A Tutorial on Best Practices and Pitfalls in Applying Machine Learning to Environmental Research

Machine learning (ML) has become a powerful paradigm for extracting structures from complex environmental data and supporting scientific inference across diverse subfields. Its potential, however, is often limited by gaps in the appropriate application of domain knowledge to machine-learning workflows, variability in data quality, and methodological choices that can distort model behavior or its interpretation. This Tutorial provides practical guidance on how domain expertise can be effectively integrated into the design of environmentally meaningful machine learning models and outlines a coherent workflow that integrates crucial stages, including data preprocessing, model development, evaluation, and interpretability. It also examines recurring pitfalls that arise along this pipeline and explains how they shape the credibility and reliability of machine-learning findings in environmental contexts. By consolidating these principles, this Tutorial aims to provide researchers with a clearer foundation for using machine learning in ways that are scientifically grounded, methodologically rigorous, and better aligned with the needs of environmental decision-making.