Field-based high-throughput phenotyping and machine learning platform for climate-resilient crop breeding

Overview

This offering combines a large-scale, field-based phenotyping infrastructure with custom machine learning analytics to help plant breeders and crop scientists identify genetic variation underlying stress-adaptive traits. By tracking crop phenotypes over time under water-limited and well-watered conditions, the platform reveals how plants respond to drought and heat and which genetic factors drive those responses. The goal is to accelerate the development of climate-resilient cultivars that maintain economic yields with reduced water and other inputs.

Technical specifications

Core infrastructure and capabilities:

• UA Field Scanner, described as the world's largest outdoor phenotyping robot, used to characterize phenotypic variation in wheat, sorghum, and lettuce under water deficit stress
• Scalable, modularized data processing workflows capable of handling more than 10 TB of imagery generated per day
• Multi-modal sensor data including RGB imagery, thermal imagery, 3D point clouds, and hyperspectral data
• Extracted traits including canopy height, canopy cover, canopy temperature, photosynthetic efficiency, solar-induced chlorophyll fluorescence, architectural features (leaf angle, width, length), and growth curves and rates
• Machine learning algorithms for individual plant identification and seasonal tracking, plus planned deep-phenotype detection and quantification
• Open dissemination of data and analytical tools to the broader phenomics community through CyVerse

Technology readiness level

The platform has already been used to generate and process field data for multiple crop species and to extract heritable traits, demonstrating proof of concept in real field environments. Future work focuses on expanding analytical capacity through novel machine learning models targeting deeper phenotypes and on validating temporal responses of sorghum to combined heat and drought stress. The infrastructure and workflows are operational, with ongoing development aimed at broadening trait coverage and community access.