Industries / Food, farming and forestry / Agriculture and farming / Viticulture
Ripeness and harvest
Deciding when grapes are ripe, picking them by hand or by machine, and handing them over at the winery's weighbridge, where each load is weighed and graded.
5
Milestones achieved, from 1820s to 2020.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1820s
Grape sugar gets a scale
Growers had used hydrometers (floats that sink less in denser liquid) to judge grape juice for more than 300 years, but in the 1820s the Pforzheim instrument maker Ferdinand Oechsle gave them a scale. It rests on the fact that sugary juice is denser than water, so a reading tells how much sugar the grapes hold and how ripe they are. The Oechsle scale is still the measure of ripeness in German and Swiss vineyards.
Credit
Ferdinand Oechsle; account by the Historisches Museum der Pfalz, Speyer
1957
A machine shakes grapes loose
Stanley Shepardson, an agricultural engineer at Cornell University, introduced a harvester that straddled the row and knocked grapes off with fast, short "bumping," so juice and wine grapes could be picked without crushing them (Cornell). The Chisholm-Ryder Company of Niagara Falls, New York, built the Cornell machine for sale in 1963 and a rival design by growers Max and Roy Orton in 1967 (Wine Spectator). Cornell's Terry Bates says New York's juice grapes went from 100% hand-picked to 100% machine-picked between the late 1960s and the mid-1970s. Machines spread to Europe in the 1970s, when farm labor ran short, and Stellenbosch University reports that more than 85% of Australia's wine grapes are now machine-harvested.
LessonMachines came first for cheap juice and table-wine grapes, and growers of premium grapes still prefer hand-picking when they can pay for it (Stellenbosch University).
Credit
Stanley Shepardson and Nelson Shaulis, Cornell; Max and Roy Orton; Chisholm-Ryder; Vito Mecca
1999
Yield mapped vine by vine
In 1999 a mechanical harvester fitted with satellite positioning (differential GPS, accurate to about 50 cm) and a HarvestMaster yield monitor mapped how much fruit came off each part of a 4.3 ha vineyard at Padthaway, South Australia. Work led by Rob Bramley of CSIRO, Australia's national science agency, found that yield inside one block managed as a single unit varies about 10-fold, from 2 to 20 tonnes per hectare, in a clear pattern that follows soil and slope. This began precision viticulture: managing parts of a block differently.
LessonTreating a block as uniform hides a tenfold spread in yield.
Credit
Rob Bramley, CSIRO; Tony Proffitt, Southcorp
2014
Cameras count the berries
A Carnegie Mellon University team, working with Cornell's Terry Bates, mounted cameras and lights on a vehicle and used software to find and count grape berries, including green berries against green leaves. In 2011 they predicted the yield of single rows, over 224 vines, to within 9.8% of the weighed crop. Their 2014 paper in the Journal of Field Robotics covers four seasons in several wine and table-grape vineyards, with an average error of 3% to 11% of total yield. A separate team (Aquino, Millan, Diago and Tardaguila, Computers and Electronics in Agriculture, 2018) repeated the approach with a vehicle that photographed five varieties at night about 100 days before harvest, and predicted yield with an error of 0.16 kg per vine.
Credit
Stephen Nuske, Supreeth Achar, Srinivasa Narasimhan and Sanjiv Singh, Carnegie Mellon; Terry Bates, Cornell
2020
Smoke shuts out a harvest
California's 2020 wine grape crop fell 13% by the US Department of Agriculture's count, and experts blamed wildfire smoke in Sonoma, Napa and Monterey counties (Sacramento Bee). A report by the law firm Downey Brand for two grower groups, Allied Grape Growers and the California Association of Winegrape Growers, put the grapes left unpicked over smoke fears at 165,000 to 325,000 tons (about 150,000 to 295,000 tonnes), with crop insurance claims of $187.9 million. Testing backlogs at West Coast laboratories stretched past harvest, and the report found some wineries rejected fruit on inconsistent standards, which winery lawyers disputed.
LessonGrape contracts had no agreed smoke test or limit, so slow tests and last-minute rejections left growers holding fruit they could not sell.
