The urgent need to better understand the impacts of climate change has highlighted the importance of interdisciplinary data resources that connect environmental science with creative practice. Existing sound archives often lack regional specificity and the environmental metadata necessary for climate-oriented modelling. AudioWeather addresses this gap; a bespoke dataset of Australian birdsong-focused soundscapes, annotated with geolocation, time, and detailed climate conditions. Developed as part of a practice-led research project in sonic arts, this dataset supports both machine learning experimentation and public engagement.

Comprising over 44,000 field recordings, AudioWeather adds novel environmental metadata into each audio entry. Originally developed to train the EAGLE generative audio model for the interactive work Synthetic Ornithology, the dataset also forms the basis of a public-facing interactive archive. This digital platform enables users to explore Australian birdsong recordings by filtering them by temporal, geographic, and environmental criteria, fostering engagement with the sonic dimensions of ecological systems and climate change.

Beyond its technical role in supporting generative audio research, AudioWeather offers a methodological contribution to the fields of soundscape ecology, sonic art, and environmental machine learning. While full dataset release is restricted due to licensing, the interactive archive, accessible at audioweather.com, presents the dataset in an interactive environment. The archive acts as a bridge between scientific data and reflection, inviting audiences to engage with climate change not through data or statistics, but through attentive listening and personal interpretation.