The site and the issue
Since 2018, an industrial site (the project site) situated on the coast has been the focus of ongoing Human Health and Ecological Risk Assessment (HHERA) investigation. The work to date has included contaminant mass flux assessment, ecological survey and habitat assessment, marine benthic macroinvertebrate survey, and surface water, porewater and sediment monitoring in the nearshore area both adjacent to the project site and at reference sites to assess ambient conditions.
These earlier assessments suggested a potential exposure pathway through the consumption of recreationally caught seafood, but only if people were actually fishing in the area. So, we set out to answer the question: does recreational fishing occur in the project site’s nearshore area?
We started with the basics, reviewing publicly available information and engaging directly with the Victorian Fisheries Authority (VFA) to build our understanding of fishing activity across this coastal area. We found that recreational fishing is common, particularly from January to May. A range of species are targeted, all with clear rules around catch limits and intertidal species. However, most available data focuses on boat-based fishing. There was limited insight into shore-based fishing and even less understanding of what proportion of recreational catch is consumed.
The takeaway: fishing occurs in the area, but whether it happens right next to the project site remained unclear, meaning the exposure pathway could neither be ruled in or out. Further information was required.
Using AI-enabled monitoring to close the data gap
To bridge the gap between regional trends and site-specific reality, we turned to AI-enabled video monitoring to provide real-world data to guide further HHERA decision-making. Our goals were simple. We needed to confirm whether fishing occurs near the project site and, if so, how often.
Cameras were installed along the project site boundary, streaming images to cloud storage for processing. We divided the camera field of view into georeferenced zones (nearshore, offshore, on-site, and horizon) based on existing environmental data, and used machine learning models to detect boats, people, vehicles, and activities within each zone. We then checked, refined and presented the outputs through an interactive dashboard with automated reporting.
We began by using a foundation model, processing two months of footage to establish a baseline for performance. From there, we manually reviewed and corrected the model’s detections to improve accuracy. Using this cleaned dataset, we then trained a custom model tailored to the project. The model was tuned for high recall, capturing as many potential events as possible, even if this resulted in some false positives. To address this, we staged detection events at the project site using field staff, which helped the model learn what to identify in a low-activity environment.
Quality assurance was embedded throughout every stage of the process. This included regular manual reviews of detection outputs, as well as monthly checks of dashboards to ensure ongoing performance and consistency. Automated reports were also cross validated to confirm reliability. Importantly, we tracked non-relevant detections, such as large ships on the horizon, to verify that the system remained active and functioning effectively, even during periods when little activity was occurring closer to shore.
The result:
So, what did we find after six months of monitoring? No recreational fishing activity was observed in the nearshore zone within 60 metres of the project site. Four fishing boats were recorded in offshore areas, while more than 800 large vessels were detected on the horizon.
The findings suggest that while the exposure pathway could theoretically exist, recreational fishing activity was not observed near the project site during the monitoring period.
Lessons learnt:
- Collaboration matters: Working with stakeholders like VFA helped build more context around the issue quickly.
- Innovation pays off: AI-based monitoring provided site-specific insights that traditional methods could not.
- Evidence strengthens decision-making: Real-world observations enable more confident, defensible decisions.
We acknowledge and appreciate our client’s openness to innovation and their willingness to explore the integration of emerging technologies, such as AI, within environmental assessments at the project site.
Authors:
Molly Lavery Senior Environmental Scientist | Anne Northway Principal Hydrogeologist |
Also contributing authors: Andras Nemes and Jacob Taylor.
This article was prepared as an abridged version of the presentation provided at the ALGA ecoforum Conference in October 2025.
Jacobs works with clients to turn contaminated sites into long-term assets by combining site investigation, risk assessment and environmental remediation design to support practical, defensible solutions. Our focus is on enabling safe land reuse by accelerating regeneration and delivering outcomes that stand up to changing regulatory requirements and community expectations.
At Jacobs, we’re challenging today to reinvent tomorrow – delivering outcomes and solutions for the world’s most complex challenges. With approximately $12 billion in annual revenue and a talent force of approximately 47,000, we provide end-to-end services in advanced manufacturing, cities & places, energy, environmental, life sciences, transportation and water. From advisory and consulting, feasibility, planning, design, program and lifecycle management, we’re creating a more connected and sustainable world. See how at jacobs.com and connect with us on LinkedIn, Instagram, X and Facebook.