About This Project

Museum bird specimens form an unintentional biosensing network: plumage colour has recorded pollution and land-use change for centuries, but existing tools are too coarse to read it. We use hyperspectral imaging and unsupervised "colour cloud" clustering to detect where, and how strongly, plumage has been disrupted by environmental change, an open dataset and method for reading disruption in any naturally coloured organism, birds to corals.

Ask the Scientists

What is the context of this research?

The project addresses the current limitations in reading biological signals from natural environments. Historically, biological monitoring from a distance required engineering organisms to emit detectable signals. However, millions of museum bird specimens already naturally record environmental conditions, such as historical black carbon pollution, within their plumage. While prior studies have extracted colour data from these collections, they relied on point measurements, patch averages, or predefined colour spaces (like RGB). These existing methods discard fine-scale spatial details, lose spatial heterogeneity, and introduce observer bias by locking the analysis into a single perspective. This project is contextualized by the recent affordability of high-resolution hyperspectral cameras and the maturation of robust unsupervised clustering algorithms in Python and R, making this new approach highly feasible.

What is the significance of this project?

The significance of this research spans scientific, practical, and institutional domains. Institutionally, it reveals that existing natural history collections latently serve as massive, distributed environmental sensor networks, extending biological monitoring backward in time and geography without requiring new fieldwork. Practically, it provides policymakers and conservationists with a low-cost, retrospective method to track historical anthropogenic pressures—such as pollution or land-use changes—complementing expensive, modern monitoring infrastructure. Scientifically, the project delivers a validated, general-purpose method that is not limited to birds. The clustering approach can be transferred to other non-engineered systems whose colouration carries an environmental signal, such as corals, algae, and biofilms.

What are the goals of the project?

The primary goal of this project is to build an instrument capable of detecting environmental disruption in naturally coloured organisms at scale. To achieve this, the first objective is to procure a hyperspectral camera and image museum bird specimens to build a foundational, pixel-level reflectance dataset across a known environmental gradient. Following this, the project will develop and benchmark an unsupervised “colour cloud” clustering algorithm to objectively detect and localise anomalous colour variations on the specimens' bodies without a priori biases. These cluster boundaries will then be projected through established visual models to test whether the environmental disruptions represent measurable and perceptible signal shifts to relevant receivers. Ultimately, the project aims to release a publicly available dataset and a transferable, open-source workflow for the wider scientific community.