Put a roof over a crop and the crop gets less light. That much is arithmetic.
What most people then assume is that the yield falls by the same fraction.
It does not, and the gap between those two numbers is the entire case for building solar over farmland.
A leaf is not a light sensor.
It is a chemical process with a ceiling, and on a bright summer afternoon it hits that ceiling early.
So what happens to the photons it cannot use?
A leaf stops paying attention long before noon
Photosynthesis rises steeply with light when light is scarce, then flattens.
Past a saturation point the machinery is running as fast as it can, and every extra photon arriving does almost nothing useful. It is absorbed, converted to heat, and has to be dealt with.
Full summer sun sits far above that point for a crop like this.
Which means taking away a large share of the light removes mostly the part the plant was already discarding.
The second half of the effect is temperature. A cooler leaf loses less water, and a plant losing less water keeps its pores open instead of closing them to survive the afternoon.
A closed pore takes in no carbon dioxide at all, so an hour of midday shutdown costs more than an hour of dimmer light.
Shade buys back that hour, and the plant keeps working through it.
The farm this was built around
The layout is the ordinary one: panel rows raised above ground, crop rows planted underneath and between, everything sharing the same acre.
The crop chosen was tomatoes, which are grown in the open across the mid Atlantic and are demanding about heat.
The site was placed in central New Jersey on a hot humid summer day, and that choice was deliberate.
It is a corridor where the demand for food and the demand for electricity sit on the same land, with very little of it spare.
Every acre committed to one use in that region is an acre argued over.
Which makes the interesting question not whether panels hurt a crop, but how much they hurt it.
What the model produced and what it was checked against
The work was published in 2026 in a modeling journal, and the method is a physical simulation rather than a field trial.
It resolves the exchange of energy, momentum and mass between the panels, the canopy, the soil and the air above them, which is what lets it produce a leaf temperature and a panel temperature from the same run.
Before being pointed at New Jersey it was checked against two real datasets: leaf temperatures recorded in California and soil temperatures from a Minnesota site.
The outputs are specific. Leaves under the array ran about 3.3 degrees cooler across daylight hours and up to 13.6 degrees cooler through the worst of the afternoon.
The panels themselves ran roughly 10 degrees cooler, which recovered close to 15 percent of the output that heat takes off a silicon module.
Water use dropped 22.4 percent, and the perceived temperature for anyone working under the array fell by about 8 degrees.
What a simulation is not
None of this is a harvest.
The model was validated on leaf and soil temperature, which is not the same as being validated on yield, and carbon uptake is not fruit in a crate.
It is also one crop, one location, one weather pattern and one panel geometry. Change the row spacing, the panel height or the latitude and the light budget underneath changes with it.
The 15 percent efficiency recovery is a model output too, not a measurement, and it will move with wind speed and humidity at any real site.
What the field evidence does support is the cooling half of the story, which is consistent across the sites where panels already share ground with grazing animals, and with what is known about temperature and cell output.
The yield half still needs a season.
The number nobody put in the headline
Two of these results are about equipment and plants. The third is about people.
An 8 degree drop in perceived temperature during working hours is not a marginal comfort gain in an agricultural field in August.
It is the difference between a workable afternoon and a lost one, in an industry where heat already sets the hours, as the research summary points out.
That finding needs no yield data to be useful, and it applies to arrays that are already standing, as the news coverage noted.
The tomato leaves may or may not deliver a crop worth selling.
The shade underneath them is already worth something.
