Crowdsourced All-Sky Imagery for AI Nowcasting and Research

Accurate and timely observations remain fundamental to meteorological research and operational forecasting, yet many regions still rely on sparse or intermittent data. Whilst low-cost sensors are becoming increasingly abundant, their utility has been marred by lack of standardisation, calibration, and accuracy. This abstract introduces a concept for building a network of low-cost, crowdsource-able “all-sky” camera observation stations to address these coverage gaps and previous shortcomings. Each station would passively capture sky imagery at frequent intervals, analyse image features and upload the data to a central repository for use by the broader weather and climate community.

By integrating this imagery with existing meteorological datasets, it may be possible to resolve localised phenomena more precisely than current observing networks. Existing research suggests that ground-based visible images could enhance high-resolution numerical weather prediction (NWP) by offering near-real-time cloud fraction, cloud genus, and atmospheric motion vectors (AMVs) at unparalleled spatial granularity. Such data also hold promise for improving nowcasting systems designed to detect rapidly evolving weather features. Moreover, the resulting datasets would present novel opportunities for advanced AI/machine learning applications, such as training algorithms to classify cloud types, track convective development, monitor the climatic impact of contrails, or nowcast surface-level solar irradiance. The network could be designed such that over-the-air (OTA) software updates allow the ‘intelligence’ of the network in aggregate to improve over time or gain new abilities as they are discovered and developed.

While the station design and data-processing pipeline remain under early development, this presentation aims to share the concept, identify potential applications, and gather community input on technical and scientific challenges. In particular, the project welcomes feedback on topics such as image standardisation, data validation, and deployment strategies. Interested colleagues are encouraged to discuss potential collaborations or sign up for project updates, including an early-access waitlist for the first stations. By co-developing these capabilities with the meteorological community, we can create a robust, accessible network that extends and complements existing observational infrastructure—ultimately enhancing our collective ability to monitor and understand the atmosphere.