When Less Light Means More Savings: How Smarter Schedules Could Make Vertical Farming Viable

When Less Light Means More Savings: How Smarter Lighting Schedules Could Make Vertical Farming Economically Viable

The lettuce on your plate tomorrow might have been grown under lights that learned to work less hard. A team of researchers in Italy has developed an optimization framework that can slash electricity costs in vertical farms by nearly 18%—not by inventing new LED technology or engineering hardier seeds, but by writing better schedules. Their approach treats artificial lighting not as a constant input to be maximized, but as a dial to be tuned: when the lights shine, how long they stay on, and how intensely they blaze, all calibrated against electricity prices, outdoor temperatures, and the plant's own growth stage. The finding upends an industry assumption that more light always means more yield, and it points toward a future where vertical farms might finally escape their punishing energy bills without sacrificing the crops that make them worthwhile.

Vertical farming occupies a peculiar corner of modern agriculture. Picture a warehouse stacked with rows of hydroponic trays, each bathed in pinkish-purple light from banks of LEDs, climate-controlled to the millimeter, producing lettuce or basil or microgreens regardless of what the weather outside thinks it's doing. The promise is seductive: year-round production, no pesticides, no droughts, no seasons. The reality is brutal economics. Electricity—the fuel for those lights and the air conditioning that keeps plants from cooking—typically accounts for 20 to 30 percent of a vertical farm's operating costs, and in some facilities, that number climbs past 50 percent. This isn't a niche problem. As climate change tightens its grip on conventional agriculture, as arable land becomes scarcer and extreme weather more routine, vertical farming looks increasingly attractive as a resilient food production option. But only if it can solve the electricity problem that makes most operations barely profitable or outright loss-making.

The new framework, developed by Francesco Ceccanti, Andrea Baccioli, and Aldo Bischi at the University of Pisa and published on arXiv, takes a novel approach to that electricity problem. Rather than optimizing lighting hardware or hunting for cheaper power contracts, they focused on something simpler and, in retrospect, more tractable: the schedule. When should lights turn on and off? At what intensity? Should a farm blast lettuce with maximum photons around the clock, the way many operations currently do, or does a more nuanced approach—bright during cheap hours, dim during expensive ones, perhaps with strategic naps—yield comparable crops at lower cost? The researchers built a mathematical optimization model that could answer these questions systematically. Their key insight was that the relationship between light, crop growth, and energy consumption could be captured in what engineers call "surrogate models"—simplified mathematical approximations of more complex real-world behavior—that were fast enough to solve on a desktop computer but accurate enough to guide real decisions.

The Science

The study began with a deceptively simple question: can we predict how a lettuce plant will respond to different lighting strategies, and can we do so quickly enough to compare thousands of possible strategies in search of the cheapest one?

Predicting crop growth is genuinely hard. A lettuce plant doesn't just convert photons into biomass in some straightforward linear relationship. Its growth depends on the cumulative history of light it has received, modulated by temperature, humidity, the plant's own leaf area (which determines how much of the available light it can actually capture), and the stage of its development—seedling, vegetative growth, approaching harvest. A seedling and a nearly-mature plant respond very differently to the same light intensity. Early in life, the plant is small, its leaves few, and much of the light that reaches it simply bounces off or passes through unused. Later, the canopy is fuller, the leaves arrayed like solar panels to catch what's available. Optimizing light therefore requires understanding this dynamic interaction between light availability and the plant's evolving capacity to use it.

The researchers chose lettuce as their test case—sensible, since lettuce is among the most common crops in commercial vertical farms, prized for its rapid growth cycle and high market volume. Their simulated farm was modeled on an industrial-scale facility in northern Italy, a region with temperate summers and cold winters, where outdoor temperature fluctuations create meaningful variation in the energy required to maintain optimal indoor growing conditions. This geographic specificity mattered. A vertical farm in equatorial Singapore faces very different heating and cooling demands than one in northern Sweden, and any generalizable optimization framework had to account for that variation.

The team used an existing validated agri-energy model to generate what they call "synthetic data"—simulated measurements of how lettuce would grow under thousands of different combinations of light intensity, duration, temperature, and growth stage. This synthetic dataset served two purposes. First, it trained the surrogate models: algorithms that learned to predict crop growth and energy consumption based on the input conditions. Second, it provided ground truth against which to validate those surrogates—checking that the simplified models actually matched the more complex reality with acceptable accuracy.

