Files in this item



application/pdfPETERSON-THESIS-2021.pdf (1MB)
(no description provided)PDF


Title:Quantifying the agronomic and water quality tradeoffs of fertilizer management with multiobjective evolutionary algorithms
Author(s):Peterson, Chelsea Marie
Advisor(s):Rodriguez, Luis F
Contributor(s):Bhattarai, Rabin; Sowers, Richard
Department / Program:Engineering Administration
Discipline:Agricultural & Biological Engr
Degree Granting Institution:University of Illinois at Urbana-Champaign
Subject(s):fertilizer, agriculture, multiobjective, evolutionary, algorithms, RZWQM2
Abstract:When combined with agricultural systems models, multiobjective evolutionary algorithms (MOEAs) can efficiently quantify optimal tradeoffs (Pareto fronts) among conflicting economic and environmental objectives to help farmers identify compromise decision alternatives. Realizing the potential of MOEAs to agricultural decision support, however, will require corroborating that approximate solutions accurately represent system tradeoffs through extensive algorithm comparisons and integrating model uncertainty analysis into the optimization workflow. This thesis contributes to these broader goals by comparing the ability of five MOEAs (e-MOEA, e-NSGA-II, OMOPSO, GDE3, and MOEA/D) to calibrate the USDA's Root Zone Water Quality Model (RZWQM2) and identify fertilizer management decisions that maximize profits and corn yields while minimizing nitrate loads for two corn-soybean production sites in east-central Illinois. For both test problems, I evaluated solution accuracy by tracking three performance metrics for all five MOEAs along with their contributions to the best-known Pareto front throughout the optimization run. After pooling the calibration results to assess model error tradeoffs, I applied the top-performing MOEA with sets of parameter vectors from the calibration Pareto fronts to stochastically optimize fertilizer rate, method, and timing decisions. e-MOEA stood out as the most effective algorithm for model calibration by achieving the highest performance metric values within the fewest generations. The accuracy differences among MOEAs were much less apparent for the fertilizer management problem, implying that the algorithm selection should not influence the optimization results. Nonetheless, I applied OMOPSO for the stochastic optimization because it achieved the highest metric values by a small margin. Despite wide uncertainty bounds and high maximum pro fit nitrogen rates, the optimization results support that sidedress fertilizer applications from 20 to 50 days after planting not only maximize profit for injection and surface broadcast but also reduce outcome sensitivity to the placement method and offer the best compromise between pro fit and drainage nitrate loads. Although testing different calibration parameters could further support these results and re fine the uncertainty bounds, this analysis provides a flexible simulation-optimization framework for stochastically optimizing best management practice selection, design, and placement at field to watershed scales.
Issue Date:2021-07-23
Rights Information:Copyright 2021 Chelsea Peterson
Date Available in IDEALS:2022-01-12
Date Deposited:2021-08

This item appears in the following Collection(s)

Item Statistics