Earth  ID: 5040

Finding Dust at Night

The NASA Short-term Prediction Research and Transition (SPoRT) Center at Marshall Space Flight Center developed DustTracker-AI a physically-based machine learning model to track dust into the night-time hours. The model ingests NASA/NOAA Geostationary Operational Environmental Satellite observations to identify the probability of dust in the imagery. Dust storms typically occur in the spring and fall months in the southwest United States as strong cyclones create high winds and loft dust into the atmosphere, especially when dry conditions have persisted. Although dust is easily identified in visible imagery during the day, these satellite bands aren’t available as the sun sets. Infrared imagery which can detect dust both day and night is also limited as the sun goes down. As the ground surface cools after dark, the surface and lofted dust are a similar temperature, and it is difficult to distinguish dust from the underlying surface. Even satellite imagery compositing techniques are limited by the physical processes masking dust at night. The SPoRT team’s use of machine learning to develop DustTracker-AI leverages the power of data science to identify relationships in the imagery indiscernible to the human eye to detect dust in difficult scenes. The random forest model is trained on a dust storm database curated through expert analysis by remote sensing scientists at SPoRT. DustTracker-AI is extensively validated with a low false alarm rate and correctly labels 85% of dust pixels in difficult night-time scenes. With a focus on transition of research to operations, the SPoRT team is running DustTracker-AI in real-time, and the product was made available to NOAA National Weather Service (NWS) forecasters to assess product utility and performance during the 2022 spring dust season.

This animated dust event occurred from April 5-8, 2022, as a low pressure system created strong winds where dust was lofted from Colorado into the Texas on the 6th and 7th. The animation follows the dust plume that developed in Colorado on the 7th and eventually reached the Dallas/Ft. Worth area by the early morning of the 8th. NWS forecasters in Texas reported that DustTracker-AI enhanced their awareness of the dust plume beyond what they could discern with information from surface observations and standard satellite imagery. Without DustTracker-AI, the extent of the dust plume is difficult to assess in standard satellite imagery and virtually undetectable.

During this event, local air quality monitoring from the Ft. Worth area showed elevated particulate matter 2.5 observations had doubled from the previous day. Airborne dust can create hazardous conditions that impact health and air quality and frequently disrupt ground and air transportation. The ability to track dust further into the night-time hours through machine learning capabilities such as DustTracker-AI has the potential to enable weather forecasters, emergency managers, and health and air quality agencies to more efficiently track dust events, issue warnings/advisories, and anticipate the disruptions and hazards that accompany dust storms.

For related feature stories, see:

Visualization Credits

Anansa B. Keaton-ashanti (NASA/GSFC): Visualizer
Kel Elkins (USRA): Visualizer
Greg Shirah (NASA/GSFC): Visualizer
Emily B. Berndt (NASA): Scientist
Dauna Coulter (Media Fusion): Producer
Laurence Schuler (ADNET): Technical Support
Ian Jones (ADNET): Technical Support
Please give credit for this item to:
NASA's Scientific Visualization Studio

Short URL to share this page:

SVS >> Dust
GCMD >> Earth Science >> Atmosphere >> Aerosols >> Dust/Ash
SVS >> Calipso
NASA Science >> Earth

GCMD keywords can be found on the Internet with the following citation: Olsen, L.M., G. Major, K. Shein, J. Scialdone, S. Ritz, T. Stevens, M. Morahan, A. Aleman, R. Vogel, S. Leicester, H. Weir, M. Meaux, S. Grebas, C.Solomon, M. Holland, T. Northcutt, R. A. Restrepo, R. Bilodeau, 2013. NASA/Global Change Master Directory (GCMD) Earth Science Keywords. Version