Pierre-Dominique Pautet, Yucheng Zhao, & Michael J. Taylor

Document Type


Publication Date


First Page


Last Page



In an effort to streamline the identification of "clean" windows of airglow images in all sky imager data for the ANGWIN experiment, we have developed a Light Gradient Boosted Machine (LightGBM) learning algorithm that sorts "clean" (marked as 0) wave images from "obscured" (marked as 1) images. These "clean" windows are then processed and undergo FFT-spectrum analysis. We have already successfully created LightGBM models that accurately sort through images taken at the Davis, McMurdo, and Halley research stations in Antarctica. Imager data from the Davis and McMurdo station has been fully processed from the years 2012 to 2022 with clean windows identified by using their respective LightGBM Models. The LightGBM model for the Halley station was recently verified and already several years' worth of data have been processed. To gauge the effectiveness of the three models, phase velocity spectrums from a season's worth of data from each station were compared against each other as well as previous findings from each station.

Included in

Physics Commons