AI in greenhouse horticulture and agri-food: where to start with yield forecasting and climate control

Why yield forecasting is the logical starting point

Harvest forecasting is not a new discipline: for decades, growers have estimated how many kilos will come off the crop a week ahead, based on experience, crop observations and historical patterns. AI does not fundamentally change that process, but adds a second estimate that is based on measurable patterns in climate and growth data rather than on experience alone.

That is valuable because buyers, packaging lines and workforce planning all call for a reliable look ahead. A model working alongside the experienced grower can flag deviations that are not yet visible to the naked eye, so that the grower can adjust earlier rather than correct afterwards.

Which data do you already have, and what is missing

Most businesses with a modern climate computer already have the basics: temperature, relative humidity (RH), CO2 level and light sum, often logged hourly or even every minute. That is a good starting point for a usable yield forecasting model, but in practice the link with harvest records in the same system or on the same timeline is often missing.

• Climate data from the climate computer (temperature, RH, CO2, light, irrigation)
• Historical harvest records, preferably per bay or compartment and per day
• Crop observations that are now on paper or in a separate system
• Labour hours around the harvest, for linking the forecast to planning

Climate control: AI as a second pair of eyes, not as an autopilot

Climate computers already control the climate automatically on the basis of preset rules and setpoints. AI can support that control by recognising patterns in the data that point to inefficient energy use or suboptimal growing conditions, and feeding that back to the grower as advice.

It is important to be clear here about what AI does and does not do: the grower sets the cultivation strategy and decides on adjustments; AI supplies signals and scenarios to underpin those decisions. A system that changes setpoints independently without human approval does not fit this approach and also calls for a careful risk assessment.

Harvest and workforce planning as the next step

As soon as a yield forecast is reasonably reliable, a logical follow-on application arises: linking the expected harvest to the planning of harvest staff and processing capacity. This is often where the time savings become most tangible, because planning is now often done on the basis of experience and telephone calls.

Start small, for example with one crop or one compartment, and only expand once the forecast demonstrably gives you something to rely on in practice when scheduling people and resources.

Quality, certification and responsibility

For businesses that work under quality frameworks such as GlobalG.A.P., the grower remains ultimately responsible for cultivation decisions and records, even when AI advice is used. AI models do not replace certification requirements or responsibility for cultivation, but they can help to flag deviations relevant to quality assurance at an early stage.

In addition, the AI literacy obligation in Article 4 of the EU AI Act has applied since 2 February 2025: anyone who deploys AI systems or has them deployed has been required since 27 July 2026 to take measures that support the AI literacy of the employees involved, without having to guarantee a specific level. For a greenhouse horticulture business, this means in practice that the person who reads the yield forecast advice and acts on it understands what the model is based on and where its limits lie. Which AI in a greenhouse, on the sorting line and in workforce planning falls under which regime is set out system by system on our page on the EU AI Act in greenhouse horticulture and agri-food (in Dutch).

Where to start in practice

Do not start with a grand data platform, but with the question of which decision you already take every week on the basis of an estimate, and whether a second, data-driven estimate can be placed alongside it. Yield forecasting based on existing climate and harvest data is the most feasible first step for most businesses, with climate control advice and the link to planning as logical next steps once the first application has proved its value. The free Practice Scan helps you choose that first step.

What that first step looks like in practice at a business, from the scan to a working workflow with a record of who reviews and who signs, is set out on our page on AI implementation in greenhouse horticulture (in Dutch).