Analyzing Atmospheric Gravity Waves and Visualizing Machine Learning Results

The ANtarctic Gravity Wave Instrument Network (ANGWIN) is an international collaboration aimed at investigating the upper atmosphere dynamics over a continent-size region, using a network of all-sky airglow imagers (ASI). In an effort to streamline the identification of "clean" (cloudless, moonless, and auroraless) windows in the extensive ASI data sets obtained since 2012, we have developed a machine learning algorithm that sorts "clean" (marked as 0) images from "obscured" (marked as 1) images. Already, we have successfully created a LightGBM (Light Gradient-Boosting Machine) models that accurately sorts through images taken at the Davis, McMurdo and Halley research stations in Antarctica. Utilizing a combination of old and new techniques, we are working on improving the LightGBM model for the Rothera station. As was done to improve other models, the misidentified windows from the Rothera model were combed through to identify weak points in the model. Once identified, those weak points were relabeled and added to the model's training set to strengthen the model's output. In addition, the development of Rothera's model uniquely utilized visual tools to document its training set and verify "clean" windows. These tools include mean images used to quickly identify weather features that are present in an entire window and graphing the model specific "scores" the machine learning's artificial intelligence (AI) assigns to each frame for a given data set (per night for a season's worth of data). This combination of techniques has allowed us to successfully improve the Rothera model, and will be used to verify the cleaning of future datasets.