Volunteer Develops Machine-Learning Tool to Identify Rare Clouds

Scientists with NASA’s Space Cloud Watch project have asked photographers worldwide to submit images of noctilucent clouds—silvery, high-altitude clouds visible after sunset or before sunrise—to study long-term atmospheric shifts. These clouds appear lower and more frequently than in the past, prompting closer examination of Earth’s changing weather patterns. Volunteers submit images to help researchers track these changes, but distinguishing noctilucent clouds from similar-looking lower-altitude clouds has created extra work for project leaders.

Volunteer Namai Chandra proposed automating the initial screening process by building a machine-learning pipeline that could pre-screen images for noctilucent clouds while preserving human oversight for uncertain cases. The tool uses image classification and confidence-based routing to flag potential noctilucent clouds and route ambiguous images for expert review. Chandra collaborated with Space Cloud Watch scientists Drs. Chihoko Cullens and Brentha Thurairajah, who supported the development and refinement of the tool.

After multiple rounds of training on labeled cloud images, the pipeline was deployed to Space Cloud Watch contributors and scientists. Users can now upload images and receive automated feedback on whether their observations are likely noctilucent clouds, reducing uncertainty before submission. The tool also helps project scientists prioritize images for detailed review, improving efficiency in processing the growing volume of cloud data.

The Space Cloud Watch project invites new participants to contribute by photographing clouds just after sunset or before dawn. The volunteer-built machine-learning tool provides an additional layer of verification, helping observers confirm noctilucent cloud sightings before sharing them with researchers studying Earth’s evolving atmosphere.