Integrated assessment of agricultural ecosystems using simulation-optimization and machine learning
Date
2018
Authors
Nguyen, Trung H., author
Paustian, Keith, advisor
Cotrufo, Francesca, committee member
Kelly, Eugene, committee member
Leisz, Stephen, committee member
Davies, Christian, committee member
Journal Title
Journal ISSN
Volume Title
Abstract
Agriculture provides many ecosystem services to human society but is also a major cause of environmental degradation. The key challenge of modern agricultural production is to meet projected increases in global demands for food, water, and energy in sustainable ways. Sustainable agricultural production requires integrated decision-support tools and rigorous assessment methods to improve the efficiency of natural resource management while minimizing its impacts to society and long-term ecosystem health. This dissertation focuses on developing methodology and modeling tools to support decision-making for sustainable agricultural resource management. The Millennium Ecosystem Assessment is used as a guiding framework for all the model development. The dissertation balances between the communication of the integrated assessment methodology and the presentation of the modeling techniques through four independent case studies. The first study links biogeochemical models with life cycle assessment (LCA) to explore the impact of regionally-specific ecosystem carbon stock changes associated with cassava cultivation for ethanol production in Vietnam. The second study couples biogeochemical models with GIS and optimization algorithms to conduct a high-resolution, spatially-explicit trade-off analysis of ecosystem services for irrigated corn production systems in the South Platte River Basin, Colorado, USA. The derived modeling platform is named the "Agricultural Ecosystem Service Optimization" (Ag-EcoSOpt). The third study integrates LCA into the Ag-EcoSOpt for a life-cycle-based optimization of feedstock landscape design for a hybrid corn grain- and stover-based ethanol production system at Front Range Energy biorefinery, Windsor, Colorado, USA. The last study develops a surrogate-based optimization framework for Ag-EcoSOpt to reduce the computational burden of large-scale landscape analyses. The study explores the trade-offs among seven management objectives of the irrigated corn production systems in Colorado, USA at different spatial scales.
Description
Rights Access
Subject
ecosystem services
machine learning
ecosystem modeling
optimization
life cycle assessment