Embedding big data and machine learning into everyday farming could soon help Australia significantly boost its food production to meet growing demands without degrading soil and water quality or overusing fertilisers.
That's the vision a University of Sydney research team, led by associate professor Thomas Bishop, will put to the test as they look for better ways to harvest information literally in the field by finding novel approaches to precision farming.
The individual datapoints might be small, but the vision is large with agriculture taking up 56 percent of Australia’s land surface and generating a wealth of data in the process - much of it currently unconnected.
To get a much clearer picture of agricultural performance, researchers will investigate combining disparate data sets surrounding crop yield, weather, and management practices to better predict crop volumes and quality.
This will guide researchers and the agricultural industry on how best to apply fertilisers to maximise grain output and quality, without increasing the amount of chemical runoff into waterways.
Limiting chemical runoff from fertilisers has been an enduring environmental problem in Australia, not least because of downstream consequences from flows that can negatively affect aquatic ecosystems.
Although localised yield forecasting tools are currently available, the researchers say these can be unreliable due to the number of complex variables whose interactions can be difficult to predict over a growing season.
Other “mechanistic” models for prediction involve collecting a large amount of data at multiple sites within a field, which the researchers say can be difficult for farmers to capture.
Bishop told iTnews the project will leverage the crop yield data collected by around 40 percent of grain growers across Australia.
The work will also be supported by Wesfarmers-owned CSBP Fertilisers, which collects over 60,000 soil samples and 20,000 plant samples from across Australia each year.
Bishop said rather than analysing one paddock at a time, the researchers will be looking at data on a continental scale.
“The approach we will adopt is that rather than building models on one paddock at a time as is typical, we will pool data from multiple paddocks across multiple farms, districts, regions to create a data-driven model to forecast yield, which can be used to vary management decisions spatially.
“This will end lead to a decision support tool using Google Earth Engine infrastructure,” Bishop said.
The mapping tool would inform more equitable “nutrient budgets” that reduce the variability of crop quality between and within paddocks, as well as across different seasons.
This is made possible by the developments in weather forecasting, regularly updated data from drones, remote sensing platforms such as Landsat and NASA’s MODIS program, and existing soil properties maps that cover Australia at a roughly 90-metre resolution.
The University of Sydney estimates the Australian agriculture industry currently supplies around 90 percent of the food consumed locally and contributes around 13 percent to the country's export earnings.