Himes leads paper on use of Neural Networks to obtain Near-Real-Time Measurements

Michael Himes (614/MSU) led a recently published paper titled “Using neural networks for near-real-time aerosol retrievals from OMPS Limb Profiler measurements” in Atmospheric Measurement Techniques (June 2025); GESTAR II colleague Ghassan Taha (614/MSU) is a co-author. 

The paper discusses the benefits of using deep learning, a branch of machine learning (ML), which is focused on neural networks, as it applies to obtaining near-real-time measurements. This form of deep learning is “a complex approximation to the physics-based algorithm the researchers have been utilizing.” Dr. Himes explains that they “used ML/AI to measure aerosols in the stratosphere from major volcanic eruptions and wildfires. This is derived from measurements by the Ozone Mapping and Profiler Suite’s Limb Profiler (OMPS LP) instrument, which is currently flying on two satellites (Suomi-NPP and NOAA-21) and is planned for two additional satellites in 2027 and 2032.” To obtain the aerosol measurements, using the physics-based method required about two hours of data analysis after the download from the satellite; however, using the neural networks, the data analysis can be performed in “near-real-time, about two minutes.” According to Dr. Himes, acquiring the information sooner is beneficial in various ways: first, the data can be applied to “improved coordination of field campaigns and ground-based follow-up measurements”; second, it can be “useful for aviation safety, such as informing the Volcanic Ash Advisory Center of the location of volcanic aerosols. Ash can cause jet engines to stop working, either by eroding engine components or by causing blockages due to the ash particles melting and resolidifying in the engines. In addition to improved flight paths, providing this data could prevent “economic consequences, as was the case with the 2023 eruption in eastern Russia that led to the cancelations of many Alaskan flights.”

NASA Worldview provides access to any near-real-time NASA satellite imagery such as images of our planet and maps of active wildfires and flooding. Dr. Himes and his team have seen visualizations utilizing their data product incorporated into this online tool along with other related data products. Users can see the latest data measurements as soon as they are available. For example, scientists could track the transport of stratospheric aerosol, such as this snapshot of aerosol over Africa from an eruption in Indonesia about 10 days prior. Dr. Himes explains, “In these images, the white-to-red dots are the team’s new data product for stratospheric aerosol, while the dark blue area shows SO2 that was also emitted by the volcano. The aerosol and SO2 are located in roughly the same place, which is expected.” 

“In the scientific community, there is a lot of excitement around novel applications of neural networks to Earth science,” states Dr. Himes.