limited to ventilation and irrigation and perhaps other minor loads. However, since greenhouse structures require translucent roofs, those arrays are assumed to be fully ground-based. Facilities with no exact location or non-electric (or unknown) energy types are excluded. Only the all-electric subset of the sample is shown, with the exception of US and Canadian cannabis sites to define the upper boundary of potential electricity demand, for which existing fuel-based heating (80% efficiency) is modeled as converted to electric heat pumps (COP 3.5). Even without area constraints, on-site applications in lower latitudes would typically require grid interties or short-term diurnal storage, and backup power that may have associated carbon emissions would be needed at higher latitudes with seasonal darkness. Existing facilities using a mix of fuels would need to first invest in costly equipment change-outs that would enable conversion to 100% electricity.

In the US, typical ratios of CEA to open-field yields per unit of land area include: cucumbers 22:1, lettuces 12-21:1, peppers 6-12:1, 4.9:1 for cannabis, tomatoes 4:1, and strawberries 1.4:1^10,69. As can be seen in Fig. 5, the required additional solar land areas are generally much higher than these multiples. Not included in the analysis is land area for electrical transmission lines, fossil fuel mining and processing, or other renewable energy land uses (hydroelectric, wind, etc.). While no all-electric open-field cases were identified, such farms would only require a minor fractional expansion of land area to achieve net-zero emissions with solar PV. Moreover, the ratios shown above compare conventional open-field results rather than best practices (which use less land area), suggesting that optimal open-field cultivation could be even more competitive on a land-use basis.

Energy-determinative dynamics of controlled environment agriculture: technologies and markets

Narratives describing advantages of CEA compared to open-field agriculture inform the assessment of energy needs and environmental impact^4,9,45,70,71. Key stated benefits include control of environmental factors (temperature, light, humidity), shortened crop cycles, an extended growing season, and increasing output with high predictability. The degree to which yields exceed those of open-field farming varies widely as a function of cultivar and practices, and outdoor conditions where competing open-field cultivation would occur¹⁰. Other stipulated benefits include reduced transportation energy in delivering crops to market, exposure to pests and natural hazards, land-area requirements, soil degradation and erosion, fertilizer use, water use, soil-derived greenhouse-gas emissions, odor nuisance, food-safety risks, and commodity shrinkage and spoilage if facilities are located closer to markets. These technical aspirations are often rhetorically tied to broader goals including environmental protection, public health, and affordable and equitable availability of food for a growing land-constrained population.

For each of these stated advantages are caveats and countervailing concerns¹⁵. In some cases, unique collateral problems are associated with energy and mechanization and bringing cultivation closer to population centers, including light pollution and odor nuisance⁷². Reduced transport distances do not predictably lower overall carbon footprints¹⁵. More broadly, in an intrinsic contradiction, greenhouse-gas emissions from CEA’s energy use—which can exceed that of open-field cultivation—exacerbate the very climate risks the practice is hoped to shield against¹¹. Disaster-resilience is an oft-cited benefit⁷³, although CEA has acutely greater vulnerability to utility disruptions (power and water) than open-field agriculture, more costly potential damage to cultivation infrastructure, and substantial added capital cost for backup electricity generation and nuisance while in operation. The advantages of CEA in terms of water and land requirements only hold when moderate open-field yields are assumed and system boundaries omit the land and water required for energy production⁶⁹.

Nonetheless, CEA has high appeal and is well-established for certain crop types and geographies. For example, most microgreens and commercial mushrooms are grown in plant factories. Geographically-isolated Iceland imports about two-thirds of its vegetables, by calorie count²⁹, while the land-constrained island state of Singapore imports 90% of its food⁷⁴. Greenhouses provide one-third of fresh vegetables imported to the US from Mexico¹⁰, ~60% of vegetable production in Iran^75,76, and virtually all non-summer cultivation in Siberia⁷⁷. In Norway, virtually all tomatoes, 90% of cucumbers, and 13% of lettuce are grown in greenhouses⁷⁸. The highest-value crops are the most viable, as in the case of cannabis for which US wholesale prices are ~$2000/kg, yet, energy costs remain a concerning factor and one-third of production continues to be conducted outdoors³¹.