Credit
a reversal; reported by the Sacramento Bee and the North Bay Business Journal
Watch for
Robots that pick by ripeness
The EU-funded BACCHUS project says its two cooperating robots judge ripeness with hyperspectral imaging (a camera that records many narrow bands of light), find each bunch's stem with a camera next to the cutter, and pick only ripe bunches with two arms. Its paper says the system was validated in real-field trials, but the published summary gives no number of bunches attempted, no success or damage rate and no time per bunch.
Achieved whena trial in a working wine vineyard publishes bunches attempted, picked, damaged and missed, and time per bunch, checked by someone outside the project.
Claimed by
the BACCHUS team (Sotiris Stavridis and 24 co-authors) in IEEE Robotics & Automation Magazine, September 2025, restated in a May 2026 report on EU vineyard projects
Open
01
Score rot at the weighbridge
A camera station over the sample tray at a winery weighbridge that gives each load a percentage of rotten fruit and of MOG ("matter other than grapes": leaves, stalks, canes, insects), logged with the load's weight, plus an open set of labeled photos so others can check it.
The briefCredit · unclaimed02
Score the ripeness forecasts
A public, season-long test of harvest forecasts. Measure sugar (in degrees Brix, the percentage of sugar in the juice) weekly from veraison (the point when berries soften and change color) to harvest on at least 30 blocks, record the real pick dates, and score each available forecast against them: paid satellite services and a free baseline built from public weather and satellite data.
The briefCredit · unclaimed
Open 01 · the brief
Score rot at the weighbridge
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A camera station over the sample tray at a winery weighbridge that gives each load a percentage of rotten fruit and of MOG ("matter other than grapes": leaves, stalks, canes, insects), logged with the load's weight, plus an open set of labeled photos so others can check it.
Why it is still open
At delivery, MOG in machine-harvested loads is still judged by eye, and the Australian Wine Research Institute's Paul Petrie said "there is always room for conjecture around the accuracy of the MOG assessment" (Wine Australia, 2021). His project showed in 2019 that cameras can pick out bunch rot and MOG in weighbridge samples, but ended as a proof of concept, recommending better lighting and a robust system; no commercial system is reported. A French team's 2023 paper on hand-picked Champagne grapes reports 58% overlap between its detected disease areas and hand-drawn ones.
Built with what exists
A fixed color camera in a light box with halogen or high-color-accuracy LED lamps (AWRI found halogen gave the most reliable results); an open object-detection model such as Ultralytics YOLO, trained on the team's own labeled photos; a labeling tool such as Label Studio; a small computer such as a Raspberry Pi 5 or NVIDIA Jetson Orin Nano; a laboratory ergosterol test (ergosterol is a sterol only the rot fungus makes) on a subset of samples as the reference (Steel and colleagues, 2020).
First step
Build the light box and agree with one winery to photograph each weighbridge sample next to its assessor's score from the next vintage, which begins in February 2027 in the Southern Hemisphere.
Done looks like
Over one vintage at one winery, at least 500 loads scored by the camera and by two human assessors, with ergosterol results on at least 100, and a published table of how far the camera's rot and MOG percentages fall from the assessors' scores and from the laboratory.
Open 02 · the brief
Score the ripeness forecasts
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public, season-long test of harvest forecasts. Measure sugar (in degrees Brix, the percentage of sugar in the juice) weekly from veraison (the point when berries soften and change color) to harvest on at least 30 blocks, record the real pick dates, and score each available forecast against them: paid satellite services and a free baseline built from public weather and satellite data.
Why it is still open
Forecast sellers publish their own accuracy figures without a method. Deep Planet's February 2024 presentation to Suisun Valley growers claims "90%+ accuracy" for its yield prediction and says its maturity predictions from satellite images reach "high accuracy" after four readings, without saying what was measured against what. Searches of Crossref, OpenAlex and GitHub found no independent comparison of these forecasts.
Built with what exists
A handheld digital refractometer (a device that reads sugar from a drop of juice), such as the Atago PAL-1; free Sentinel-2 satellite images from the Copernicus Data Space Ecosystem; free weather data from Open-Meteo; Python with pandas for scoring.
First step
Collect last vintage's sugar records and pick dates from five growers and test a simple heat-sum model (one that adds up daily warmth) against them this month.
Done looks like
Weekly samples of 200 berries from 30 or more blocks across one season, and a published table giving each forecast's error in degrees Brix at 7 and 14 days ahead and its error in days on harvest date.