Environmental Machine Learning, Baseline Reporting, and Comprehensive Evaluation: The EMBRACE Checklist

Abstract

M achine learning (ML) is transforming environmental research with its powerful functionality and broad applicability.The number of papers in high impact environmental journals using ML is growing exponentially. 1 Recent innovative environmental applications ranging from forecasting beach water quality, 2 designing polymer for resource recovery, 3 to modeling rainfall-induced landslide susceptibility. 4 Given the complexity of ML and the various assumptions and data manipulations often required, it is essential to follow best practices for reporting methods and results from ML applications. [5]6][7] Our recent critical review 5 highlighted that many environmental ML papers could enhance their impact and clarity by adhering more closely to these best practices.In the review of closely examining 148 highly cited environmental ML papers, we found only 24%, 48%, 37%, and 26% of the studies clearly reported their methods for missing data management, feature selection, feature scaling, and hyperparameter optimization, respectively.After publishing the review, 5 we received numerous requests for easy-to-use guidelines to assistant authors, reviewers, and editors to better conceptualize, prepare, conduct, and evaluate ML research.In response, here we introduce the Environmental Machine-learning, Baseline Reporting, And Comprehensive Evaluation (EMBRACE) Checklist in this Viewpoint.The EMBRACE Checklist, along with accompanying instructions in the Supporting Information and a newly constructed GitHub repository, 8 are designed to help researchers in maximizing the comprehensive assessment and impact of their ML research while adhering to best practices for reporting.By using the checklist, researchers can ensure that they provide important methodological details, identify critical problems, implement robust and explainable models, and improve the overall quality and impact of their ML research.We encourage researchers using ML methods for environ-

Topics & Concepts