Jeffery Thomas Mitchell — Brookhaven National Laboratory
Scott Smith — Brookhaven National Laboratory
Andrew McMahon — Brookhaven National Laboratory
Stephen R. Springston — Brookhaven National Laboratory
Richard Wagener — Brookhaven National Laboratory
Laurie Gregory — Brookhaven National Laboratory
—
Category
General topics – Aerosols
Description
A time series of measurements from 3 co-located instruments: an ozone monitor, a carbon monoxide monitor, and a condensation particle counter. The measurements are taken during the same single day.
The Atmospheric Radiation Measurement (ARM) Climate Research Facility’s Aerosol Observing Systems (AOS) characterize aerosol properties at both fixed and mobile locations around the world. In order to relate in situ surface observations to the large-scale distribution of aerosols that influence cloud properties, it is important to identify and document measurements caused by concentrated local sources of particles, e.g., power generators or vehicles and airplanes. The detection of these local sources in the measurements is difficult and time consuming because it requires matching visual sightings of a local source, e.g., aircraft on runway, with sharp spikes in the measurements, e.g., CPC concentration levels (see figure below). We present a method using machine learning and image classification to detect these local sources for an ARM AOS facility located proximate to an airport on Graciosa Island, Azores. Two site cameras were recently installed at the AOS to help identify local source emissions on regional representativeness of measurements. Using visual images from the camera as confirmation, a neural network classification algorithm is developed to identify the presence of aircraft or vehicles in the vicinity. During the identified times of vehicle/airplane traffic we combine data from anemometers, trace gas monitors, and particle counters in a supervised classification algorithm to identify the unique signatures so those data can be accurately flagged and reported as local source emissions in much less time than could be identified by human intervention.
