Work

Surrogate-assisted Bayesian optimisation for multi-objective parameter estimation in process-based agroecosystem models, with application to DNDCv.CAN. A Manitoba case study.

Year: 2026
Type: article
Abstract: This data repository contains all inputs and extended results of DNDCv.CAN model calibration study. This case study is a part of our manuscript titled: Surrogate-assisted Bayesian optimisation for multi-objective parameter estimation in process-based agroecosystem models, with application to DNDCv.CAN. Abstract: Process-based agroecosystem models such as DNDCv.CAN support carbon accounting and agronomic decisions, but their parameters must be calibrated before use. Manual trial-and-error calibration is slow, and calibrating against one output can degrade the fit of others. This article presents a surrogate-assisted Bayesian optimization framework for DNDCv.CAN, implemented in Python with a graphical interface so that calibration runs without scripting. It couples a Gaussian process surrogate with an Expected Improvement acquisition function to estimate parameters with few model evaluations. It calibrates several targets at once through a normalized weighted objective, and it ties crop-level parameters across all occurrences of a crop. We demonstrate the framework at a Manitoba crop rotation site by calibrating annual yield and daily net ecosystem exchange together. Yield fitted well, while net ecosystem exchange was weakly constrained by the crop parameters used. Calibrating on one target reduced the fit of the other, and the framework quantified this trade-off directly. The approach estimates a best-fitting parameter set rather than a full posterior, and its adapter design allows the same workflow to be applied to other command-line models.
Language: English
Open Access status: green
OpenAlex ID: W7204240674
FWCI:
Cites:
Cited by:
Related to: 10