Clouds play an important role in influencing the dynamics of local and global weather and climate conditions.
Continuous monitoring of clouds is vital for weather forecasting and for air-traffic control. Convective clouds such as
Towering Cumulus (TCU) and Cumulonimbus clouds (CB) are associated with thunderstorms, turbulence and
atmospheric instability. Human observers periodically report the presence of CB and TCU clouds during operational
hours at airports and observatories; however such observations are expensive and time limited. Robust, automatic
classification of cloud type using infrared ground-based instrumentation offers the advantage of continuous, real-time
(24/7) data capture and the representation of cloud structure in the form of a thermal map, which can greatly help to
characterise certain cloud formations. The work presented here utilised a ground based infrared (8-14 μm) imaging
device mounted on a pan/tilt unit for capturing high spatial resolution sky images. These images were processed to
extract 45 separate textural features using statistical and spatial frequency based analytical techniques. These features
were used to train a weighted k-nearest neighbour (KNN) classifier in order to determine cloud type. Ground truth data
were obtained by inspection of images captured simultaneously from a visible wavelength colour camera at the same
installation, with approximately the same field of view as the infrared device. These images were classified by a trained
cloud observer. Results from the KNN classifier gave an encouraging success rate. A Probability of Detection (POD) of
up to 90% with a Probability of False Alarm (POFA) as low as 16% was achieved.