Citizen-Built AI Tool Helps NASA Spot Night-Shining Clouds

A citizen scientist has turned a repetitive cloud-identification problem into a practical machine-learning tool for a NASA-supported research project. The system helps volunteers and researchers distinguish rare noctilucent clouds from ordinary clouds that can look similar near dawn or dusk.

The tool was developed by Space Cloud Watch volunteer Namai Chandra. Rather than replacing expert judgement, it screens images, classifies likely cloud types and routes uncertain cases for human review. That human-in-the-loop structure is designed to save time while preserving scientific oversight.

Why noctilucent clouds are unusual

Noctilucent clouds are also called night-shining clouds because they scatter sunlight after the Sun has set or before it rises. Their silvery appearance can make them striking, but identification is not always simple. Lower-altitude clouds may catch light in ways that produce a similar look in photographs.

The rare clouds form much higher in the atmosphere than common weather clouds. Researchers are interested in reports that they may be appearing more frequently and at lower altitudes than before. Understanding those observations could help scientists study changes in the upper atmosphere and possible links with long-term weather and climate patterns.

Space Cloud Watch asks people around the world to photograph the sky shortly after sunset or before dawn and submit their observations. That broad participation gives researchers more geographic coverage than a small professional team could gather alone. The cost is a large volume of images that must be checked carefully.

From volunteer observation to software

Chandra noticed that project leaders were manually verifying suspected noctilucent-cloud images. He proposed a pipeline that could handle routine screening while sending the most important or uncertain photographs to specialists.

Working with Space Cloud Watch scientists Chihoko Cullens and Brentha Thurairajah, he trained the system on varied cloud images. The training set included both confirmed noctilucent clouds and the lower-level formations most likely to be confused with them.

The finished pipeline performs several steps. It first screens submitted images, then estimates a classification and attaches a confidence level. A high-confidence result can help a contributor decide whether an observation is likely to be relevant. A low-confidence or scientifically interesting result can be flagged for human examination.

After development, testing and refinement, the tool was released for use by contributors and project scientists. It now gives newcomers a way to check a photograph before submission and helps the research team prioritise its review queue.

What this says about citizen science

Citizen-science projects are often described as a way for the public to collect data. This example shows a wider role. A volunteer did not merely supply an observation; he identified an operational bottleneck and built technology to improve the research process.

That contribution is possible because modern machine-learning tools are increasingly accessible outside large laboratories. But accessibility does not remove the need for scientific validation. Cloud images vary with camera settings, location, season, weather and light. A classifier trained on one collection may perform differently on unfamiliar examples.

The decision to keep humans involved is therefore important. Researchers can correct mistakes, inspect unusual cases and update the model as more verified images become available. Those corrections can in turn improve later versions of the system.

How people can contribute

NASA invites interested observers to join Space Cloud Watch and photograph the sky during the periods when night-shining clouds are visible. Useful observations need accurate timing and location information, and contributors should follow the project's submission instructions rather than relying only on the classifier.

The noctilucent cloud AI tool will not by itself explain why the clouds may be changing. Its value lies in making a distributed observing network more efficient and consistent. Better screening can produce a cleaner collection of observations, allowing scientists to spend more time investigating patterns in the data.

At a moment when artificial intelligence is often presented as a replacement for human work, this project offers a more grounded model: software handles repetition, volunteers expand coverage and scientists retain responsibility for interpretation. The result is a small but useful advance in understanding a remote part of Earth's atmosphere.

This Novexa News science report is based on information published by NASA's official citizen-science programme.