Bayesian Optimization for Anything: An Open-Source Framework for Accessible, User-Friendly Bayesian Optimization

The Science

A team of researchers developed Bayesian Optimization for Anything (BOA), a new high-level Bayesian optimization model wrapping toolkit addressing common barriers in implementing Bayesian optimization (BO). BOA is language-agnostic and can interface with models written at any coding language.

The Impact

Numerical models play an indispensable role in environmental science. Models such as Earth system models, land surface models, ecosystem models, hydrological models, and watershed models are crucial for understanding and predicting complex environmental processes. Despite significant advancements in model development and the inclusion of increasingly complex processes, these models remain approximations of the systems they represent and inherently require parameterization.

Given the complexity and potential computational expense associated with these models, there have been concerted efforts within the scientific community to develop and refine techniques for parameterization, such as BO. A team of researchers aimed to bridge the gap between nondomain experts and BO by introducing BOA.

Summary

BOA, a high-level BO framework and model wrapping toolkit, presents a novel approach to simplifying BO with the goal of making it more accessible and user-friendly, particularly for those with limited expertise in the field. BOA addresses common barriers in implementing BO, focusing on increasing ease of use, reducing the need for deep domain knowledge, and cutting down on extensive coding requirements. A notable feature of BOA is its language-agnostic architecture. BOA’s features enhance its applicability, allowing for broader application in various fields and to a wider audience.

The study showcases BOA’s application through three examples: a high-dimensional optimization with 184 parameters of the Soil and Water Assessment Tool (SWAT+) watershed model, a highly parallelized optimization of this intrinsically nonparallel model, and a multiobjective optimization of the Finite-difference Ecosystem-scale Tree Crown Hydrodynamics (FETCH) model. These test cases illustrate BOA’s effectiveness in addressing complex optimization challenges in diverse scenarios.

Principal Investigator

Gil Bohrer
The Ohio State University
[email protected]

Program Manager

Daniel Stover
U.S. Department of Energy, Biological and Environmental Research (SC-33)
Environmental System Science
[email protected]

Funding

The work was funded in part by National Science Foundation award 2036982, U.S. Department of Energy awards DE-SC0023084 and DE-SC0021067, U.S.–Israel Binational Agricultural Research and Development Fund (BARD) award IS-5304, and National Oceanic and Atmospheric Administration award NA18NOS42000079 (Davidson Fellowship, OWC-NERR ODNR Subaward N18B 315-11).

References