Adaptive sampling in the Philippine Sea using autonomous profiling floats and FloatCast

Area of Science:

• Oceanography
• Robotics
• Machine Learning

Background:

• Ocean-profiling floats are crucial for studying oceanographic processes.
• Current floats lack precise horizontal positioning, limiting data relevance.
• Deploying floats is costly and logistically complex.

Purpose of the Study:

• To present the FloatCast framework for actively managing the drift of ocean-profiling floats.
• To enhance the sampling effectiveness of float arrays through intelligent control.
• To improve the quality of scientific data products from autonomous oceanographic instruments.

Main Methods:

• Utilizing vertical depth control to manipulate float horizontal position.
• Employing an echo state network for ocean surface flow forecasting.
• Combining machine learning, optimization, and feedback control for dive command selection.
• Ranking candidate dive commands using a mapping error scoring metric.

Main Results:

• FloatCast demonstrated promising results in both simulation and real-time experiments in the Philippine Sea.
• The framework successfully predicted and influenced float trajectories.
• The control framework showed potential for increasing sampling effectiveness.

Conclusions:

• FloatCast offers a viable method for improving the spatial sampling of free-drifting floats.
• Active drift management via vertical depth control can enhance oceanographic data collection.
• The framework has the potential to improve scientific data products despite inherent underactuation.