Western Farm Press Logo

Mason Earles’ team is training AI machine-learning models to find a data set on the ground.

Lee Allen, Contributing Writer

June 24, 2021

3 Min Read
GL0624-UC-big-data.jpg
The current prototype of a sensing kit does a run through a Lodi, Calif., vineyard block to collect crop data.Mason Earles/UC Davis

Adapt or die.

Adaptation is a key requirement for survival.  George Bernard Shaw noted that, “Reasonable people adapt themselves to the world,” while Charles Darwin went even more basic — “It’s not the strongest of the species that survives, nor the most intelligent — it’s the one most adaptable to change.”

So scientist Mason Earles looks to change things in his research at the University of California, Davis by bringing algorithms and agriculture together combining big data and vineyards.

“We’re on our 2021 iteration of a sensing kit that will strap onto tractors, just like a sprayer, and tell us more about the grapes hidden in the canopy,” Earles said.  “Our Alpha version last year covered about 200 acres and returned harvest data.  This year’s upgraded version will roll out early in an attempt to predict yield.  I don’t think anyone’s been successful doing this with grapes because there’s so much occlusion.

“We’re taking the big data approach, training AI machine-learning models to find a data set on the ground.  Up to this point, this hasn’t been done in terms of high-resolution yield in specialty crops.  While some studies get a single yield value for a whole block of a few acres, we’ll be getting a yield value every meter.”

That data becomes way more valuable pre-veraison versus post-harvest.  “This way you get your labor dialed in and plan your operation to get contracts pinned down,” Earles said.  This year’s field work will be on some 550+ acres in Lodi and Napa/Sonoma, obtaining data every three weeks from the front of a tractor as the plants and berries are forming and growing.

“We’ll take that yield monitor data prior to veraison and build models that will try to accurately predict what the yield will be at harvest. So, 500 acres this year and hopefully scaling that up to a thousand or 1,500 acres next year in different big block locations in Lodi and Sonoma to obtain a large enough data set of ground monitoring sensing data leading up to harvest.”

How would it help?

The logical on-the-ground question is if this concept were implemented today, how would it improve the life of grape growers?

“The most immediate way is if you could project, three months out, what your yield would be --- give or take 10 percent accuracy at the block level --- you could save a lot of money in not over- or under-scheduling harvest labor.  It could help in planning early season contract negotiations so you don’t end up with a surplus or a deficit.  It’s not uncommon now to be off by plus or minus 30% and this would help reduce that guesstimate, generating the variability of your vineyard and managing for that variability.

“If you could predict three months out, you could start managing your canopy earlier, checking more closely for diseases that might be driving projected yield values down so you could bring them back up before harvest and not take a financial hit in some of the block zones.”

With Earles and fellow researchers doing first-phase research and development into the piloting phase, commercialization would be the next step in this new technology.  “Our goal is to have this in 20+ vineyards in the next couple of years as a pilot phase refining its reliability.  The agricultural environment is a rugged and very dusty one and a unit strapped to the front of a tractor in the middle of the field for four months needs some engineering for reliability before moving towards commercialization.”

About the Author(s)

Subscribe to receive top agriculture news
Be informed daily with these free e-newsletters

You May Also Like