Scientific knowledge advances through the interplay of empiricism and theory. Empirical observations of environmental dynamics provide insights into system behavior, while theory organizes these insights into conceptual frameworks that guide understanding and prediction. The rise of powerful machine learning (ML) tools—particularly deep learning and generative AI—has enabled the extraction of complex patterns from large datasets, leading some to question the continued relevance of theory in scientific discovery.

However, a growing body of knowledge-guided machine learning (KGML) research demonstrates that integrating theory with data-driven approaches enhances both predictive performance and scientific understanding. KGML approaches can improve parameter estimation, identify knowledge gaps, and enable more reliable predictions in unseen or data-scarce conditions.

This Research Topic invites contributions that explore how scientific knowledge can be embedded into ML models to improve outcomes in water resources. We welcome submissions that address, but are not limited to:

• The use of process-aware architectures or custom loss functions that enforce physical or hydrologic principles.

• Differentiable modeling or ML-based parameterization of process-based models.

• Hybrid approaches that combine process-based and ML components.

• Techniques for interrogating ML models to ensure physical realism or extract insights into hydrologic processes.

• Methods for balancing confidence in data with confidence in scientific knowledge.

• Knowledge-guided multi-source data integration into ML models, including remote sensing, in-situ observations, and outputs from process-based models.

We particularly encourage submissions that demonstrate how KGML can bridge the gap between predictive accuracy and process understanding, enabling both robust predictions and scientific discovery.

Are you interested in contributing a manuscript to this Research Topic?

Improving Machine Learning for Water Resources Through Scientific Knowledge Integration