Cross-scale climate regionalization for crop phenotype and hybrid selection predictability in the high plains

Overview

This solution addresses the challenge of predicting extreme precipitation, drought, and crop performance across the High Plains by integrating cross-scale climate regionalization with genotype-by-environment (GxE) analytics. The approach identifies sources of sub-seasonal climate predictability and uses them to improve phenotype forecasts and hybrid selection for maize and other crops tested at more than 100 U.S. locations. By coupling climate science with agricultural modeling, the offering provides breeders, seed companies, and agricultural decision-makers with a more reliable framework for evaluating how climate variability drives crop performance across diverse growing environments.

Technical specifications

Key features:

• Climate regionalization: Statistical regionalization of meridional wind and geopotential height data from the Climate Forecast System and observed sources to identify sources of rainfall and drought predictability across the High Plains
• GxE modeling: Aggregates genotypic, phenotypic, and hydroclimate data from the Genomes-to-Fields (G2F) initiative covering more than 100 sites since 2014
• Global sensitivity analysis (GSA): Couples with the GxE model to propagate climate data errors into simulated phenotypes, revealing spatio-temporal attributions of climate influence on model performance
• Deep neural network gap-filling: Uses deep learning to intelligently fill climate data gaps in the G2F database, expanding usable environmental coverage beyond the current constraint of approximately 30 sites per year
• Enhanced environmental covariance matrix: Produces a richer climate database that supports regionalization and improves hybrid selection predictability

Technology readiness level

The methodology has been validated using climate data from 2014 to 2016 across the G2F network, with initial results showing mild improvements in GxE model performance when regional hydroclimate data is aggregated. Current limitations include data availability and quality constraints that restrict usable sites to roughly 30 per year over three years. The next validation phase will deploy GSA-coupled GxE modeling and deep neural network gap-filling to build an expanded climate database, define the limits of climate regionalization on model performance, and further refine phenotype and hybrid selection predictions. The approach is positioned for collaborative refinement and pilot-scale application with breeding programs and agricultural partners.