Data-Driven Study of the Probabilistic Characteristics of Wind Waves in Latvia

Wind waves are inherently irregular and random, making the goal of finding a fully deterministic description practically impossible. However, knowing their probabilistic properties is crucial for engineering applications and for understanding ocean dynamics. To deepen this understanding and build more efficient wind-wave models, machine-learning approaches are likely to become increasingly valuable.  Recent progress in physics-informed machine learning (PIML) has transformed fluid mechanics by combining data-driven approaches with physical fundamental equations, enabling more robust and generalizable models.

In our study, we apply PIML techniques to identify probabilistic characteristics of wind waves. Our research is based on learning probability distributions directly from data, which allows us to avoid restrictive assumptions or classical approximations.

We utilize field measurements collected in Skulte, Latvia, during August–September 2022. The dataset includes pressure time series and 3D velocity profiles, providing a detailed description of wave dynamics. Building upon existing PIML architectures, we developed a framework capable of inferring an accurate and efficient probabilistic model of wind waves. Preliminary results show promising agreement with theoretical expectations and previous studies.

The dataset was provided by Kevin Parnell and colleagues from Tallinn University of Technology (TalTech), together with the Latvian Institute of Aquatic Ecology. Our findings highlight the potential of PIML for improving probabilistic wave modelling and set the foundation for future applications in coastal engineering and environmental monitoring.