AI Serves a Dual Purpose: Forecasting Marine Health and Safeguarding Oceanic Data

The ocean serves as one of the planet’s most formidable natural systems for carbon sequestration. Among the most potent contributors to this process are blue carbon ecosystems—coastal habitats like mangrove forests, seagrass meadows, and salt marshes. These verdant environments effectively capture and store significant amounts of carbon through plant growth and sediment accumulation. However, unlike many terrestrial ecosystems that release carbon back into the atmosphere, a considerable amount of the carbon sequestered by our oceans remains trapped in underwater soils for centuries or longer. Protecting and restoring these ecosystems is crucial for enhancing the ocean’s role in climate regulation and for Canada’s goal of reaching net-zero carbon emissions by 2050.

With support from the National Research Council of Canada’s (NRC) Ocean program, a University of Victoria research team has joined forces with the NRC to monitor ocean health and its capacity for carbon sequestration using innovative sensors, machine learning, and a remotely operated vehicle (ROV). The project also incorporates a cybersecurity aspect to evaluate vulnerabilities in the ROV’s sensor network to ensure data privacy.

Dr. Mohammad Mamun, a research officer at the NRC, notes that the project “utilizes interdisciplinary expertise in artificial intelligence to develop models that integrate data confidentiality, cybersecurity, and underwater systems, creating a comprehensive framework for monitoring blue carbon ecosystems and furthering ocean health intelligence.” At its essence, it employs a transformer-based machine learning model pre-trained on publicly available historical data from Ocean Networks Canada to predict chlorophyll concentrations—a vital indicator of phytoplankton activity and carbon sequestration potential. Elevated chlorophyll levels signify a higher potential for carbon absorption. This model aids researchers in evaluating the health of marine ecosystems and their carbon uptake capabilities.

The remarkable aspect of this project lies in how these models are constructed through federated learning, a method for training AI models without transferring all the data to a central hub. This distinction is significant, as not all organizations are keen on sharing their data; however, many are open to utilizing it for AI training as long as it remains protected. In this method, instead of sending data to a centralized AI system, the AI model travels to the devices where the data resides—here, a set of sensors. The model learns directly from the local data. Consequently, only the insights gained are transmitted back, preserving the raw data’s privacy while enabling continual AI improvement.

This approach becomes particularly advantageous in intricate environments like underwater systems. While many are familiar with the Internet of Things (IoT), the Internet of Underwater Things (IoUT) poses added complexity. In this aquatic context, the sensor suite itself contains the predictive AI model, comprising both hardware and software. These sensors link to a ROV, which communicates with the cloud. Every additional connection point—from sensor to ROV and ROV to cloud—heightens the risk of cyber attacks, making the privacy-preserving aspect of federated learning not merely beneficial, but essential.

“Our goal was to understand ocean health, which is intricately connected to carbon sequestration—a process most efficiently executed by a healthy ocean,” emphasizes Dr. Navneet Kaur Popli, an associate teaching professor of electrical and computer engineering at the University of Victoria. She played a crucial role in developing the sensor suite, the ROV that housed the sensors, and the AI algorithms employed for federated learning.

Ocean-going drones for predicting changes in ocean health

The researchers utilized publicly available data from Ocean Networks Canada collected at four key regions in the Salish Sea, a marginal sea of the Pacific Ocean situated in British Columbia. By leveraging this data, researchers implemented machine learning techniques to forecast chlorophyll levels. Subsequently, by examining patterns in environmental variables such as rising water temperatures, ocean acidification, and increasing salinity, the team devised models capable of accurately predicting changes in chlorophyll concentration—and thus, ecosystem productivity—30 days, one year, and two years into the future. These predictions are vital for comprehending how ecosystems like mangrove forests, seagrass meadows, and salt marshes in coastal British Columbia may react to climate change and how their carbon sequestration capabilities might evolve over time.

Access to this information can offer a competitive edge for certain industries, identifying optimal locations for seaweed aquaculture or eco-tourism. By employing federated learning, these sensors help maintain data privacy, which is crucial for companies seeking to protect their information from competitors.

Cybersecure ocean monitoring

The team’s efforts also focus on implementing machine learning in cybersecurity to formulate better techniques for detecting potential cyber attacks on ocean monitoring systems.

Attackers can target websites, the cloud, drones, and wired or wireless communication networks. Dr. Popli explains, if an adversary tampers with or obstructs values from the sensors while proper cybersecurity measures are lacking, the data might still appear normal—even when alarm bells should be ringing. “If an application lacks security, it becomes vulnerable to attacks, and its predictions are invalid. Without security, I can’t trust whether the data represents reality.”

To ensure the integrity and security of underwater data collection, a Federated Learning-based Intrusion Detection System—originally designed for Internet of Underwater Things networks—is employed to facilitate distributed, privacy-preserving anomaly detection. This protects sensor data, which is crucial for accurate carbon accounting. Additionally, an AI-enabled simulation framework, adapted from autonomous ships (known as Maritime Autonomous Surface Ships, or MASS), is utilized to model cyber attack scenarios targeting automatic identification system protocols. This framework allows for monitoring the resilience of the sensing and communication infrastructure essential for blue carbon monitoring.

By incorporating all of these technologies into the project, we are advancing the scientific understanding of blue carbon processes, fostering the blue economy, and contributing to the establishment of secure, scalable, and intelligent ocean monitoring systems. These systems aid in climate change mitigation, enhance the sustainable management of marine resources, and support evidence-based policy development.

Fostering research and skill development

This project not only has commercial applications and furthers the NRC’s research objectives, but it also serves as an excellent learning opportunity for students interested in careers in ocean-related fields where data privacy is paramount. Furthermore, this research can benefit future generations. “By training highly qualified personnel and equipping a future workforce with the knowledge and confidence to navigate the underwater cyber realm, we are also paving the way to ensure governments and industry have access to reliable and essential data sets—especially as we transition to a carbon economy,” states Dr. Mamun.

To date, the sensor suite and ROV have undergone testing at the University of Victoria’s laboratory. For the next phase, scheduled for completion by the end of 2025, the team intends to deploy the ROV in Burrard Inlet within the Salish Sea.