Research
Since 2006, all-sky imagers deployed as part of the THEMIS mission have been capturing ground-based images of the night sky every three seconds at sites across North America. While the broad phenomena that lead to the aurora are well-understood, the smaller-scale shapes captured in these images and their connection to important underlying magnetospheric processes are not.
We use machine learning to automatically classify all-sky images into one of six distinct categories (arc aurora, diffuse aurora, discrete aurora, cloudy, moon, or clear/no aurora), with the overarching goal of enabling large-scale statistical studies of the aurora that will lead to a better understanding of the relationship between auroral morphology and the underlying dynamics of the magnetosphere. In order to accurately assess the performance of our algorithms on the wide variety of images in the THEMIS dataset, though, we need a representative sample of images that have been accurately labeled.
So far, only a very small proportion (less than 1/10000 of 1%) of the dataset has been labeled. This is where you come in - the classifications you make here will enable us to develop the next generation of machine learning algorithms for better understanding the connection between observed aurora and the driving magnetospheric processes.
AuroraClick is independent of the THEMIS mission but would not exist without it. The University of California, Berkeley Space Sciences Laboratory and the University of Calgary Space Physics Group developed and deployed THEMIS-ASI, and continue to operate and produce the auroral images used in AuroraClick. UCalgary and UCBerkeley acknowledge funding from the Canadian Space Agency and NASA.
We acknowledge NASA contract NAS5-02099 and V. Angelopoulos for use of data from the THEMIS Mission. Specifically: S. Mende and E. Donovan for use of the ASI data, the CSA for logistical support in fielding and data retrieval from the GBO stations, and NSF for support of GIMNAST through grant AGS-1004736.