Evaluation of sugar beet plants in the field (Credit: USDA ARS)
COMPUTER VISION AT SCALE

High-Throughput Phenotyping

Turning terabytes of drone and sensor imagery into biological insight, field by field.

Image: USDA ARS

The Bottleneck Has Moved: From Genotype to Phenotype

Sequencing a plant's genome is now routine. Measuring how that genome actually performs — growth rate, leaf health, drought response, yield — across thousands of plots and multiple growing seasons is the harder problem. High-throughput phenotyping replaces slow, manual field measurements with drone, satellite, and sensor imagery processed by deep learning pipelines at a scale no human team could match.

The HPC Architecture Behind Field-Scale Phenotyping

A simplified view of how raw imagery becomes actionable trait data:

Image Capture Drones & ground robots Multispectral / satellite Preprocessing Orthomosaic stitching 3D point-cloud generation GPU-Accelerated Cluster CNN segmentation per plot Trait extraction (height, cover) Multi-terabyte throughput Trait Database Per-plot phenotype values Time-series growth curves Genotype-Phenotype Correlation Matched against genomic markers Feeds back into breeding selection Next flight targets flagged plots Terabytes of imagery per season become a structured trait database ready for genomic analysis.

Simplified data flow: field imagery is stitched, segmented, and quantified by GPU-accelerated deep learning, then matched against genomic data.

TOP500 · Rank #89 (2025 debut)

Who Works on This: DeltaAI at NCSA (University of Illinois)

DeltaAI, operated by the National Center for Supercomputing Applications (NCSA) at the University of Illinois Urbana-Champaign, entered the TOP500 list at No. 89 — a strong showing among a list that includes systems run by private industry as well as national labs. Alongside its sister system Delta, it is one of the NSF ACCESS program's most heavily used resources for AI and machine learning research.

NCSA's Center for Digital Agriculture uses this infrastructure to support researchers like plant physiologist Lisa Ainsworth, who studies how crops such as maize and soybean respond to rising CO₂ and ozone — work that increasingly depends on processing large-scale drone and sensor phenotyping data alongside climate models.

Voices from the Field

Lisa Ainsworth

University of Illinois Urbana-Champaign / NCSA

On how NCSA's computing resources enable better understanding and modeling of crop responses to climate change.

Read more
Claudia Kamphuis

Wageningen University & Research

On what computer vision can — and can't — reliably measure in living organisms, twenty years into the field.

Read interview
SDSC Researchers

San Diego Supercomputer Center

On turning vineyard sensor data into irrigation and microclimate predictions using machine learning.

Read article