Climate and genetic data enhancement using deep learning analytics to improve maize yield predictability.

Area of Science:

• Agricultural Science
• Plant Genomics
• Climate Science

Background:

• Limited spatiotemporal data hinders accurate prediction of plant responses to climate change.
• Genomics, phenomics ('omics'), and environmental data are crucial for understanding plant performance.
• Large-scale experiments like Genomes to Fields (G2F) offer valuable datasets.

Purpose of the Study:

• To quantify the impact of enhanced climate data on maize yield predictability.
• To improve the G2F database by filling climate data gaps using deep neural networks.
• To assess the contribution of genetic and climate data enhancements to genotype by environment (G×E) models.

Main Methods:

• Utilized deep neural networks to reduce climate data gaps in the G2F database.
• Employed genotype by environment (G×E) modeling to analyze environmental covariance structures.
• Evaluated yield predictability using three trial selection schemes: randomization, ranking, and precipitation gradient.

Main Results:

• Achieved a 12.1% increase in maize yield predictability through climate and 'omics' database enhancement.
• Observed enhanced covariance structures in G×E models across all train-test schemes, indicating improved predictability.
• The 'random-based' trial selection scheme showed the largest predictability improvement by incorporating environmental variability.

Conclusions:

• Enhancing climate and 'omics' data significantly improves maize yield prediction accuracy.
• G×E modeling with improved datasets and appropriate trial selection strategies is key to robust yield forecasting.
• Integrating diverse environmental data, particularly through randomization, maximizes predictive gains in agricultural research.