How model predictive control just made agrivoltaics — solar panels sharing farmland with food — nearly 90% more productive than either alone.
Here's a land-use riddle that's about to define the next decade of energy development: a parcel of land can grow food, or it can generate solar power. Choose one, and you're wasting the other. That's the tension at the heart of every environmental assessment for utility-scale solar — and it's why "agrivoltaics" (raising panels above crops) keeps coming up as the compromise that lets you have both.
But agrivoltaics has always had an awkward problem: panels want to face the sun; crops want the sun too. A panel that tracks the sun all day casts a deep shadow that starves the plants below. A panel that stays out of the way produces far less power. Somebody has to negotiate.
Until now, the negotiation was fixed at design time — you picked a tilt, a row spacing, and lived with it. A new paper from researchers at the University of Michigan and Michigan Tech throws that out the window. Their system renegotiates the deal every hour, in real time, using a control strategy borrowed from robotics and process engineering.
The Idea: A Controller That Weighs Power vs. Produce
The paper, led by Anna Stuhlmacher and Johanna Mathieu, applies Model Predictive Control (MPC) to dual-axis agrivoltaic panels.
MPC is the control philosophy behind self-driving cars, chemical plants, and grid balancing. Here's how it works in this context:
- Every hour, the controller looks at the weather forecast — irradiance, temperature, cloud cover.
- It simulates the entire remaining growing season: where the sun will be, how much power the panels would make, how much shade the crops would get, and how that shade converts into lettuce growth.
- It finds the panel positions that maximize the weighted score: power revenue + crop yield (a knob, ω, sets how much you care about each).
- It moves the panels to the first hour's optimal position, then re-solves everything next hour with fresh forecasts.
The panels are never perfectly facing the sun, and they're never perfectly out of the way — they're doing something smarter: giving the crops exactly enough light to grow well, and keeping the rest for electricity.
To make this tractable, the team did some genuinely clever math. The full problem is non-convex (the geometry of shadows is nasty), so they reformulated it as a convex Second-Order Cone Program — using a trick where tilt angles become (cos θ, sin θ) pairs with a relaxed unit-circle constraint, and the shading factor is approximated by an affine function of cos(tilt deviation). The approximation errors turn out to be negligible: for reasonable weight settings, the final energy and yield numbers are off by only ~0.01%.
The Headline Number: Land Equivalent Ratio of 1.897
The metric that matters is the Land Equivalent Ratio (LER) — how much land you'd need to get the same food and energy separately, compared to doing it together on one parcel.
- LER of 1.0 = no benefit (same as splitting the land).
- LER of 1.897 = you're getting 89.7% more total value from the land than using it for just one thing.
In the perfect-forecast case, the system achieves 0.899 crop LER and 0.998 PV LER — meaning the crops keep ~90% of their full yield, while the panels give up almost nothing (~0.2% power loss) versus pure sun-tracking. That's the negotiation working: the panels sacrifice a hair of output, and the farm keeps nearly everything.
Even better, the result is robust to forecast error. With deliberately noisy irradiance and temperature forecasts (5–15% noise), the LER only drops from 1.897 to ~1.851. The MPC keeps re-planning, and the damage from bad forecasts is minimal. This matters enormously in the real world, where weather forecasts are never perfect.
The case study: a 280-panel, 60-day lettuce-growing season in Ann Arbor, Michigan, using real NSRDB solar data and the EPIC crop model. Lettuce is a smart choice — it's shade-tolerant, which is why it's the standard test crop in agrivoltaics research.
Why This Matters for Environmental Due Diligence
If you're evaluating a utility-scale solar project, agrivoltaics is no longer a feel-good footnote — it's becoming an engineering solution with numbers attached:
- It changes the land-use math. An LER above 1.5 means solar developers can make a credible case that their project increases agricultural output per acre, which is a dramatically different environmental impact story than "we paved over farmland."
- It de-risks permitting. In jurisdictions where agricultural land is protected, "we can keep farming under the panels" is often the difference between approval and rejection. Now there's control-theoretic rigor behind that claim, not just anecdote.
- It reframes the trade-off as tunable. The ω knob means a project can be operated as "power-first" or "crop-first" — or re-tuned seasonally based on crop prices and electricity prices. That's operational flexibility due diligence teams should know about.
- Dual-use land = dual revenue. For lenders, a farm that produces both crop income and energy revenue has a more resilient cash flow profile.
The Bottom Line
This paper is the first to treat agrivoltaic panels as a real-time controlled system rather than a static design choice — and the numbers justify the approach. Nearly 90% land-productivity gains, robust performance under forecast uncertainty, and a mathematically sound path to scale (the SOCP reformulation solves fast enough for hourly re-planning over a full growing season).
Solar and farming aren't fighting over land anymore. They're negotiating — and thanks to MPC, the panels now have a very good negotiator.









