Predictive agriculture: The art of understanding uncertainty

Predictive agriculture has always been a part of agriculture. A farmer must make predictions before planting crops or selecting animals for breeding.

For more than 10,000 years, experience and the human eye have been important to make these predictions and in the last 100 years, scientists have developed many means of measurement and mathematical insights about the climate and soil drivers of environment and the genetics and physiology of crops.

However, in the digital era, decision-making is increasingly predictive - informed by a suite of data-driven tools that provide more precise information – such as genetic/genomic markers for breeding values; or improving operational efficiencies using technologies such as automated planting and harvesting technology.

Predictive agriculture differs from precision agriculture in its scope.

While drawing upon precision data, predictive agriculture also integrates a vast array of agricultural, biological, climate, and hydrological data and sources into a full system model – using artificial intelligence and algorithms to predict outcomes, manage inputs, and plan for system shocks and changes decades into the future.

“What we try to do as scientists is provide a whole series of tools that help plant breeders make better predictions about how to design a new variety that will work on that farm or work in that particular environment,” says Scott Chapman, Professor in Crop Physiology at The University of Queensland.

“We want to support informed decision-making during the season – things like whether to add more nitrogen to increase the protein to get a particular outcome, or how much specialty noodle wheat we're going to produce in Australia and where are we going to sell it?

“UQ has people working across multiple domains from developing the genetics, understanding the agronomy, understanding the farming system to how do we deliver into the supply chain. And that's all informing prediction.”

There has never been a more exciting time to work in predictive agriculture, Professor Chapman says.

“The changes that have happened in the last five years are incredible, particularly in artificial intelligence and in our capacity to calculate, to measure, to monitor systems and plants and genetics.

“For example, the same kind of technology that's used to identify people in crowds in an airport is the technology that we use to count 1000s of plants when I fly a drone over that field.”

For Professor Mark Cooper, Chair in Crop Improvement at The University of Queensland, the goal is integrating genomic prediction and crop growth models into an ‘end to end’ framework for crop improvement.

Professor Cooper has pioneered the development of novel genetic modelling methodologies, based on gene networks, to study important properties of quantitative traits in biology, and demonstrated how this new genetic modelling framework can be successfully used in plant breeding to improve prediction of important traits under the influences of selection.

Professor Cooper’s work at DuPont Pioneer/Corteva on drought adaptation in one of the largest maize breeding programs in the world led to the AQUAmax hybrids that presently cover millions of hectares worldwide.

“Essentially this was achieved by melding together two different modelling capabilities,” Professor Cooper says.

“There were quantitative genetics models that compare the performance of genetically diverse plants in field trials to detect the DNA sequences that can account for high-performing traits but are not as accurate in predicting how plants will respond to variation in the environment, or to varied management practices.

“Key projects using these predictive technologies for crop improvement are also being developed in the new ARC Centre of Excellence for Plant Success.”

In the 1990s, researchers including Professor Graeme Hammer at the University of Queensland further developed applications using a different kind of computational tool – a crop growth model – which simulates physiological processes that are essential to growth, including processes that interact with the environment, such as the availability of light, water and soil nutrients.

“Rather than trying to associate a gene with yield, we work with those clusters of genes that underlie the ‘intermediate traits’ – the same traits captured in crop growth models,” Professor Hammer explains.

“The advantage is that the combined models can better predict how genes selected during a breeding program are likely to interact with seasonal weather variation and agronomy.

“This allows us to scale a breeding program even when it involves complex traits (such as drought tolerance) and the predictive software can subsequently even be used on-farm to predict crop performance and better manage risks.”

On the animal side of the equation, Professor Ben Hayes, Director of the Centre for Animal Sciences at UQ and the co-inventor of Genomic Prediction, says genomic selection was first implemented in dairy cattle.

“We had great success in estimating the breeding value of bulls for milk production early and precisely through genomic prediction equations rather than later through a costly progeny test,” Professor Hayes said.

“Now we are doing this for the north Australian cattle industry, through a suite of genomic prediction and Precision Beef strategies relevant for the industry – like fertility and calf mortality.

“What it does is connect the dots, bringing together researchers that deal with different industry issues on one hand, and combining our understanding within one integrated computer system on the other.”

Computer algorithms make it possible to backtrack the quality data against how cows were reared, their genetics, the nutritional value of pastures, the associated methane production and even stress levels experienced by herds.

Examples of Predictive Modelling