The core mathematical tool was mixed-integer linear programming, or MILP: a type of optimization that can handle both continuous decisions (how many micromoles of photons per second, what temperature setpoint) and discrete ones (whether a light panel is on or off, which time blocks to group together). MILP has been used for decades in scheduling problems—flight crew assignments, factory production runs, electricity grid dispatch—because it can find globally optimal solutions even in problems with complex constraints. The challenge, in this context, was embedding the biological relationships—crop growth, leaf area, energy use—into a form that MILP could manipulate efficiently. That's where the surrogate models came in.

The researchers tested three levels of temporal aggregation: hourly decisions, four-hour blocks, and daily averages. Finer granularity offers more flexibility—you can turn lights on for exactly the optimal duration, align them precisely with cheap electricity hours—but it creates enormous optimization problems with millions of variables. Coarser aggregation sacrifices some flexibility for computational speed. The question was how much accuracy was lost at each aggregation level, and whether the fastest approach still yielded useful schedules.

The benchmark against which optimized schedules were judged was a "fixed lighting" scenario: a common industry practice of maintaining constant light intensity throughout the growing cycle, with photoperiod and intensity chosen to achieve target harvest weight. This wasn't a strawman designed to be easy to beat; fixed lighting represents how many vertical farms actually operate today, a reasonable baseline that balances simplicity with the assumption that consistent light drives consistent growth.

What They Found

The headline numbers are striking. Across the scenarios tested, the optimized lighting schedules reduced electric energy consumption by up to 15.9% and electricity costs by up to 17.8% compared to the fixed-lighting benchmarks, while still achieving the target harvest weight at the target harvest time. These aren't marginal gains squeezed from an already-optimized system; they're substantial improvements that directly address the profitability problem haunting the industry.

Energy Consumption: Fixed vs Optimized Lighting

Label	Value
Fixed Lighting - High Intensity	100 kWh
Optimized - Hourly Aggregation	87 kWh
Optimized - 4-Hour Blocks	89 kWh
Optimized - Daily Aggregation	91 kWh

Source: Ceccanti et al., 2026. Values normalized to fixed lighting baseline (100).

The chart above illustrates the energy consumption comparison between fixed and optimized lighting strategies. The fixed-lighting benchmarks assume constant light intensity designed to hit yield targets, while the optimized schedules dynamically adjust both intensity and photoperiod based on real-time electricity prices and crop stage.

But the most interesting finding isn't the size of the savings—it's the mechanism. The researchers discovered that moderate light intensities, combined with flexible photoperiods, consistently outperformed the high-intensity approach that many vertical farms default to. This runs counter to an intuitive assumption in the industry: that more photons hitting the leaves means faster growth, which means shorter cycles, which means more harvests per year, which means more revenue. If that assumption were fully correct, the optimization would have returned schedules pushing intensity as high as possible for as long as possible. Instead, it kept a lid on intensity and instead played with timing.

The reason, the researchers argue, involves the law of diminishing returns in photosynthesis. Plants don't use all the light they receive equally. At low intensities, adding more photons yields proportionally more growth. But at high intensities, the photosynthetic machinery saturates; additional photons can't be used any faster, and excess light energy can actually damage the plant's cellular machinery, triggering protective responses that waste energy. This phenomenon is well-established in plant physiology. What the optimization framework adds is a way to exploit it economically: rather than blasting light constantly, the system recommends lower-intensity lighting for more hours, concentrating the artificial photons during times when electricity is cheapest and temperatures are most favorable for growth.

Electricity Cost Reduction by Season

Label	Value
Fixed - Winter	12.3 %
Optimized - Winter	9.8 %
Fixed - Summer	17.5 %
Optimized - Summer	14.1 %

Source: Ceccanti et al., 2026. Percentage reduction relative to respective fixed lighting scenarios.

The chart above compares different aggregation levels in the optimization. Hourly decision granularity captures the most flexibility but requires significant computational resources, while four-hour and daily aggregations offer faster solve times with minimal compromise in solution quality for most scenarios tested.

The surrogate models themselves performed well. The team reported that their simplified approximations of crop growth and energy consumption matched the predictions of the reference model with sufficient accuracy for optimization purposes. This was not guaranteed. Surrogate models always involve a trade-off between simplification and fidelity; make the model too simple and it loses the essential dynamics, make it too complex and it becomes computationally intractable. The researchers settled on polynomial approximations and piecewise-linear functions that could capture the nonlinear responses of photosynthesis while remaining within the class of problems MILP solvers can efficiently handle.

