Dissertation
Reinforcement learning-based agricultural management subject to climate variability
University of Iowa
Doctor of Philosophy (PhD), University of Iowa
Spring 2026
DOI: 10.25820/etd.008478
Abstract
Agriculture, a cornerstone of global food security and economic stability, faces mounting challenges from rapid population growth, limited land availability, and the escalating impacts of climate change. This research advances sustainable farming by developing adaptive management strategies that address unpredictable weather patterns and environmental concerns, particularly greenhouse gas emissions from soils. Within the framework of Partially Observable Markov Decision Processes (POMDPs), the thesis integrates advanced machine learning methods to improve decision-making under uncertainty. Specifically, Recurrent Neural Networks (RNNs) and Deep Reinforcement Learning (DRL) are employed to design management strategies responsive to diverse climatic conditions and soil dynamics. Probabilistic deep learning models are further explored for predicting nitrous oxide N2O emissions—a major contributor to greenhouse gases—while incorporating climate variability and precision agriculture practices. By synthesizing heterogeneous data sources, including real-world weather and soil properties, the study demonstrates how modern AI can significantly enhance both the efficiency and sustainability of agricultural systems. To enable long-term adaptability, a continual reinforcement learning (CRL) framework is introduced, combining DRL with continual learning techniques such as Elastic Weight Consolidation and rehearsal to mitigate catastrophic forgetting across changing environments. Additionally, a human-centered, trust-aware optimization framework bridges the gap between algorithmic recommendations and farmers’ practical needs, promoting transparency, usability, and adoption. Collectively, these contributions advance precision agriculture by delivering adaptive, environmentally conscious, and trust-centered management strategies, aligning cutting-edge AI with the realities of farming communities.
Details
- Title: Subtitle
- Reinforcement learning-based agricultural management subject to climate variability
- Creators
- Zhaoan Wang
- Contributors
- Shaoping Xiao (Advisor)Rachel Vitali (Committee Member)Kishlay Jha (Committee Member)Marc Linderman (Committee Member)
- Resource Type
- Dissertation
- Degree Awarded
- Doctor of Philosophy (PhD), University of Iowa
- Degree in
- Mechanical Engineering
- Date degree season
- Spring 2026
- DOI
- 10.25820/etd.008478
- Publisher
- University of Iowa
- Number of pages
- xv, 172 pages
- Copyright
- Copyright 2025 Zhaoan Wang
- Language
- English
- Date submitted
- 03/27/2026
- Description illustrations
- Illustrations, graphs, charts, tables
- Description bibliographic
- Includes bibliographical references (pages 160-171).
- Public Abstract (ETD)
Agriculture is essential for feeding the world and supporting economies, but it faces growing challenges from climate change, limited farmland, and the need to produce more food for a rising population. This research explores how artificial intelligence (AI) can help farmers make better decisions in the face of uncertainty, such as unpredictable weather and environmental concerns like greenhouse gas emissions from soils. By combining real-world data on weather and soil with advanced AI techniques, this study develops new tools to improve crop management while reducing environmental impacts. Special attention is given to nitrous oxide (N2O), a powerful greenhouse gas released from soil, and how farming practices can be adapted to reduce its emissions. The research also introduces ways for AI systems to continually learn as conditions change, and it places strong emphasis on building trust with farmers by making recommendations that are practical, transparent, and easy to use. Together, these innovations aim to create more sustainable farming practices that protect the environment, support food security, and benefit farming communities worldwide.
- Academic Unit
- Mechanical Engineering
- Record Identifier
- 9985177377002771
Metrics
1 File views/ downloads
1 Record Views