With Europe investing strongly in the systematic development of remote sensing and imaging technology to monitor and predict crop performance, Australia has launched a parallel project in partnership with European laboratories.
The Australian complement of the EU initiative was made possible with GRDC investment in INVITA – Innovations in Variety Testing in Australia.
The initiative leverages advances being made within the Horizon 2020 European initiative INVITE – Innovations in Plant Variety Testing in Europe.
Heading INVITA at the University of Queensland is Professor Scott Chapman, who has long-standing ties with a key INVITE partner, Wageningen University in the Netherlands, which is one of the top-ranked universities in the world for agricultural research. CSIRO Agriculture and Food is the other major collaborator.
Professor Chapman says INVITA embraces the use of drones, satellite images and paddock-based sensors and cameras, with experts in statistics, modelling, machine learning, crop physiology, crop computer simulation and satellite imaging taking part.
The goal is to utilise this technology for agricultural applications, in particular within the National Variety Trials (NVT). The two main goals are to improve:
- the prediction of performance at single sites, through use of imaging on the ground and by drone, and using this data to account for spatial field variability; and
- the ability to account for variation in crop performance in relation to environmental effects.
“The crop monitoring technology will focus on capturing data on environmental conditions in addition to crop growth, yield and health characteristics,” Professor Chapman says. “That will aid us in developing analytical tools capable of predicting crop performance in ways that account for impacts from environmental variables.”
Ultimately, this is research that seeks to understand what is the right variety for the wide mix of environmental conditions experienced in the grains industry.
An invitation too good to refuse
INVITA launched in March 2020, with field trials focused on bread wheat germplasm in the first instance. Preparations are underway to include sorghum and other crops over summer. In the more advanced European arm, data is also being captured for canola, maize and sunflowers.
“A total of 90 field sites have been planted to host the remote-sensing technology,” Professor Chapman says. “Included are a subset of about 50 NVT locations which were selected to host ancillary trials.”
The trials fall under two classes: BioCal and SatCal sites. The SatCal sites include satellite data collection to provide a constant point on the Earth to align the satellite images with NVT trials. These sites have been planted with the same wheat type.
The BioCal (short for biomass calibration trial) sites involve six different wheat types (sourced from the NVT) planted at different densities and sampled three times during the season. These sites are imaged using drones equipped with RGB (standard red-green-blue cameras) or with multispectral cameras that include NIR (near infra-red) and NRE (near red edge) wavelengths to match with satellite imaging. The drone data is processed to generate detailed field maps at less than one-centimetre resolution.
In addition, the BioCal sites include fixed 4G cameras that transmit single-plot images four times a day, canopy temperature sensors and, at a few sites, Arable Mark weather station/reflectance sensors.
Three trial sites managed by the University of Queensland, CSIRO and the South Australian Research and Development Institute are also subject to more intensive within-season crop performance assessment (or phenotyping) than the NVT sites.
The INVITA sites are distributed across Queensland, Victoria, New South Wales, South Australia and Western Australia.
“This aspect of INVITA is about understanding if there is a small amount of technology that we can add to NVTs that allows for a step change in the information we derive from the trials to explain cross-site performance,” Professor Chapman says.
Examples of the new capabilities include software that uses drone-derived data to estimate early crop cover or count wheat heads or, alternatively, uses satellite images and drone-derived imagery to estimate leaf area and biomass. These estimates can then be run against environmental conditions and other data to mine deeper correlations.
Along the way, project activities are helping to upskill NVT Service Providers (TSPs) in the use of remote sensing technology. As this becomes established technology, the broader agricultural research community also stands to benefit through the availability of a steady stream of paddock-relevant data.
“The strength of INVITE and INVITA’s approach is the ability to integrate the satellite, drone and observational data while advancing the ability to extract useful information from images,” Professor Chapman says. “The result is an integrated platform that allows a new understanding about the way genotypes and environments interact that will prove especially useful when analysing performance od germplasm in the NVT trials.”
Learning to process Big Data
The resulting Data is uploaded to a cloud database and shared among the collaborators, who are developing advanced statistical and computational analytical methods. Wageningen University has a particularly strong track record in this space. For parts of the work, the methods used include machine learning and artificial intelligence algorithms.
This analytical work is already underway using historical NVT datasets.
“We ultimately want to develop environmental indices that provide a better explanation of genotype-by-environment interactions across the NVT,” Professor Chapman says. “So, one of the key questions we want to answer is whether better environmental typing can improve our ability to predict the performance of germplasm undergoing variety trials.”
This kind of understanding would ultimately provide insights about why the unique genetic make-up of a variety – its genotype – works better in some areas and what aspect of that environment influences that adaptation.
“The idea is to break away from a geographical partitioning of genotypes and cultivars to an environment-driven prediction of performance,” Professor Chapman adds. “Underlying this shift is the capability to simulate NVT trials based on a variety’s genotype and a range of simulated environmental indices.”
The 2020 launch of INVITA required the purchase and distribution of drones to the NVT Service Providers, training in their safe operation, training in imagery collection and data submission for analysis (contracted to Hiphen, France), writing of field, sensor and data management protocols and data collection templates.
Sensors were also installed and satellite data collection was contracted (0.5m satellite data; RGB, near-infrared wavelengths).
A project manager (Dr Chelsea Janke) and data scientist (Dr Swaantje Grunefeld) have been appointed at the University of Queensland, with the reference trials being largely coordinated by Daniel Smith (GRDC PhD scholar).
Dr Andries Potgieter and Dr Yan Zhao oversee the satellite data collection and processing. CSIRO staff (Dr Bill Bovill and Jamie Scarrow) are overseeing the phenotyping at CSIRO Boorowa Research Station in NSW, while Dr Kenton Porker oversees trials at the SARDI Palmer site.
The team of Dr Bangyou Zheng (CSIRO), Dr Karine Chenu (University of Queensland) and Professor Chapman lead a component of the research that utilises crop simulation to generate environment indices and creates ‘virtual trial datasets’ for use by the Wageningen team.
Datasets from previous GRDC/CSIRO trials (Dr Anton Wasson) have been provided to Wageningen (Professor Fred van Eeuwijk, Dr Daniela Bustos-Korts, Dr Martin Boer) to test their preliminary methods to analyse within-trial spatial variability. Dr Vivi Arief (University of Queensland) is coordinating this work and is organising requests for historical NVT datasets on which to apply new analytical methods. Associate Professors Fred Roosta and Yoni Nazarathy of University of Queensland School of Maths and Physics will provide support on machine learning methods in the latter part of the project.
More information: Professor Scott Chapman, firstname.lastname@example.org