The scale of CEA practices is increasing. Globally, greenhouses occupy an estimated land area of 1.3–3.7 million ha – an unknown fraction being heated or artificially illuminated – of which closed glass houses comprise ~70 kha^79,80. No reliable estimates have been identified for plant factories, but they likely number in the low thousands of hectares. As an illustration at the national scale, CEA production in the US occurs in in ~11,000 greenhouse farms, spanning 19.4 Mm² plus 3.1 Mm² for mushrooms³⁰ and 4.6 million m² for cannabis cultivation in a mix of greenhouse and plant-factory settings⁸¹ (Table S2).

On-farm energy use is a leading determinant of overall climate impacts from crop production (Table 1). Energy uses include electricity, diesel, natural gas, propane, coal, biofuels, on-site renewable energy, and industrial waste heat^{76,82,83,84,85,86,87} Energy needs are driven by artificial lighting, heating, space cooling due to solar gains in greenhouses and substantial waste-heat from lighting, and dehumidification. For certain crops and geographies, greenhouses can avoid the need for artificial lighting, but since artificial lighting can increase yields, and provide sometimes-useful waste heat, many greenhouses rely heavily on it⁸⁷. Secondary energy-using activities include ventilation, irrigation water temperature control, crop drying, water pumping and irrigation, snow-melting, carbon-dioxide enrichment, post-harvest soil steaming for disinfection, water recovery and on-site treatment, harvesting and other on-site crop handling or processing, cold storage, robotics or other automation, and AI or other computing associated with process control.

Table 1 Role of direct on-farm energy in total energy use and environmental footprint

In addition to technical issues are energy-related market challenges. CEA capital costs–much of which trace to energy infrastructure–far exceed those for open-field agriculture and are reported to range from $550–$1600/m² for greenhouses and $1600–$4300/m² for plant factories⁸⁸. Some authors note that capital costs are orders of magnitude higher for CEA than for open-field operations, and a single facility can cost hundreds of millions of dollars^4,89. Land costs are also higher in urban and peri-urban areas, which is where CEA is often proposed as a means of reducing food-transportation distances.

As a reflection of these costs, many CEA producers fail to attain profitability⁸⁹. Operating costs for energy constitute a larger share of expenses than most other industries (Table 2) – and were at the root of early vertical farming start-up bankruptcies^{5,9,15,20,67,90}. A late-1970s effort known as Phytofarms was abandoned by General Mills in 1982, purchased by another entity (for $20 M in today’s dollars), and operated until 1992, with utility costs named as a major factor in its ultimate bankruptcy⁴. Other well-capitalized companies (Google, General Electric, GTE, Sylvania, and Weyerhaeuser) also abandoned early efforts⁴. The current-day cannabis industry also faces solvency issues, tied in part to energy intensity³¹. Energy costs are ultimately borne by consumers in the form of elevated commodity prices, which can negate socioeconomic benefits of growing nearer to the point of demand, particularly for lower-income consumers^35,91,92.

Table 2 CEA facility energy expenses as a percentage of total operating costs

Contextualizing aggregate energy demand for indoor agriculture

It is helpful to put the energy intensity of CEA in context with other uses, and to similarly assess the aggregate energy demand. As an illustration, the median energy intensity (MJ/m²-y) for food crops grown in plant factories is 5-times that of the average US office building, while for the most intensive food crops the ratio is 90-times greater. Righini et al.²⁶ estimate of 16 MWh to provide a year’s worth of protein per person with soybeans grown in plant factories is on par with the requirements of an all-electric home with two electric vehicles. Moreover, while the energy embodied in producing 1 kg of usable protein from open-field soybeans ranges from 1.5 to 3% that of producing that same protein from beef, soybeans grown in plant factories require 4500- to 9000-times more energy than open-field farming and thus 60- to 120-times more than producing protein via beef (Table S3).

The literature review revealed no prior estimates of aggregate national energy use associated with CEA, aside from cannabis^31,93. Results of this meta-analysis enable the determination that current on-farm energy use for 15 primary CEA food crops (representing 7.4% of all yields of those crops) is 735 PJ/year, exceeding that to produce all crops under open-field conditions (Table S4)⁹⁴. Approximately half this energy use is attributable to cannabis cultivation. The total CEA food- crop production is <1% of all US food-crop supply (~7% of vegetables), by weight, and much less by calorie content. The cultivation area represented in this analysis is only about one-fifth the total protected area when including crops such as bedding plants, florist products, hemp, nursery plants, and sod (Table S2).