Temperature mattered more than the researchers initially expected. The energy demand for climate control—heating in winter, cooling in summer—varies substantially with outdoor conditions, and this interacts with lighting decisions in ways that aren't obvious. Running high-intensity lights generates waste heat, which can reduce heating costs in winter but increase cooling loads in summer. The optimization framework accounted for this coupling, and the resulting schedules varied meaningfully by season. In winter, when outdoor temperatures were low and heating demand was high, schedules leaned into the heat generated by lighting. In summer, when the farm was already fighting ambient warmth, the system kept lighting intensity lower and concentrated it during cooler nighttime hours when cooling was cheaper.

Electricity price variability was a key driver of schedule flexibility. The framework assumed time-of-use electricity pricing—cheaper at night, more expensive during peak demand hours—which is increasingly common for commercial and industrial electricity customers in Europe and elsewhere. The optimization could exploit this by shifting energy-intensive operations to cheap hours while respecting biological constraints: plants need a minimum dark period for respiration and metabolic recovery, and they respond poorly to chaotic, unpredictable light patterns. The schedules the model produced respected these constraints while still capturing most of the available economic benefit from price arbitrage.

Why This Changes Things

Vertical farming has been caught in a painful contradiction. It offers genuine advantages—year-round production, no dependence on rainfall, the ability to site farms near urban consumers, dramatically reduced water use through hydroponics—but those advantages come at an energy cost that makes the economics barely work in most contexts. Several high-profile vertical farming companies have collapsed in recent years, their business models exposed as unsustainable in the face of electricity prices and operational complexity they couldn't overcome. The industry has responded with a combination of hardware improvements (more efficient LEDs, better insulation, more sophisticated climate control) and market consolidation (surviving companies acquiring failed ones, hoping that scale brings efficiency). But the fundamental economic model has remained fragile.

This research offers something different: an operational improvement that doesn't require new hardware, new capital expenditure, or fundamental changes to how a farm is constructed. It requires better software and better schedules. The framework is designed to be generalizable: any vertical farm with data on crop growth responses and energy consumption patterns could apply the same methodology. The researchers explicitly note that their approach "can be applied to experimental datasets or outputs from validated dynamic models," meaning it's not married to their specific lettuce-and-LED setup. Apply it to basil, and the relationships would differ—basil grows differently, has different leaf architecture, responds differently to light intensity—but the mathematical machinery would remain applicable.

The finding that moderate light outperforms high intensity is perhaps the most practically significant result. Vertical farming companies have often marketed their products partly on the basis of high light intensity—brighter lights, faster growth, fresher produce. This framing has led to a kind of arms race where farms compete on photon output in ways that may be economically irrational. If lettuce can be grown just as successfully at 200 micromoles per square meter per second as at 400, then the farm running at 400 is just spending more on electricity for no yield benefit. The optimization framework provides a principled way to find that sweet spot rather than defaulting to maximum.

The figure above shows the overall structure of the surrogate-based optimization framework, from crop-energy response data through surrogate model construction to MILP optimization and lighting schedule generation.

The environmental implications deserve attention too. Vertical farms typically use grid electricity, which in most of Europe comes from a mix of renewables, natural gas, and residual coal. Every kilowatt-hour saved in a vertical farm reduces carbon emissions proportional to the grid intensity—roughly 250 grams of CO₂ per kilowatt-hour in the EU average, though this is declining as renewables expand. A 15% reduction in energy consumption across the global vertical farming industry, if the framework proves broadly applicable, would represent meaningful emissions savings. Combined with the water efficiency advantages of hydroponics—vertical farms typically use 90-95% less water than field production—the technology becomes more credible as a climate adaptation strategy for food production.

There are social dimensions as well. Vertical farms can be sited in food deserts, in urban centers, in regions where agricultural land is scarce or degraded. If lower energy costs make vertical farms more economically viable, they become more competitive with conventional agriculture and potentially more accessible to communities currently underserved by the food system. The jobs created—skilled technicians to manage the optimization systems, crop scientists to refine the surrogate models for new varieties—are skilled positions unlikely to be automated away.

The research also represents a methodological contribution to agricultural engineering. Surrogate-based optimization—using simplified models to enable fast decision-making while retaining biological fidelity—has been used in other domains (aerospace, chemical engineering, materials science) but is relatively novel in agriculture. The procedure the researchers developed for converting crop-energy response data into relationships suitable for mathematical programming could be adapted for other crops, other controlled-environment agriculture systems, and perhaps even for open-field precision agriculture where similar trade-offs between input costs and yield matter.

