A multi-faceted project that has been exploring how to use sensors to inform vineyard management strategies may well be on the cusp of a pioneering technology: a low-cost, foliage penetrating radar that can measure fruit mass on the vine at scale immediately before harvest.
‘If it works, it will be a gamechanger in terms of accurate yield prediction – but it’s early days yet’, said Dr Everard Edwards, from CSIRO Agriculture and Food.
Dr Edwards is leading a project in the Wine Australia and the CSIRO Strategic Partnership Agreement, which continues work previously supported by Wine Australia and the Australian Government’s Rural R&D for Profit program and was co-led by CSIRO’s Dr Mark Thomas.
Dr Edwards has targeted three phenological periods to determine how best to accurately predict yield for growers – each with distinct advantages and disadvantages.
‘Bud dissections are the earliest indicator, but are limited in scope, expensive to do and are not amenable to being replaced by a vehicle mounted on-the-go sensor[1]. Therefore, our earliest estimate is of yield potential and we are doing this with simple consumer action cams (e.g. GoPro) and machine learning tools to identify, track and count inflorescences after budburst.’
The team’s second target is yield prediction from fruit that is pea-size onwards. To estimate this, they have been using video cameras and machine learning tools to count not only developing bunches, but also berries per bunch.
‘This provides a more direct prediction but is more difficult once the canopy reaches its maximum size and covers many of the bunches. If a bunch can’t be seen from a particular angle it can’t be assessed’, Dr Edwards explained.
The team’s third target is yield estimation in the weeks leading up to harvest.
‘This can be done using the same techniques as previously, but we are also looking at the foliage penetrating radar to directly assess fruit mass which, if successful, would negate the problems of occlusion by canopy.’
Dr Edwards said the technology to determine yield potential post-budburst was probably the most well developed application – and worked well even in vigorous vines.
In other work, the use of hyperspectral imaging/sensing in the near infra-red range has allowed the team to remotely examine the chemical composition of the target.
‘This can be fruit (for estimating maturation and quality) or canopy (for estimating nutrition status)’, Dr Edwards said.
‘However, hyperspectral imaging requires very expensive and quite sensitive equipment, so we are developing a simpler and more robust system more suited to use by vineyard managers.
‘We are concentrating on measuring fruit composition at present and hope to be able to give growers the ability to accurately measure fruit maturity and quality simply by driving down the row in a vineyard vehicle.’
Finally, the team has been using LiDAR to examine the structure of the whole canopy, ‘seeing between leaves to the inside of the canopy and producing quantifiable measures of the canopy structure, such as density, size and light interception.’
‘For growers that use detailed canopy management, such as leaf plucking, this will allow them to associate a canopy structure with a desired outcome and reliably reproduce that season after season, with less dependence on the skills and expertise of the vineyard manager.’
Dr Edwards said the LIDAR, ground-based video imaging and drone-based imaging all provided methods to estimate canopy size and/or ground cover.
He said the tools would help growers finesse canopy management strategies to maximise fruit quality under both high and low production targets.