CEA energy demand is currently ~1% of the US total (all sectors). In a scoping estimate in which all applicable crops are produced in plant factories (Fig. 6), this would increase to up to 7% of all US energy consumption (10-times current energy use for all open-field crops) (Table S5), exceeding that of many other segments of the economy (Fig. 7). The hypothetical land area required to produce this CEA energy with solar photovoltaic arrays would be triple that currently used for outdoor production of these same crops.

Fig. 6: Scoping estimate of present and future US energy demand for controlled environment agriculture (CEA).

Aggregate national CEA energy use is estimated by applying average energy intensities from this meta-analysis to US current and projected CEA output plus net imports. Excludes crops known to be grown in CEA but for which sufficient energy data are unavailable: blueberries, chard, eggplant, string beans, hemp, floriculture, ornamental plants, and home cultivation or greenhouses for producing transplant starts for any crop. To derive the estimate for current CEA cultivation, crops are mapped to the dominant facility types in today’s market (plant factories for mushrooms and microgreens, other food crops to greenhouses, and cannabis to a 36% plant factory/27% greenhouse mix per Mills³¹. In each future-oriented scenario, crops are progressively mapped to higher-yield facilities at maximum penetration so as to define an upper limit of potential energy use. The sole identified estimate for national energy use for open-field food crops is 528 PJ/y⁹⁴. Estimated current-day total CEA energy use is 735 PJ/y, 0.9% of 2023 US final energy consumption. These results are expressed in terms of energy consumed at the site, enabling focus on facility performance and processes therein, independent of the thermal losses or fuel-mix of upstream power-generation. Scenarios exclude energy use for staple crops such as grains, legumes, and root crops, nuts, and most fruits, which are not suitable for CEA and thus continued to be grown in open-field conditions. Data and assumptions in Table S5.

Fig. 7: Energy use of controlled environment agriculture in context: US.

Adding energy for cultivation of tobacco and malting barley increases Beverages and Tobacco value by ~14%. Cryptocurrency mining is a subset of data centers. Open-field agriculture, current CEA, and maximum saturation CEA scenario from this study. Data and sources in Table S6.

Improving assessment practices: Nuances of energy analysis and benchmarking

The choice of energy metrics is critical to meaningful performance benchmarking^16,95. In the energy literature, CEA facilities are typically rated in terms of energy per unit of floor area^{44,62,67,68,87,96}, a metric useful for energy-infrastructure planning, assessing burdens on the grid, etc. The choice of net cultivation area or gross facility area (e.g., including aisles or office areas) can have a marked effect on computed intensities, but the distinction is often not made in the CEA literature. Metrics based on yield are standard in the agronomy and LCA literature – and in the agricultural industry itself – illuminating the energy-efficiency of the cultivation process while being insensitive to definition of planted area or cultivation periods. Occasionally, other functional units are used, e.g., sweetener metabolite for Stevia⁴⁵, kilocalories for mushrooms⁹⁷, protein for soybeans²⁶, or potency for cannabis⁴¹.

Defining a system boundary that articulates which energy-using processes are considered is essential to achieving clear and usable metrics. In practice, the literature is often imprecise in these definitions or otherwise quantitatively or conceptually ambiguous, e.g., regarding inclusion of energy inputs during production of seedlings or clones. Some studies omit specific energy types such as fuel for heating, end uses such as cooling, on-site electricity generation by diesel generators, or energy used on-site to produce CO₂, post-harvest processing, cold storage, etc. Others normatively include only single energy uses, e.g., heating^98,99, or lighting^27,100, or only measure total energy use while lighting is on^13,101,102, heating and cooling^103,104, during only part of the cultivation cycle¹⁰⁵, only seasonally¹⁰⁶, or during even briefer periods of monitoring thus missing the seasonal effects of fluctuations in weather and daylighting¹⁰⁷. Test facilities are sometimes located inside other buildings^47,108, with studies failing to measure the heating and cooling provided from the surrounding conditioned space.

