Skip to main content
Ongoing project

Advancing the delivery of national mapping applications and tools (AV21006)

Key research provider: University of New England

What's it all about? 

This project will further position Australia’s avocado industry as world leaders in the sector, by delivering growers commercial tools for improved yield forecasting and mapping from the orchard block to the national scale.

The project provides further improvements and investigations to support the delivery and long-term relevancy of AV18002, delivered by the University of New England (UNE). This project saw the implementation of a precision agriculture solution in Australian avocado production systems and the creation of a multi-scale monitoring tool for managing Australian avocado tree crops.  

AV120006 seeks to contribute to this body of work by achieving the following objectives:

  • Continue to update the web mapping applications with improved accuracy and usefulness to the avocado industry.
  • Yield forecasting model to support a benchmarking project, crop forecasting, the investigation into crop developments relationship with the climate and inform the remote sensing climate-based yield prediction model.
  • Use the CropCount mobile application to improve productivity by expanding testing and structured feedback processes, improving decision-making and orchard management.
  • Use the CropCount mobile application to support avocado growers who have new plantings or no historic data.

In terms of the ongoing mapping of avocado orchards, the second regional update of ‘The Australian Tree Crop Map Dashboard’ has been completed, with mapping updates completed in the Sunshine Coast and South-east Queensland, Northern and Central NSW, including the Comboyne. The mapping team has also updated Bundaberg and Childers. Updates elsewhere around Australia continue to be supported via the location-based tools (survey) hosted on the Avocados Australia website. 

The ‘CropCount’ or 18 tree sampling methodology was evaluated over four orchards in Western Australia with the accuracies when compared against pack house data ranging from 74% to 98%. To achieve greater adoption, the original methodology was amended to better align with the current practice of the participating growers. This included the measure of yield from only 9 trees instead of 18 trees within two of the orchards and growers using a ‘panel’ count, where the yield of 6 trees was combined in 3 locations across the orchard, in the remaining two orchards. Although both methods reduced the number of calibration points (n=9 and n=3 per orchard), the accuracies achieved exceeded current practice i.e. growers estimate.
The ‘Time series’ forecasting approach was further validated, with new models developed for 20 orchards (3 farms) in Western Australia. Since the actual yield (t/ha) data for all blocks were not yet supplied, the models accuracy were compared against growers estimates. Due to the effect of ‘Irregular bearing’ (most blocks being ‘off season’ in 2022), the models over predicted 1.15 to 3.35 times higher than the grower estimation. To better understand and therefore predict the occurrence of ‘on’ and ‘off’ years from remotely sensed data, Machine Learning (ML) algorithms have been initially developed and trialed over 35 orchard blocks in Bundaberg. Two new farms (31 blocks) from Western Australia have also been added for the upcoming season, with predictions already provided to the corresponding growers.

For the extension of results, the AARSC presented at the recent World Avocado Congress New Zealand 2023. With some 1,200 delegates from 32 countries in attendance, Dr Moshiur Rahman and Craig Shephard each presented in the academic program and a keynote address was delivered by Prof. Andrew Robson. Andrew's presentation highlighted the need for agritech issues to be application led rather than technology led, "Only believe half the hype and ask for evidence of real world application". Considering Australia only produces 1% of global avocado production, the privilege of presenting these talks demonstrates the impact and respect that the Australian research receives on the global stage. This was confirmed by additional farms in Australia requesting involvement in the current project and commercial contracting being developed for yield forecasting in New Zealand (NZ Avocado) and South Africa and Africa (Westfalia). This commercialisation aligns with the required outcome of this project to identify commercial partners and to see the research commercialized and as such Hort. Innovation are working with AARSC to best develop these opportunities.

The first regional update of ‘The Australian Tree Crop Map Dashboard’ has been completed in the intensive growing regions in Queensland, including Atherton Tablelands, Bundaberg, and Childers. Ongoing updates continue to be supported via the location-based tools (survey) hosted on the Avocados Australia website. The annual tree census undertaken by Avocados Australia encourages growers to contribute by completing a survey, which continues to provide an ongoing source of industry engagement to action updates on the map, available here.

For the evaluation of 18 tree sampling ‘CropCount’ methodology for yield forecasting, three farms were selected in South Australia and six farms in Western Australia. Due to the early harvest of avocado crops in South Australia, the methodology was, unfortunately, unable to be applied as the orchards were harvested before the 18 trees could be sampled. For the WA region a Pleiades NEO-3 image (30 cm 6-band panchromatic imagery and 120 cm 6-band multispectral imagery) was acquired over 87 sqkm area on 24 and 30 September 2022 to cover 5 farms (at the harvest maturity stage of the orchards). An additional high-resolution Pleiades NEO-3 image as well as a Planet image were acquired over an additional WA farm on the 24 September 2022. Classified maps showing variability in tree vigour across the orchard blocks as well as the 18 tree sample location were generated from the satellite imagery captures and provided to the growers as ‘WebApps’. The harvested yield data of the 18 sample trees from 1 WA farm has already been provided by the participating grower and the subsequent yield prediction for the entire orchard block returned. For the second farm the measured yield from the 18 trees is yet to be provided. For the third WA grower (operating 4 farms) a slightly varied image calibration method is being adopted to better align with their current forecasting methodology. Instead of 18 sample trees per orchard, the grower is providing yield weights from clusters of trees across their orchard. When the yield data has been received, it will be used like the 18 trees to calibrate the satellite imagery to deliver yield variability maps and total yields per orchard block to the grower.

Corresponding with the ‘CropCount’ methodology, ‘Time series’ models were developed for 20 orchard blocks in Western Australia (from three farms). The predicted yield will be compared to actual yield following the commercial harvest. The ‘Time series’ model prediction for 37 blocks in the Bundaberg region produced a 92.2 per cent accuracy for total farm yield when compared to actual harvested data. The benefit of the ‘time series’ method is no fruit counts are required, it uses freely available imagery and forecasts can be provided months before harvest. The downside is that predictions are more aligned with farm level predictions, rather than within orchard blocks as with the ‘CropCount’ method and at least five years of historic yield data is required.

Related levy funds

This project is a strategic levy investment in the Hort Innovation Avocado Fund