New results demonstrate greenhouses can automate important tasks using artificial intelligence (AI). The agriculture technology company, Motorleaf, has new findings and a testimonial demonstrating the industry-changing performance of their greenhouse automation services. "Of particular significance are novel findings that prove our technology is able to not only learn greenhouse tasks but indeed becomes increasingly precise once implemented within farming operations", they explain. 

Motorleaf’s initial service resides in software that provides automated, highly accurate harvest yield forecasts weeks in advance for cultivators of tomatoes and peppers. In addition to enabling greenhouses to cut costs and gain greater control over their daily operations, their solution requires no significant changes to greenhouse infrastructure and adds value to pre-existing greenhouse control systems.

"Other greenhouse technologies lose value over time from wear-and-tear. What's remarkable about AI is that it gets better and better over time with more greenhouse data, providing more and more value to greenhouses. That's a world first," said the CEO of Motorleaf, Alastair Monk.

Increase in precision after one year
Once implemented at SunSelect Produce--a 70-acre greenhouse in California--the AI technology reduced errors in estimating weekly tomato yields by 50% at the start; following one year, the greenhouse now benefits from a 72% reduction in error. Such precision enabled the company to fully automate harvest forecasting. In a new video testimonial, co-owner of SunSelect Produce, Victor Krahn, states this will save each member of his grow team a day-and-half of work each week, being gained time they now use to improve their product line and expand business opportunities.

“In the marketplace right now, as far as technology-based products, there’s nothing I’ve seen that is available to even touch this type of large-scale greenhouse production", he explains. "We’re truly believers in the AI model … way more accurate results than our human scouting team.”

Error in manual estimates of tomato harvest yields at SunSelect Produce fluctuate between 17% to 33%. From continuous learning with greenhouse data, Motorleaf produced increasingly precise versions of their AI algorithm for automated harvest forecasts, where the error in harvest estimates now average less than 8%.

Data acquisition
Motorleaf’s current phase of data acquisition from SunSelect now aims to strengthen the automation technology so it can surpass highly accurate 2-week harvest yield predictions. Sights are set on providing accurate predictions more than a month into the future.

SunSelect acquired an accurate forecast for tomato harvest 2 weeks into the future 

Greenhouse data across the globe
"Developing the automation service starts by collecting data on internal growing conditions within commercial hydroponic greenhouses that many have available on hand. We use that information to develop AI algorithms custom made for each greenhouse and plant variety", the Motorleaf team explains. The company continues to replicate their observations with data from greenhouses in Europe and North America, with additional results from Asia expected by the end of October.

The data from a European producer, reducing the error in harvest forecasts by 35%. mape= average error in predicting harvest yield; harvest yield measures on y-axis removed to protect confidential crop production outputs.

And this one is for the cherry tomato yield forecasts. mape = average error in forecasting.The AI algorithms reduce error in harvest forecasts for a large-scale cherry tomato greenhouse from 29% to less than 9%. The name of the greenhouse is withheld to protect confidential business information. 

Crop pest and disease scouting
In addition to expanding harvest yield estimates to new crop varieties, Motorleaf is now beta testing their AI technology to automate crop pest and disease scouting. "Initial results are promising in predicting the onset of clavibacter and whitefly. We now seek additional greenhouses to test automated scouting for a growing list of common pests and diseases." 

For more information:
Jason Behrmann