The literature can also present incommensurate values with respect to the system boundary. For example, in some cases a portion of the injected CO₂ is made on-site and part off-site^53,109. Similarly, one study counted irrigation energy and substantial post-harvest weight loss due to commodity shrinkage during transport for open-field operations but not for CEA operations⁹¹. In an assessment of mushroom cultivation, the most energy-intensive process (composting) was conducted on-farm in one case and off-farm in another, resulting in a ten-fold difference in energy intensity between facilities that reflected energy accounting rather than actual energy efficiency¹¹⁰. Some assessments include energy input from human labor and sunlight (in the form of heat as well as illumination).

Other nuances arise from study duration and net yields. Most CEA crops can be grown multiple times each year. To extrapolate annual energy use from a single reported cycle one must know the cycle duration and any possible seasonal limitations or yield variability. The energy use embodied in pre- and post-harvest crop losses should also be added to the calculations, but rarely is. Cannabis is particularly impacted in this latter respect through a combination of losses due to disease (mold outbreaks are common), interdiction, and product destruction or recalls due to consumer safety testing failures that can collectively result in 20% losses nationally³¹.

Studies reporting only carbon footprint^19,111,112 have limited usefulness for energy analysis since the underlying energy intensity and types of constituent fuels and electricity are obscured, which is important when evaluating tradeoffs between fossil-fuel transport and electrified cultivation. Even studies providing both views show that emissions comparisons can obscure differences in underlying energy use⁹¹. This is particularly problematic when outcomes from countries with low-emissions electricity are compared to those with more carbon-intensive grids¹¹³. An emissions focus can mask caveats to conclusions and policy recommendations, as in a study set in Sweden which found lower CO₂ emissions for vertically-farmed lettuce and basil than for field-grown lettuce commonly imported from southern Europe¹⁰⁸. As Sweden’s grid is almost entirely based on zero-emissions nuclear and hydroelectric power, recalibration of those results to US conditions results in a 20-fold increase in total CEA emissions. Another Swedish study found domestic emissions below those of crop imports only in summer, excluding cooling energy¹¹⁴.

A model-based assessment of tomato production in Norwegian greenhouses further illuminates how assumptions can shape findings. Surprisingly, summertime cultivation in un-lit greenhouses was found to be more carbon-intensive than year-round cultivation. This arose from inappropriately comparing summer heating at low efficiency with fossil fuels and in some cases less facility insulation to heat pumps and lighting in winter at high efficiency powered with carbon-free hydroelectric electricity⁸². Wintertime heating is thus decarbonized and waste heat from electric light further substitutes for fuel. The comparisons also appear to involve variations in the mix of CO₂ produced on-site with fossil fuel vs. purchased CO₂ canisters for which the embodied energy and carbon is outside the system boundary and thus not counted. The evaluation of facility characteristics and seasonality are thus conflated with the influences of fuel, heating efficiencies, lighting, and perhaps varying system boundary conditions. With identical heating fuels and efficiencies, seasonal summer production would have the lowest carbon footprint per unit yield.

Improving assessment practices: Joint of measurement and modeling

Some CEA energy estimates are measurement-based while others are modeled. Measurement provides a “ground-truth” of sorts, albeit often with caveats, and helps identify potentially unmodeled uses (e.g., snow-melting on greenhouses). Modeling allows explicit articulation of system boundaries; annualized estimation under long-term average weather; and assessment of sensitivity to interior climate setpoints, building geometry, and outdoor climatic conditions while accounting for energetic trade-offs between factors such as lighting and cooling loads as a function of solar gains. Co-modeling of energy together with evapotranspiration, photoperiod, and other crop dynamics captures implications for energy-using processes such as dehumidification or lighting needs and how yields may vary as a function of energy inputs (Table 3).

Table 3 Uses of controlled environment agriculture energy model studies

Model studies also offer advantages insofar as they control for or at least make transparent the “noisy” drivers underlying measured data, and enable rapid and less costly parametric analyses and scenario studies^44,67 but are also susceptible to specification error and unrealistic assumptions. Energy-agronomy interactions are complex and challenging to model, while empirical data intrinsically captures the effects.

