Intelligent All-sky Cameras for Dense Mesoscale Observations: From Field Trials to Operational Pilots

Meteorological calls on the mesoscale often hinge on change detection such as towering cumulus building, a gap opening for solar energy, or the first smoke plume rising under a convective cell. One of the most cost-effective datasets for detection of these phenomena is real-time, on-site imagery. The meteorological community has built remarkable capability with reliable, trustworthy surface/PBL observations spanning decades, and some network owners have recently begun to include webcam imagery at their sites as well, discovering many tangible benefits. However, for full situational awareness at a given location, the best imaging device is one that can capture the sky in all directions seamlessly.

This talk will introduce the WxStation, a new all-sky observing system currently undergoing extensive field testing within academic projects. The device combines professional grade in‑situ sensors with a 180° all‑sky camera to improve situational awareness. The addition of all-sky imagery into meteorological workflows can raise the effective resolution of mesoscale monitoring, particularly for cloud‑driven variability that impacts nowcasting, convective initiation, solar ramps, aviation ceiling/visibility, and wildfire situational awareness. A short image sequence quickly conveys a wealth of information to a trained observer—often faster and more intuitively than numbers alone can—making on-site imagery a powerful complement to conventional variables.

Furthermore, low‑power inference compute within the WxStation can automatically derive real‑time meteorological products (e.g., cloud fraction/type/motion) for assimilation into NWP and for wider analysis/climatological studies. In the long-term, these devices could provide insights equivalent to that of having a trained meteorological observer at every site 24/7/365. Two factors influence this outcome: (1) the deployment of enough devices to collect large datasets of high-quality imagery for labelling and (2) the development of AI interpretation models that are more capable per watt of compute power, specifically for meteorological tasks.

Prototype WxStation devices have been installed in real-world environments, including examples that will be highlighted during the talk. A deployment during the TEAMx intensive observing period (~3 summer months) in the European Alps captured frequent thunderstorms and rapid cloud‑field transitions. Another unit at the University of Leeds Farm (UK) has demonstrated day‑to‑day reliability in an agricultural setting. Some cold-weather testing down to –50 ºC will also be discussed. These deployments informed enclosure hardening, uptime targets, remote management, and early data intercomparisons against co‑located reference instruments.

The goals of this talk are to both engage and collaborate with the ground-based monitoring network community to propel this idea forward into new climates. Ideally, we seek network operators, research observatories, boundary‑layer testbeds, and emergency management–oriented networks to co‑design pilot evaluations in 2026. Pilots would quantify installation/ops burden (PoE power, mounting, maintenance), data pathways and latency, and product skill (cloud fraction/type/motion) versus co‑located references (e.g., ceilometers, trained observers). The aim is publishable evaluation results, a reusable ingest/QA template, and a clear path to broader operational use; including, if possible, exploratory NWP assimilation experiments (OSEs) as a stretch goal.