Robust learning algorithms for capturing oceanic dynamics and transport of Noctiluca blooms using linear dynamical...

Area of Science:

• Marine Biology
• Oceanography
• Machine Learning

Background:

• Intensifying Noctiluca blooms in the Gulf of Oman and Arabian Sea pose a significant threat to regional fisheries.
• These blooms impact the ecosystem health of a region supporting nearly 120 million people.
• Blooms occur annually during the winter monsoon, necessitating a deeper understanding of their dynamics.

Purpose of the Study:

• To investigate the onset and patterns of Noctiluca blooms using local-scale data analysis.
• To identify causal factors and latent dynamics driving these blooms.
• To test macroscopic oceanographic hypotheses and uncover local trajectory-scale dynamics.

Main Methods:

• Combined physical and biological oceanography with machine learning techniques.
• Developed and applied a novel algorithm, the variable-length Linear Dynamic Systems (vLDS) model.
• Utilized vLDS to analyze irregular surface velocity drifter datasets with variable-length time series.

Main Results:

• The vLDS model effectively extracts causal factors and latent dynamics from local drifter trajectories.
• Generated predictive plots and tested macroscopic scientific hypotheses regarding Noctiluca blooms.
• Provided local-scale statistical evidence supporting and refining existing oceanographic hypotheses.

Conclusions:

• The vLDS model offers a robust method for analyzing complex, irregular oceanographic datasets.
• Identified local trajectory-scale dynamics influencing Noctiluca blooms, often missed at macroscopic scales.
• Demonstrated the generalization capability of vLDS for studying biogeochemical processes and population-level dynamics in marine environments.