Modeling drifting seaweed rafts that clog ports and beaches

Sargassum is a type of seaweed often found in large rafts floating on the surface of the ocean. It is so prolific that an entire region of the Atlantic Ocean, the Sargasso Sea, is named for it. However, Sargassum is not constrained to only these waters, and often floats into harbors and along beaches in the Caribbean and coastal South America. There, it can interfere with shipping and damage local ecosystems.

Francisco Beron-Vera and Gage Bonner employed two machine learning models to develop equations of motion for Sargassum rafts directly from collected data. Their goal was to explore the effectiveness of each model at inferring a physical law from data alone.

The two models the team investigated were a Long Short-Term Memory (LSTM) recurrent neural network and a Sparse Identification of Nonlinear Dynamics (SINDy) algorithm.

“LSTM is a popular deep-learning architecture for time series. It learns patterns directly from data by passing information through a chain of specialized units that can ‘remember’ or ‘forget’ past signals over time,” said Beron-Vera. “SINDy, on the other hand, is a data-driven modeling method that looks for a simple mathematical formula to explain the observed motion.”

The group fed each model a collection of data generated by another algorithm, called eBOMB, which can accurately produce simulations of Sargassum movements but is too complex to supply simple equations of motion.

The authors found that both methods performed well, although each method had its own advantages. While LSTM is a lighter model than SINDy that uses fewer neurons and layers, it functions more like a “black box” with results that are harder to interpret. SINDy, in contrast, provided explicit mathematical relationships from the eBOMB data.

The authors hope that with further improvements, these lightweight models will be able to provide forecasts to costal areas in advance of a Sargassum invasion, giving authorities more time to prepare a response.

Source: “Discovering the dynamics of Sargassum rafts’ centers of mass,” by F. J. Beron-Vera and G. Bonner, Chaos (2026). The article can be accessed at https://doi.org/10.1063/5.0292965 .