• Ready for review
  • Source: Night Lights

     

    Night Light Vessel Detections in Skylight come from VIIRS, the Visible Infrared Imaging Radiometer Suite onboard two satellites - Suomi NPP, and JPSS-1 (NOAA-20) (link to NOAA.gov). VIIRS is a low resolution electro-optical imaging system with high spectral fidelity that capture images of the earth at night. It’s possible to detect vessels from the light they emit at night.

    While the two VIIRS satellites may detect typical deck or running lights on vessels, the brightest and most reliable detections tend to be from the working deck lights or the lights used to lure plankton or squid in industrial fishing fleets. Using the data from VIIRS, Skylight displays those vessels detected operating with ‘Night Lights’.

    Likely a single vessel (“likely” because we cannot distinguish multiple vessels within a single pixel)
    Likely a fishing fleet

     

    Images are collected globally at 750 meters per pixel and displayed in Skylight on 15x15 km image chips as seen above. It is important to note that each bright spot does not necessarily correlate with one vessel. A single lighted vessel pixel may or may not correspond to one vessel - we cannot distinguish multiple vessels within a single pixel. Skylight correlates the nearest vessel to the center of each VIIRS pixel, but 750 meters is a very coarse resolution to begin with.

    What are the benefits of Night Lights?

    Night Lights have several benefits over other remote sensing methods:

    • They are collected several times nightly and globally. This is a big advantage for areas not regularly covered by other satellites.

    • They are collected multiple times per night, generally between 01:00 and 04:00.

    • The latency, time from data capture to its availability in Skylight, is about 3 hours from the time the data is collected until it is displayed in Skylight. That’s quite low compared to other data sources.

    • If the vessel is transmitting AIS, Skylight can determine the identity of a Night Light vessel detection.

    How has Skylight addressed known issues with or limitations of VIIRS data?

    There are many sources of light at sea, both naturally occurring, such as moonlit clouds, lightning, and the aurora, and from other light sources like oil platforms and lights close to shore. Therefore, Skylight uses strategies to address these other sources of light and ensure they’re not displayed as a vessel detection.

    Land light

    • False detections caused by light spreading from land over the ocean. Skylight has incorporated a sea land mask to remove false detections caused by light from land.

    Full Moons

    • False detections caused by full moon illumination. Because VIIRS senses light, the illumination from a full moon, especially if there is cloud cover, produces false detections.

    VIIRS imagery on a moonlit night
    • One solution would be to remove all detections during full moons. However, if the cloud cover is minimal during a full moon, vessels can still be detected.

    • The following two images are examples of the impact cloud cover can have on moon illumination and vessel detections. On the left (Jan 31) is an image captured on a cloudless, moonlit night; vessels can be detected. On the right (Feb 1) is a cloudy moonlit night; vessels cannot be detected.

    • Skylight uses a strategy to remove the false positives that occur when moonlight shines on small clouds as we see in the example above from February 1. In instances when there is high moon illumination but the atmosphere is cloud-free as is seen in the January 31 example above, vessels can be detected. To avoid removing true detections Skylight computes moonlight illumination and cloud cover. If moon illumination factor > 80%, detections under clouds are rejected and detections beneath cloud free skies are kept.

    Lightning

    • False detections caused by lightning are identified and removed from Skylight.

    Gas Flares

    • False detections caused by gas flares are identified and removed from Skylight.

    South Atlantic Anomaly

    • False detections caused by the well-known South Atlantic anomaly (link to Wikipedia) are identified and removed from Skylight

    Ionospheric Noise

    • False detections caused by ionospheric noise (>1000 nanowatts) that occur when the Satellite flies through the auroral zone are identified and removed from Skylight.

    Are larger vessels more likely to be detected?

    VIIRS is sensitive to a broad range of light intensity. In practice, brighter vessels will be more visible. The size of the vessel is less of a factor. For example, multiple squid fishing boats or multiple purse seiners packed very tightly together, will result in the brightest signal.

     

    How real-time is VIIRS data?

    Like any satellite-based data source, there is a delay - called “latency” - between when the image is collected by the satellite and when it appears in the satellite platform. Generally, Night Lights data is displayed in Skylight about 2-hours after it’s been collected.

    What is the revisit rate for VIIRS?

    How often a satellite or group of satellites come back to the same location is called the “revisit rate”. For Night Lights, every location is imaged at least once a night. Some areas may be imaged multiple times a night - the revisit rate is latitude and longitude dependent.

    Why are there no collection areas for Night Lights as there are for other detection sources?

    There are collection areas describing the satellite’s footprint but Skylight has chosen not to display Night Light frames in Skylight. Because Night Lights are collected globally, at least every 24-hours, displaying the collection areas in Skylight would quickly overwhelm the globe, making the rest of the underlying map almost unreadable. Below is an example of what Skylight would look like if we displayed the collection areas for Night Lights.

    Why do I see two detections of the same vessel very close to each other?

    There may be times when there are two detections displayed for the same vessel that are very close in time and distance. This can happen for a couple of reasons:

    1. The Hoyu Maru 37 may have been captured by both VIIRS sensors, Suomi NPP and NOAA-20 so Skylight displays two detections for the same vessel.

    2. It’s also possible that the Hoyu Maru 37 was captured twice by the same sensor but in different, adjacent swaths. In this case Skylight will display both detections for the vessel.

    The video below shows the orbits of the two VIIRS sensors, NOAA 20 and Suomi NPP.

     

    Sources and additional documentation:

    Black Marble User Guide This was consulted extensively by Skylight Machine Learning team as the VIIRS model was developed

    https://viirsland.gsfc.nasa.gov/PDF/BlackMarbleUserGuide_v1.2_20210421.pdf

    Correlation of vessel category and vessel brightness:

    https://www.mdpi.com/2072-4292/11/9/995  

    Cloud detection (SSEC docs)

    https://ladsweb.modaps.eosdis.nasa.gov/missions-and-measurements/products/CLDMSK_L2_VIIRS_SNPP#overview

    Land sea masking (NASA docs)

    https://landweb.modaps.eosdis.nasa.gov/QA_WWW/forPage/MODIS_C6_Water_Mask_v3.pdf

    Night Lights Open sourced code/model (this is Skylight’s public codebase)

    https://github.com/allenai/vessel-detection-viirs