Weather Classification using Machine Learning

By Greg Furlich

Telescope Array Collaboration, University of Utah

The Telescope Array (TA) cosmic ray observatory located in Millard County, Utah is the largest cosmic ray observatory in the northern hemisphere. It operates 507 surface detectors and 3 Fluorescence Detector (FDs) sites, Black Rock (BR), Long Ridge (LR), and Middle Drum (MD) to detect ultra high energy cosmic rays (UHECR). The FDs operate on clear, moonless nights to best observe the cosmic ray Extensive Air Shower that excites the Nitrogen in the atmosphere. However, sometimes the detector operates when the night is cloudy and this affects the scattering fluorescence light in the atmosphere diminishing our ability to properly reconstruct or simulate the cosmic ray event. In order to flag and remove cloudy weather from the FD data, neural networks were trained on snapshots of the night sky created using the FADC pedestals of each PMT at BR. Starting with simple neural networks and building up complexity, we were able to achieve high accuracy of weather classification and classifying each part allows for better time resolution of the operation night's weather progression.

System Status

Cluster Utilization