Distributed Cooperative Estimation and Modeling of Fields with Applications for Underwater Sensor Networks
Author(s)
Mayberry, Scott Timothy
Advisor(s)
Editor(s)
Collections
Supplementary to:
Permanent Link
Abstract
Environmental phenomena, such as algal blooms, wildfire propagation, pollutant dispersion, and more, evolve as continuous spatiotemporal fields that demand real-time monitoring for effective management, rapid response, and/or policymaking. While modern sensing and modeling frameworks have improved environmental monitoring through satellite imagery, sensor networks, and deep learning, underwater environments present unique challenges due to severe communication constraints, costly infrastructure, and the absence of shared research platforms. This dissertation addresses these challenges by co-designing accessible aquatic hardware infrastructure with distributed estimation algorithms for real-time monitoring of spatiotemporal fields.
A primary contribution of this work is the development of μNet, an open-source aquatic testbed that integrates custom acoustic and optical modems, particle-filter-based localization, and miniature underwater robots to enable decentralized sensing experiments. Building on this platform, we develop a comprehensive suite of distributed cooperative filters that transition from relying on known field structures to accommodating unknown field structures. First, we embed known Poisson and advection-diffusion constraints directly into a filter using finite-volume discretization, proving convergence and demonstrating adaptive gradient-based formation control. Next, we extend to unknown parameters through a cascaded architecture that decouples field estimation from online parameter identification of diffusion and flow coefficients while maintaining a distributed implementation. We then generalize the estimation process by augmenting states with arbitrary-order spatial and temporal derivatives, employing extended Kalman filtering to discover governing dynamics from data, and constraining the final estimates by the discovered model. Finally, we enhance the prediction step by embedding a stochastic discontinuous Galerkin solver using polynomial chaos expansion, yielding uncertainty-aware, physics-informed forecasts even during active model learning.
This dissertation bridges the gap between theoretical advances in distributed estimation and practical deployment in resource-constrained aquatic environments. By providing both an accessible hardware platform and a modular suite of algorithms that scale from physics-constrained to data-driven estimation, this work enables accurate, physically consistent field reconstruction under realistic communication and computational limits. The open-source nature of all components—from circuit schematics to algorithm implementations—lowers barriers to entry for underwater research and fosters reproducible, community-driven innovation in environmental monitoring.
Sponsor
Date
2025-07-22
Extent
Resource Type
Text
Resource Subtype
Dissertation