Skip to content
 

ML4DNB

CIRA ML Home
ML Overview
ML4Clouds
ML4TC
ML4DNB
ML4DA
ML4RT
ML4Soundings
AI Institute
AI4NWP
Team
Publications
Internal

Machine Learning for the VIIRS Day-Night Band

Detecting Gravity Waves

One of our goals is to develop machine learning algorithms that can identify gravity waves from the Day-Night band, e.g., from images like the one below.

Gravity waves (concentric circles) visible in Day-Night Band imagery

A key challenge for the task of detecting gravity waves is the presence of many other signals in the background of typical day-night imagery, such as city lights, ship lights, lightning, clouds, coastlines, etc.  We are currently exploring several different techniques for the automatic detection of gravity waves.

Observations and Detection of Milky Seas

VIIRS DNB observations of milky seas published in Scientific Reports:
The paper titled, “Honing in on bioluminescent milky seas from space” was published in Scientific Reports on 29 July 2021. This paper discusses VIIRS Day/Night Band observations of 12 milky sea events. Milky seas are a rare phenomenon where the surface of the ocean appears to glow at night with the water taking on a milky appearance. Recorded observations of milky seas from mariners, dating back as early as 1864, often claim this glowing, milky water stretches from horizon to horizon around the ship. Milky seas are a form of bioluminescence believed to be caused by the saprophytic relationship between a luminous bacteria and microalgae that occurs on a large scale. However, given the rare nature of these events, little in situ research has been performed. Previously, the only satellite-based observations of a milky sea event came from the Operational Linescan System (OLS) onboard the Defense Meteorological Satellite Program (DMSP). Now, for the first time, researchers led by Steve Miller (CIRA) have searched exhaustively through the VIIRS Day/Night Band (DNB) record, dating back to 2012, and identified 12 milky sea events. It is hoped that, moving forward, routine inspection of DNB imagery in the regions where milky seas are most common can be coordinated with research vessels to directly observe a milky sea event as it occurs, thus answering many of the remaining scientific questions surrounding this mysterious phenomenon.

One natural extension of this work is to develop machine learning algorithms that detect this phenomenon automatically, and alert researchers.  A first step though is to identify enough imagery with this phenomenon that makes training of machine learning techniques feasible.

Full reference:

Miller, S.D., Haddock, S.H.D., Straka, W.C., Seaman, C.J., Combs, C.L., Wang, M., Shi, W. and Nam, S.-H. Honing in on bioluminescent milky seas from space. Sci Rep 11, 15443 (2021). https://doi.org/10.1038/s41598-021-94823-z

Interested readers are encouraged to check out the supplemental materials provided at the bottom of the above article (below the references) for a list of mariner reports of milky sea sightings and the full list of DNB-detected milky sea events – so far.

Comparison between Defense Meteorological Satellite Program (DMSP) Operational Linescan System (OLS) and Visible Infrared Imaging Radiometer Suite (VIIRS) Day/Night Band (DNB) imagery for a bioluminescent milky sea south of Java.  The glowing ocean surface, seen only on moonless nights and whose area exceeds the size of Iceland (> 100,000 square km) persisted offshore of Java for several weeks, drifting slowly westward between two ocean eddies.  The new DNB capability affords the first practical detection of milky seas via high-quality, calibrated low-light imagery–paving the way for future research expeditions.

Publications

Primary contact

Steve Miller (CIRA director, Professor – Atmospheric Science)