Yield forecasting technology
Dr Mark Whitty and colleagues have every right to feel pretty pleased with their current research project; it’s on track, on time, has already ruled out the most expensive of three options being tested, and has thrown a bit of world-first research into the mix as a bonus.
The aim is to find a reliable and inexpensive way to use image sensing technology to forecast block-level crop yields that can replace the slow and very unreliable manual approach – and actually provide more information to grape growers.
Funded by Wine Australia, the three-year project led by the NSW Department of Primary Industries is being investigated by Dr Whitty and his team in the Faculty of Engineering at the University of NSW.
The brief was to develop three separate systems for comparison – one based on a high-precision sensor suite, one on a low-cost sensor suite, and the third on a smart device such as a mobile phone.
The good news is that the low-cost option of a GoPro camera mounted on a vehicle moving through the vineyard is performing well and that the more expensive option of using a GPS system with special sensors can be discounted as impractical.
The option of using a mobile phone app is also showing great promise, though the exact format is still being worked out.
‘The advantages of a smartphone are that pretty well everyone already has one and the inbuilt processing capability is great for doing spot-based analysis, whereas with the GoPro you achieve a wider coverage but in less detail’, Dr Whitty said.
‘A smartphone camera can capture in more detail specific things like flower count and berry numbers on individual bunches, and it may well turn out that a combination of the phone app and the GoPro camera is the best solution.’
Both systems allow relative yield maps to be generated from when the first leaves separate. That means the maps can be used to adjust management practices such as trimming, thinning or mulching during the season. The processed images from the GoPro can also be used to map non-bearing sections of canopy and to identify missing vines, which has the potential to help in the detection of trunk diseases such as eutypa.
The relative yield maps are converted into absolute yield maps using estimates from sampling locations on the ground or historical data. Forecasts using these maps not only provide an overall tonnage for the block, but also an indication of the variation within the block without using a yield monitor.
‘The aspect that went particularly well was developing the new algorithms for shoot counting really early; we hadn’t anticipated doing that’, Dr Whitty said. ‘We had anticipated it being just from around fruit set stage onwards but we are able to bring that estimation back by a month or so. As far as I’m aware, that’s a world first.’
It has been a comprehensive process, with more than 30 people involved in data collection and analysis in the last season alone. Trials have been run on Shiraz and Chardonnay blocks in Orange and Clare Valley and on two trellis types – single wire VSP and single wire sprawl.
‘We can’t just go out there and take imagery and use it for estimating yield’, Dr Whitty said. ‘As a research project we need to not only video every row but to manually count fairly large portions of that to be able to say this imaging process is working to a defined accuracy.’
The project will run until the middle of next year.