Comparing model results to measured data can help design, validate, and calibrate models and ensure realism^{35,44,67,68,115,116,117,118,119,120}. Measurement and modeling can be combined to achieve more comprehensive assessments^8,31,56. While estimating the benefits of energy efficiency measures is far easier with modeling – especially for ensembles of measures — some studies^44,46,121 have done so in actual facilities, providing added confidence in the findings.

Assumptions often favor CEA over open-field cultivation and transport energy use

The literature occasionally compares CEA outcomes to open-field practices (Fig. 3c, d), often citing values for different crops and geographies and comparing optimized and high-tech indoor methods to average open-field practices (rather than best practices). CEA yields for “protected” practices are reported to be 1.5 to 20-times greater than open-field cultivation, depending on crop¹⁰. More intensive open-field methods have been documented to increase yields by 2- to 3-fold^122,123 above standard practices, which would reduce the yield gap by half to two-thirds and correspondingly widen the differences in energy intensity in favor of open-field cultivation. It is rare that analyses offer sensitivities to baseline open-field energy use or emissions reductions as a function of energy efficiency or renewable energy.

Reduced transport energy is a prime argument given in favor of CEA practices. In practice, transport emissions can vary by nearly a factor of four, even among internal-combustion vehicles¹⁵, but truck shipping at low fuel economies is the typical default assumption. Actual outcomes are tempered by stipulated transport distances, mode (e.g., truck, train, ship, air), energy efficiency (energy per tonne-km), fuel choice (e.g., diesel, electric), logistics (return cargo), and other factors. In contrast to typical stipulations, one of the larger US CEA facilities routes its products through a Walmart distribution center hundreds of kilometers from the ultimate point of sale¹²⁴. Nicholson et al.⁹¹ isolate transportation and cultivation energy use for lettuce production in the US, concluding that the combined total in greenhouses and plant factories is 24 to 69% lower for open-field cultivation, even with pessimistic assumptions of cross-country transport distances 1400-times greater for field-grown than plant-factory crops, and inefficient diesel truck transport (3 km/l). Van Ginkel et al. ¹²⁵ find that cross-country US transport energy use does not exceed that of plant-factory energy use closer to the point of demand.

Discussion

On-farm CEA energy use is often the primary contributor to life-cycle environmental impacts of crop production, which, in aggregate, exceeds that for open-field cultivation in the US despite providing less than 1% of the nation’s harvest. In that country CEA today uses more energy than crypto-currency mining, beverage and tobacco products manufacturing, and many segments of the buildings sector (supermarkets, restaurants, hospitals, hotels, or warehouses) (Fig. 7). Per the US example, if the trend towards energy-intensive continues, CEA energy use could grow tenfold, exceeding the energy use of the commercial aviation or the chemical industry.

The most compelling applications of CEA are where import dependency and transport energy needs are high or the climate is ill-suited for outdoor crop production. These are often extreme or isolated locations, and can correspond to lower-income countries²⁹, examples of which are rare in the literature. CEA practices are sometimes particularly energy-intensive in these contexts, have the greatest dependency on fossil fuels, and can be the most problematic in terms of available technical support and operator skill^8,126, and have lower solar energy availability. In many cases, these areas are populated by lower-income populations, where energy availability and food security challenges are often greatest, capital for building and operating such facilities is scarce, and the resulting higher food prices could be particularly unaffordable.

This meta-analysis illuminates enormous variability in the patterns of CEA energy use and efficiency. While maximum yields often occur in windowless multi-level plant factories, greenhouses typically consume substantially less energy per unit yield. Energy intensities are lowest for simple “open” greenhouses (i.e., with minimal mechanical equipment and artificial lighting). Progressively increasing yields can lead to diminishing returns and other unintended environmental impacts. Farming in plant factories is also by far the most energy-intensive method on a building-footprint basis and thus the most difficult and costly to serve with solar power. Cannabis produced in plant factories is the most energy-intensive crop documented in the reviewed literature, requiring 300-times more energy than that to produce food crops in similar facilities.

Simple greenhouses thus emerge as the most energy-efficient CEA facility type for most contexts, consistent with the findings of smaller studies^20,127,128. Yet, greenhouses have thermally inefficient envelopes and heating, if any, is typically provided via inefficient fuel combustion. With optimized design and siting¹²⁹, the advantages would become even more pronounced, making greenhouses the most promising CEA facilities to become carbon-neutral since much smaller land areas are required to provide solar electricity than for more energy-intensive plant factories.

