Enhancing Atmospheric Measurements and Site Monitoring with Camera Imagery and Machine Learning

Description

The growing network of site cameras deployed across the Atmospheric Radiation Measurement (ARM) facilities presents a unique opportunity for site monitoring and to develop new data products using machine learning techniques. The high frequency of images captured at the ARM sites enables the application of convolutional neural networks (CNNs) for diverse image classification and object detection tasks. We demonstrate the potential of site camera imagery by successfully categorizing various cloud and sky conditions using a Visual Geometry Group (VGG) type Convolutional Neural Network. We also showcase the identification and characterization of a power plant plume using the same CNN architecture.

By combining the classified sky and plume characteristics with data from a colocated Meteorological Temperature Profiler (ATTEX MTP-5), we demonstrate that boundary layer temperature profiles can be more accurately constrained. Furthermore, we present preliminary work on detecting and classifying common objects captured in several ARM camera datastreams, highlighting potential applications for site monitoring and data quality control. These applications include monitoring vehicle traffic around instrumentation such as the Aerosol Observing System (AOS) to identify possible sources of contamination within the data, or providing notifications of polar bear activity to North Slope of Alaska (NSA) site operations for safety and logistical planning. The integration of machine learning with site camera imagery offers a promising approach to enhancing atmospheric remote sensing measurements and improving site monitoring capabilities.