Seeing the Sky: Progress Toward Routine All-Sky Camera Networks

In 2025 Wx Labs introduced an all-sky imaging and sensing platform intended to help fill a gap in routine meteorological observing. While operational networks measure core meteorological variables well, they rarely observe the visual sky state itself. Yet cloud amount, cloud evolution and overhead conditions matter for nowcasting, radiation studies, aviation, solar applications and local decision-making. With recent advances in low-power machine learning, camera-based observations are becoming more realistic as routine measurements. By developing scalable all-sky hardware designed specifically for meteorological applications, Wx Labs is addressing this observing opportunity.

Development over the past year has moved the device beyond a prototype camera system and towards a deployable observation node: optics and camera software have significantly improved image quality, particularly at the horizon; a sloped lid and dedicated internal heater have strengthened weather resilience and icing; custom, swappable electronics and over-the-air software updates have improved longevity and maintainability; and detailed logging with CF-compliant variables have improved interoperability. From initial engagement with the NMHS community, development has also begun on a low-power all-sky camera variant that can operate on 0.2 W to support off-grid national observing stations.

Testing and deployments have been undertaken across multiple environments, including Antarctica, a climatic chamber down to -50 °C, the Alps during TEAMx, and several UK campaigns. Applications have included academic researchers, UAS operators and solar farms. The talk will discuss these deployments and the applications they support, including cloud monitoring, local nowcasting, radiative studies, and augmentation of existing observing networks. Rather than treating sky cameras as ancillary devices, this talk argues that they can become a meaningful new observation layer in meteorology when paired with reliable engineering, usable metadata, and machine learning for automatic interpretation. The main challenge is no longer whether all-sky observations are interesting, but how to make them robust, standardised and interoperable enough to support routine scientific and operational use.