Adaptive Sampling of River Plume Fronts: Integrating Statistical Modeling and Autonomous Path Planning for Enhanced Oceanographic Exploration

Abstract

The ocean remains largely unexplored and presents a great challenge for scientific re­search. Ocean fronts have shown importance for understanding both biological and physical oceanographic phenomena, with river plume fronts being particularly intrigu­ing. These fronts are a complex combination of freshwater and oceanic systems and provide a unique perspective on the dynamics of frontal systems. Despite the acknowledged significance of ocean fronts, their comprehensive sam­pling is still a goal not yet achieved. To address this issue, this study tries to develop an approach that utilizes statistical techniques, robotics, and oceanographic knowledge to create an intelligent and ASS tailored to investigate river plume fronts or other similar frontal systems. This system is designed to address the spatial and temporal complexi­ties of ocean fronts, allowing for more efficient and representative sampling. The basis of our approach is the incorporation of Gaussian random fields. This modeling technique provides a robust proxy to the intricate field dynamics, allowing us to capture underlying patterns and forecast evolving behaviors. This surrogate mod­eling approach enables a nuanced understanding of the river plume front, reducing the computational burden associated with real-time decision-making, this means that it can be conducted on-board a robotic agent such as an autonomous underwater vehicle. Building upon this proxy model, we develop and implement both myopic and non­myopic path planning algorithms. Myopic planning guides autonomous agents in im­mediate decision-making, capitalizing on localized data to direct sampling efforts. In contrast, non-myopic planning offers a broader perspective, considering the entirety of the available information to optimize sampling across the entire field. The harmoniza­tion of these algorithms presents an unprecedented opportunity to balance immediate responsiveness with long-term strategic sampling. We engage in a rigorous evaluation of our system through a series of simulation studies, modeling different scenarios and conditions that represent real-world com­plexities. Furthermore, we conduct experimental validations in authentic marine en­vironments, leveraging our system's adaptability to capture the transient and spatially heterogeneous nature of river plume fronts. Autonomous underwater vehicles are used in our field deployments. The results of our investigation demonstrate the robustness and versatility of our approach and its potential to improve oceanographic sampling. By seamlessly weav­ing together statistical modeling and strategic path planning, we present an adaptive sampling framework for exploring our ocean more efficiently and intelligently. This work showcases the possibility of adaptive ocean sampling and contributes to our un­derstanding of river plume fronts.