Computational corn hybrid selection integrating phenotypic, environmental, and genomic information

Chow, Tsz Yau Iris

Permalink

https://hdl.handle.net/2142/129925 Copy

Description

Title

Computational corn hybrid selection integrating phenotypic, environmental, and genomic information

Author(s)

Chow, Tsz Yau Iris

Issue Date

2025-07-10

Director of Research (if dissertation) or Advisor (if thesis)

Martin, Nicolas

Committee Member(s)

• Dokoohaki, Hamze
• Monteverde Dominguez , Eliana

Department of Study

Crop Sciences

Discipline

Bioinformatics

Degree Granting Institution

University of Illinois Urbana-Champaign

Degree Name

M.S.

Degree Level

Thesis

Keyword(s)

Maize hybrid prediction, Genotype-by-environment interaction (G×E), Corn yield modeling, Genomic selection, Machine learning in agriculture, Random forest classification , Principal component analysis (PCA), Regression and classification models , Feature importance analysis ,Genomes-to-Fields (G2F) Initiative , High-throughput genotyping , Public yield prediction competitions

Abstract

Accurate prediction of maize yield has been a persistent goal in plant breeding and agronomic research. Traditional approaches use primarily inferential stiatistics and mixed models, providing fundamental understanding of genotype × environment (GXE) interactions, albeit with relatively simplified statistical assumptions. Over time, technological breakthroughs—such as genomic selection, advanced remote sensing, and rich environmental datasets—have revolutionized the predictive modeling landscape. These modern frameworks harness machine learning (ML) algorithms, including random forests, gradient-boosted trees, and neural networks, to capture non-linearities in genetic, environmental, and management variables. Collaborative initiatives, including the Genomes-to-Fields (G2F) Initiative, have significantly expanded data availability and driven methodological innovation. Despite these advancements, challenges persist in addressing complex GXE interactions, the expansive nature of hybrid × location testing, and the interpretability of complex ML models. The integration of phenotypic and multi-modal environmental data, including soil characteristics, weather metrics, and remote-sensing data, has shown potential for improving the prediction of yield variability. However, scaling these innovations across diverse agricultural systems presents substantial practical and computational challenges. Subsequent chapters explores a suite of modeling strategies aimed at improving hybrid yield prediction under varied environmental conditions. The first chapter reviews historical evolution of predictive modeling in maize breeding, underscoring the shift from traditional statistical paradigms to advanced ML-based frameworks. Chapter 2 approaches the yield prediction problem as a classification task by determining whether a hybrid can outperform standard local checks, using an extensive dataset of over 180,000 field plots and 2,500 hybrids across 162 unique environments. Chapter 3 introduces a regression-based approach in the context of a public Genotype-by-Environment competition, wherein genetic, environmental, and phenotypic data are integrated to forecast future yield performance—even for hybrids and years not represented in the training dataset. Collectivelly , these chapters illuminate both the opportunities and obstacles in modern maize breeding research. Key findings include the efficacy of ensemble ML for GXE capture, the potential of synthetic APSIM variables, and the value of diverse methodologies (e.g., logistic regression, deep neural networks) in understanding yield dynamics. Ultimately, this thesis aims to guide future efforts in refining predictive frameworks that can robustly and transparently support breeders, farmers, and researchers in meeting the global demand for sustainable and high-yielding maize production.

Graduation Semester

2025-08

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/129925

Copyright and License Information

Copyright 2025 Tsz Yau Iris Chow

Owning Collections