Interpretable machine learning for shoreline forecasting.

Area of Science:

• Environmental Science
• Geophysics
• Computational Science

Background:

• Machine learning (ML) has advanced scientific modeling but often lacks interpretability in physics-based applications.
• Traditional physics-based models may not generalize across diverse environments, limiting their applicability.
• Symbolic regression (SR) provides transparent, interpretable mathematical expressions, suitable for uncovering physical principles.

Purpose of the Study:

• To demonstrate the application of symbolic regression for physical modeling in environmental sciences.
• To develop interpretable models for shoreline prediction using observational data.
• To uncover region-specific physical drivers of coastal evolution.

Main Methods:

• Evolved a population of interpretable models using symbolic regression.
• Optimized models for both predictive accuracy and complexity.
• Utilized global observational data for shoreline change analysis.

Main Results:

• Successfully applied symbolic regression to physical modeling for shoreline prediction.
• Discovered region-specific mathematical formulations representing dominant physical drivers.
• Generated interpretable models that align with physical intuition.

Conclusions:

• Symbolic regression enables data-driven discovery in physics-based domains while maintaining interpretability.
• This approach offers new insights into the physical dynamics of shoreline change across scales.
• The methodology supports understanding coastal evolution influenced by climate change and human activities.