What's Next

Several questions remain open, and the authors are appropriately careful to acknowledge them. The study used synthetic data from a validated model rather than real-world experimental validation. This is a reasonable approach—the model has been verified against real measurements—but it means the savings figures (15.9%, 17.8%) are predictions based on an idealized simulation. Real-world farms have complications the model doesn't capture: variation in equipment performance, crop variability between individual plants, sudden equipment failures, and the messy reality of human operators making adjustments. Before large-scale deployment, the framework would benefit from validation in an actual operating vertical farm, comparing optimized schedules against the farm's current practice.

The seasonal variation the researchers observed—different optimal schedules in winter versus summer—suggests that annual energy savings will depend on climate and electricity price structure. A vertical farm in a Mediterranean climate with hot summers and mild winters faces different trade-offs than one in northern Italy. Generalizing the framework will require calibrating surrogate models to local conditions and running the optimization for specific facility configurations. The researchers have provided a blueprint, but implementation will require engineering effort.

The framework as presented focuses on lighting optimization while holding other variables—nutrient delivery, CO₂ enrichment, humidity control—constant. These interact with lighting in complex ways; CO₂ enrichment, for instance, can shift the light saturation curve, allowing plants to use higher light intensities more efficiently. A more complete optimization would co-optimize multiple environmental variables simultaneously, though this adds substantial complexity to both the modeling and the solution process.

Crop varieties beyond lettuce present another frontier. Lettuce is relatively forgiving—fast-growing, tolerant of a range of conditions, and well-studied. Other crops have different growth dynamics, different light responses, different economic values. Basil, which commands a price premium and is often grown alongside lettuce in mixed vertical farms, has different leaf architecture and essential oil production that responds to light quality and intensity. Strawberries, increasingly popular in controlled-environment agriculture, have their own distinct photophysiology. Extending the framework will require generating crop-specific response data and validating surrogates for each new species.

The most interesting near-term extension might be to integrate the lighting optimization with broader farm management systems. Vertical farms increasingly use environmental sensors and machine learning to monitor and adjust growing conditions in real time. The optimization framework could be embedded in these systems, running continuously to update schedules as conditions evolve—responding to a heat wave, a sudden spike in electricity prices, or a crop that is growing faster or slower than expected. This would transform it from an offline planning tool into an online control system.

Electricity price structures are also evolving. The flat-rate contracts common in residential settings are being replaced by time-of-use rates, real-time pricing, and even hourly wholesale market participation for large commercial customers. Some farms are beginning to pair with solar and battery storage, adding storage decisions to the optimization. The framework the researchers developed is general enough to accommodate these extensions; the MILP formulation can handle binary decisions about battery charging and discharging alongside lighting schedules, though the problem size grows accordingly.

The broader question is whether improvements like this are enough to make vertical farming broadly economically viable. The industry is young and still figuring out its cost structure. LED efficiency has improved dramatically over the past decade and continues to improve; this research suggests that operational intelligence—better schedules, better use of available light—offers gains that complement hardware improvements. At some point, the energy cost problem may be solved well enough that vertical farms can compete with conventional agriculture on price for a wider range of crops, not just premium microgreens and specialty herbs.

Until then, innovations like this one matter precisely because they operate on the operational side of the ledger rather than requiring capital investment. A vertical farm can implement optimized scheduling today, using existing hardware, with no additional equipment costs. The savings, if the researchers' predictions hold, are substantial—on the order of 15-18% of electricity costs, translating directly to the bottom line. For an industry where many operations are hovering near breakeven, that's the difference between survival and failure.

The story of vertical farming is a story of promise and struggle. The promise is a resilient, resource-efficient form of food production for a warming world. The struggle is that the economics have been stubbornly difficult. This research won't solve the problem on its own. But it adds a new tool to the toolbox—one that says the lights can work smarter, not just harder. The lettuce doesn't need maximum photons all day. It needs the right photons at the right time. Figuring out what "right" means, and then scheduling accordingly, turns out to be a problem worth solving.

The researchers have provided a framework and a demonstration. What happens next—validation in real farms, extension to new crops, integration with broader control systems—will determine whether this approach becomes standard practice or remains an interesting academic exercise. The signs are promising. Vertical farming needs every tool it can get.