Speaker
Description
UiO has been operating all-sky imagers in the arctic regions for over 20 years. Between October and March images are taken at two different wavelengths every 5-30 seconds, resulting in a collection of almost 8 million images taken in Ny-Ålesund and Longyearbyen since 2006. These images are supplemented by magnetometer measurements, information about GNSS-signal quality of satellites in the field of view of the imager and ceilometer measurements giving the cloud base height near the imager.
Using pre-trained neural networks, we extract numerical features from the images. These features are used to predict the type of phenomenon visible in the image and relate the image to its supplementary data.
Preliminary results show good correlation between the prediction of cloudy images and data from the ceilometer, as well as good classification of aurora into arcs, diffuse or discrete aurora. On smaller timescales, relationships between the features and magnetometer data can also be established.