KGML-N: a knowledge guided machine learning modeling framework for efficient simulation of N-yield response
Liu, Qi
This item is only available for download by members of the University of Illinois community. Students, faculty, and staff at the U of I may log in with your NetID and password to view the item. If you are trying to access an Illinois-restricted dissertation or thesis, you can request a copy through your library's Inter-Library Loan office or purchase a copy directly from ProQuest.
Permalink
https://hdl.handle.net/2142/132712
Description
Title
KGML-N: a knowledge guided machine learning modeling framework for efficient simulation of N-yield response
Author(s)
Liu, Qi
Issue Date
2025-12-12
Director of Research (if dissertation) or Advisor (if thesis)
Guan, Kaiyu
Committee Member(s)
Lee, DoKyoung
Peng, Bin
Department of Study
Natural Res & Env Sci
Discipline
Natural Res & Env Sciences
Degree Granting Institution
University of Illinois Urbana-Champaign
Degree Name
M.S.
Degree Level
Thesis
Keyword(s)
machine learning
nitrogen fertilizer-yield response
Abstract
Applying the optimal amount of nitrogen (N) fertilizer is a critical challenge in corn farming, balancing crop yield against fertilizer costs and environmental risks. While scientific tools exist to guide these decisions, they face a trade-off. Complex process-based (PB) models are detailed but inefficient for large-scale use, while simpler data-driven models are fast but often unreliable under new climate conditions.
This thesis confronts this challenge by developing and validating a surrogate model KGML-N—a highly efficient machine learning model trained to emulate the functions of a comprehensive PB model (ecosys). We designed a Knowledge-Guided Machine Learning (KGML) model, named KGML-N, whose architecture is inspired by the known scientific principles of crop growth and N cycling. This approach aims to retain the scientific integrity of the original model while dramatically increasing its computational speed. Our research question focuses on whether KGML-N can maintain high fidelity to ecosys and achieve consistency with existing knowledge about environmental impacts on N-yield response.
To achieve this objective, we generated a large synthetic dataset over the 3I (Illinois, Iowa, Indiana) states by running ecosys across thousands of scenarios covering 21 years of weather regimes from 2000 to 2020 and various soil types sampled for each county. The KGML-N model was trained on this data to predict yield and a dozen key environmental variables. The model was then calibrated using real-world experimental data and tested through a series of virtual experiments involving changes in temperature, precipitation, and soil organic carbon.
The results demonstrate the high performance of KGML-N. Our KGML-N model achieved high fidelity, accurately reproducing the predictions of the original ecosys model and correctly capturing complex interactions between crop characteristics, management, and the environment. Furthermore, its responses to climate and soil changes were consistent with decades of established agronomic knowledge. This work validates the KGML surrogate as a powerful, fast, and scientifically sound tool, providing a new pathway for developing more accurate and scalable N management recommendations for a sustainable agricultural future.
Use this login method if you
don't
have an
@illinois.edu
email address.
(Oops, I do have one)
IDEALS migrated to a new platform on June 23, 2022. If you created
your account prior to this date, you will have to reset your password
using the forgot-password link below.
