Shoreline memory: under the reserve

Question

Asbestos-containing material turned up in a bay in New Zealand. Over several months of field visits, lab work and conversations with the residents affected, it became clear the public record told a different story from the ground.

Environmental health records note what was reported. They miss what was never tested. This project asks how large that gap is, and whether a model can point to where hidden risk is most likely before anyone digs.

Approach

A building-level risk model. Each building near the shoreline gets a risk score from what public records already hold: construction era, building type, land use history and distance to confirmed finds. Field and lab results serve as ground truth. The scores are mapped in GIS so the gap between recorded and likely risk is visible street by street.

A blind test for AI models. Language models are increasingly asked about places they have never seen. I give a model only a coordinate and one question, risk or not at risk, then score its answers against the field data. The test shows what models assume about a place, and where those assumptions would leave people unprotected.

Anything that could identify a resident or a private property is generalized before it reaches the map.

Why it matters

Asbestos is still turning up in schools, museums, beaches and older homes across New Zealand and Australia, and families are often unsure what has been checked. A risk map built from the records that already exist can tell councils and communities where to look first.

The same fieldwork became Asbestos: The Fiber Clue, a children's book that teaches readers to notice the materials around them, document hazards that aren't obvious, and report what they find.

Links