Pairing genomics and weather data to make crop predictions more powerful

All farming is, in a sense, a prediction business. Crops are planted in anticipation they'll grow and produce a bountiful harvest. Research led by Iowa State University agronomy professor Jianming Yu aims to give a major boost to agricultural predictions with powerful modeling tools that would benefit both breeders and farmers.

In a recent study published in Plant, Cell & Environment, Yu's team built a model to predict the flowering time and height of sorghum plants based on genomic analysis and early-season weather data. In blind tests of sorghum fields that weren't included in the modeling data, Yu and his colleagues predicted flowering time with up to 74% accuracy and plant height up to 96%.

Similar models integrating genomics and environmental conditions could be used in other crops and to predict equally complex traits such as yield, said Yu, the Pioneer Hi-Bred Distinguished Chair in Maize Breeding and director of Raymond F. Baker Center for Plant Breeding.

"Getting information about what's likely to happen ahead of time has tremendous value," he said.

The study was based on data from the Sorghum Association Panel, a collection of 400 variants of sorghum selected to represent the plant's global diversity and genetically sequenced for research and breeding. The group of 16 researchers from eight different institutions analyzed 14 growing seasons to see which combinations of weather factors over what timespans connected most closely with plant height and flowering time, which is also called the environmental index.

For flowering time, researchers established the environmental index as heat accumulated in the second month of growing. For plant height, it was the diurnal temperature range—the difference between the daily high and low temperatures—from day 25 to 31.

Yu's team then used a method of statistical analysis called genome-wide association studies to search about 265,000 genetic markers for groups of genes linked to plant height and flowering time, identifying seven groups for flowering time and 69 for plant height. Another analytical method, called genomic prediction, was used to build models that predict performance based on genetic similarities.

Researchers integrated an environmental index into their modeling to account for the differing ways genetically identical plants respond to growing conditions, a dynamic called phenotypic plasticity that's a focus of Yu's work in recent years. Considering the varying range of performance—as opposed to only the average—makes it easier to compare large data sets across a wide variety of environments, leading to more effective models, Yu said.

"Incorporating an environmental index makes genomic prediction more dynamic and accurate. You can study all the data together and have a deep and holistic view for finding patterns," he said.

That should speed up and broaden the geographic reach of plant breeding, improving efforts to make crops more resilient to the weather extremes of climate change.

"We really hope to establish processes that help breeders use our phenotypic plasticity framework in their own practices," Yu said.

But accurately modeling crop traits also holds direct potential for farmers, who could use reliable in-season forecasting to make important field management and marketing decisions, Yu said. That could be even more possible as his research group looks to further bolster prediction models with additional input, such as drone-collected in-field data.

"We'll keep thinking about how to make these predictions more powerful," he said.