EventsThe 2nd International Online Conference on Marine Science and Engineering
Published
This submission belongs to the session S1. Ocean Engineering of the event The 2nd International Online Conference on Marine Science and Engineering
Published date
18 Nov, 2026
Academic Editor
author-avatarWeicheng Cui
Citation
Manuele Zanaboni, Riccardo Mandolini, Kim Bender, Carlos Santini, Camilo Velez, Mahendra Samarawickrama, A Trustworthy Edge-to-Cloud Sensor Network for Ocean Observation with Human-in-the-Loop Assurance, in Proceedings of The 2nd International Online Conference on Marine Science and Engineering, 23 November–25 November 2026, MDPI: Basel, Switzerland
Share
Email
Facebook
Twitter
LinkedIn

A Trustworthy Edge-to-Cloud Sensor Network for Ocean Observation with Human-in-the-Loop Assurance

Manuele Zanaboni 1
Kim Bender 3
Carlos Santini 4
Camilo Velez 4
Mahendra Samarawickrama 5
1. Technology, Conservation and Machine Learning, Project OceanOasis Ltd, Kelso, QLD, 4815, Australia
2. Technology, AI and Machine Learning, Project OceanOasis Ltd, Kelso, QLD, 4815, Australia
3. IoT and Engineering, IOTIS Pty Ltd, Smithfield, QLD, 4878, Australia
4. Assembly Cloud Solutions, Brisbane, Australia
5. Sydney Program, Boston University (USA), Sydney, NSW, 2008, Australia
Abstract

Coral reefs and coastal ecosystems are declining faster than conventional monitoring can track. Diver surveys, moored cameras and episodic sampling remain costly and intermittent, while streaming raw underwater video to cloud infrastructure is prohibitively bandwidth and power intensive at ocean scale. We present a reference architecture for scalable, trustworthy ocean observation that relocates analysis to the edge whilst embedding human oversight as a core design principle. The system integrates three layers. The edge layer runs on-device AI inference for species and event detection on low-power passive-acoustic and stereo-vision sensors, transmitting only structured detections and validation imagery over satellite or cellular links and reducing transmitted volume by orders of magnitude. The human-assurance layer routes low-confidence detections to qualified domain experts for rapid, auditable review. Adjudications are logged to create an accountability trail and fed back to retrain models, so human oversight functions as a load-bearing design element rather than post-hoc validation. The cloud analytics layer aggregates vetted findings into biodiversity and environmental-pressure indicators. We demonstrate the architecture through two instantiations, an acoustic node and a reef-vision node, validated end to end under bench and simulated-fleet conditions. The acoustic blast-detection model achieves an F1 of 0.955 and ROC AUC of 0.994 on a held-out test set, with in-water trials in preparation.

Keywords
Ocean observation systems
edge AI
passive acoustic monitoring
human-in-the-loop assurance
responsible AI
A Combined CFD-FEA Approach Using Open Water Simulation to Enhance the Structural Efficiency of A 3D-Printed PLA Propeller Reinforced with Carbon Epoxy Surface Coatings
The Emerging Arctic Shipping Routes and Their Influence on Chinese Ports