Modern space research uses both satellite-born and ground-based instruments to measure the near-Earth space environment. Studying the auroral display provides information of the electric currents in the ionosphere, which is why automated imaging stations capture millions of auroral all-sky images every year. However, due to the nature of the aurora, these images are difficult to analyse automatically: photon-limited images are noisy, and objects are irregular and difficult to identify. We used hierarchical attribute trees in a large scale experiment with over 350,000 auroral allsky images. Tree-to-tree distances were utilised in classifying images and in locating similar images in content-based image retrieval fashion.