Open-field cultivation remains by far the least energy-intensive form of crop production. While CEA facilities tend to be highly optimized, no energy-focused studies have been identified that compare best practices of indoor to those of outdoor cultivation, which have benefitted from far less research and development aimed at reducing energy use. Longer-distance transportation systems can also achieve significant carbon reductions, shifting the calculus of benefits for CEA farm proximity to markets.

The potential for further improving energy efficiency of CEA processes is constrained. While the studies reviewed here span a period of three decades, half were published since 2020 and three-quarters since 2017. The results are thus largely representative of current-day technologies, practices, and market conditions. The model studies typically assume idealized conditions and the most advanced technologies or even practices that have not yet been demonstrated in practice. A sense of best practices can be gleaned by looking at the least energy-intensive values in Fig. 3, generally, and in the panels there which identify studies that included energy efficient technologies. Meanwhile, that the inclusion of energy efficient equipment is higher in the sample than in the market at large, indicates this baseline already includes significant added efficiency.

It is nonetheless worth asking whether the inclusion of older studies overstates current-day energy needs or obscures the potential for future energy savings. Surprisingly, energy intensities in this data set (1993–2024) show no downward trend over time, in aggregate and even for subgroups (greenhouses, plant factories) or for modeled versus measured assessments (Fig. S2). Evidence of learning curves is thus not readily evident. Indeed, the highest energy intensities tend to be found in the most recent literature. These observations suggest that a fundamental shift from current practices would be required in order to achieve substantially more efficient use at scale across the industry. The industry’s reluctance to invest capital to improve efficiency stands to slow progress in that direction.

Some authors present a grand vision for CEA, yet address the energy issue with only qualitative and normative statements that renewable energy and energy efficiency will suffice^130,131 or categorical statements that CEA has a lower carbon footprint than open-field cultivation⁷⁰, with the inference that the approaches are viable in their current form. Others cite literature that some categorize as aimed at fostering legitimacy¹⁴ and focusing uncritically on the industry’s “world-saving” potential^4,15,20.

This suggests that pursuit of CEA remains an expression of significant technological enthusiasm, if not irrational exuberance. Today, more than two millennia after the advent of greenhouses and more than a century after the introduction of electric lighting into CEA environments, significantly increased yields have indeed been achieved. However, due primarily to energy demands, essential staple crops cannot be cost-effectively produced with CEA, as such higher-calorie crops would require several orders-of-magnitude more energy inputs. Even for current practices, CEA business developers are experiencing increasing economic burdens (and insolvencies) associated with both capital and operating (primarily energy) costs.

To more effectively grapple with these challenges, and in keeping with the findings of other reviews^11,14,15,16, there is a need for more methodological nuance, transparency, and rigor in CEA energy assessments. One of the most widely cited studies is also among the most elementary in terms of simplified model approximations of energy use¹³². Methodological needs include clearer and more rigorously defined system boundaries, with more detailed energy accounting including disaggregation of energy types and uses, explicit identification of omitted energy uses, separate reporting of transport and heating fuels, and documentation of embedded thermal conversion factors. Analyses and cross-study comparisons are impeded by ambiguities in the treatment of energy-determinative factors such as definitions of cultivation areas (net, gross, etc.), inclusion of seedling stage energy inputs, uses of CO₂ for growth enhancement, numbers of cropping cycles per year, crop/weight losses, and post-harvest processing or storage. Energy-models require more attention to validation and should always employ coupled crop models in order to properly characterize interactions between environmental conditions and crop yields. Where economics are considered, care should be taken to avoid unrealistically low energy price assumptions³⁴. Complicating matters, producers may be reluctant to share what they perceive as proprietary data and even not-for-profit data aggregators have put the pooled information behind paywalls. Improved energy analyses would elevate the quality of LCA assessments, dispel entrenched and inaccurate rules of thumb, and inform efforts to quantify and manage national energy use and greenhouse-gas emissions associated with crop production.

Critical research needs include developing finer-grain benchmarks for comparisons among facilities together with more analysis of transportation and other off-farm energy uses, process change as a promising pathway to energy savings, energy and emissions modeling at regional and national scales, techno-economic assessments for lower-income countries, and energy optimization of outdoor cultivation. As author teams developing the existing literature tend to either have expertize in energy, agronomy, or LCA but rarely all three, more cross-disciplinary work would likely yield new insights.

Policymakers should champion tighter integration of agricultural planning and energy planning, e.g., fostering process-level reductions in energy intensity and siting CEA facilities with local electrical grid conditions and solar access in mind. Research and incentives should be focused on both open-field and CEA strategies. More comprehensive data on markets and industry structure are needed: for many countries, insufficient information is available to estimate current national energy use and emissions or make forecasts, thus hamstringing policymakers. In the US and no doubt elsewhere, this is particularly problematic in the case of cannabis, where research at the federal level is prohibited and limited at the local level where institutions dependent on federal support for other purposes are uneasy.

The agricultural industry faces an inherent contradiction between the allure of climate resilience promised by CEA advocates and the adverse effect that CEA’s energy use has on climate if based on fossil fuels, or on other environmental factors if based on land-intensive renewable energy (particularly in the case of vertical factory farms). A perhaps more pragmatic concern is the juxtaposition of CEA’s business goals and the pattern of failed CEA companies, the latter attributable in no small part to the high cost of energy using systems and the energy bills they generate. New efforts to enhance resilience and efficiency of open-field agriculture could yield better economic and societal outcomes.

Methods

To identify publications that provide sufficient data for conducting standardized energy analysis, a Scopus search was conducted with no date constraints. This included journal articles, conference proceedings, and book chapters using the string: Energy AND kWh OR MJ AND “plant factory” OR “vertical farm” OR “controlled environment agriculture” OR “indoor agriculture” OR “greenhouse” (Table S7). Other repositories were examined to identify technical reports and other document types not indexed by Scopus. References within the identified articles yielded further primary sources. Only English-language publications were evaluated.

The data was then extracted in original units and converted to common units to obtain uniform metrics (i.e., yield in kg/m²-y and on-farm energy intensities in MJ/kg yield and MJ/m²-y). Each case (model scenario or measured crop cycle) is a datapoint in the analysis. The data were binned by crop type, facility type (plant factory, greenhouse, open field), presence of standard versus energy efficient equipment and processes, and modeled versus measured analysis. Energy use was counted at the facility level: 3.6 MJ/kWh for electricity, 38.29 MJ/l for diesel fuel and heating oil, 33.6 MJ/l for gasoline, and 38.622 MJ/m³ for natural gas. Weights of the final marketed product (wet or dry) are used, per industry practices. Co-products (e.g., compost with mushrooms or straw with wheat) were excluded⁹⁷.

Results are expressed in terms of energy consumed at the site, enabling a focus on facility performance and processes therein, independent of the thermal losses or fuel-mix of upstream electricity generation. This is also necessitated because studies frequently fail to separate electricity from other energy types. Where not specified, a ratio of growing area to total footprint of 0.7 is assumed and, for plant factories, six “vertical” cultivation layers per storey for non-vining crops, and yields annualized based on days per crop cycle and other information provided in the source document, with one non-cultivation day between cycles. The energy embodied in human labor is occasionally reported, but is not included here. Nor is energy embodied in non-energy inputs (e.g., fertilizers, growing media, or building materials) or that required for off-farm transport.

Estimates of the required land area for solar photovoltaics providing 100% of required on-site energy are based on system output calculated with the National Renewable Energy Laboratory’s PVWatts calculator (v8.4.1 - default system characteristics) for the latitude and longitude nearest to the facility (assuming no shade obstructions). Power densities are based on the US industry-average value of 0.67 MW_DC/ha (1.01 GWh/ha-y), the stock-weighted average of fixed-tilt and tracking panels installed nationally as of 2019¹³³. In the case of grid-based solar, these results exclude roads and other land uses surrounding the arrays (except that for inverter pads and other ancillary electrical equipment), which can be substantial. Panel spacing is represented by a ground-cover ratio of 0.55 (active area/total occupied area). The same factor is applied for plant factory rooftops, although surveyed small, medium, and large commercial buildings typically devote 26%, 49%, and 66% of their roof area to solar, respectively¹³⁴. Any additional required panels beyond what cultivation-facility rooftops can accommodate are assumed to be ground-mounted.