Skylight is a tool for surfacing suspicious events in the maritime domain. Skylight is focused on addressing illegal, unreported and unregulated (IUU) fishing, but can be applied to many types of maritime crimes. Skylight was developed to enable maritime analysts to identify suspicious vessel behavior and alert authorities who can investigate, and take enforcement and compliance action when necessary. We strive to help analysts focus their time on the most suspicious activities in their domain rather than analyzing all vessels and vessel tracks. Because our focus is on surfacing anomalous activity, the Skylight visualization presents events rather than the last known positions of vessels.
Skylight is accessible in two ways. The first is a web-based application that any account with internet access can reach. The second is via an Application Programming Interface (API) that can connect the Skylight system with other maritime domain awareness tools.
The Skylight application works best when accessed on the most recent version of Chrome, Safari, Edge, or FireFox.
Skylight events are automatically generated events that may be of interest to a user. Some events are generated globally without any action by the user (e.g. Standard and Dark Rendezvous events). Other events require user configuration or input before they start to be generated (e.g. Entry events, Dark Detection events).
Alerts are events that are notified to a user via email or in-app. Users decide what types of events should be alerts when they create alerts. Alerts require users to select an area of interest and optionally select filter conditions that make a subset of those events particularly suspicious. Without setting up alerts, users will only view events on the map without any additional notification.
Skylight uses Automatic Identification System (AIS) and Synthetic Aperture Radar (satellite radar or “SAR”). AIS is necessary for all of its real-time and near-real time monitoring, whereas satellite radar carries a delay.
Skylight receives real-time AIS data from SPIRE, so elements like the Last Known Positions of vessels are updated immediately (within seconds or minutes in most cases). These events need some time before the model has confidence that the event is occurring:
These events are near real-time:
When you are on the main map, there is a bar at the top that informs how long it has been since the map has refreshed. The map must be refreshed in order to see new events.
Skylight is available through partners like the UN Office of Drugs and Crime, National Geographic Pristine Seas, and the Wildlife Conservation Society. If you have not reached this page through the introduction of a partner, please reach out to firstname.lastname@example.org about your interest and we will contact you with next steps.
You can create your own AOIs via drawing, typing lat/lon coordinates, or uploading a KMZ file. Drawing AOIs is helpful for analyzing whether events are occurring inside or outside those areas, as well as for setting up alerts. When you create an area, it will show up under the “My Areas” layer.
Note that there are limitations to creating AOIs:
Use the Areas menu on the top left to get a dropdown list of all areas that have been created on your account.
First, create a custom area in the Skylight system for your area of interest (AOI). During the creation process, you will have the option to “generate entry events” for that area. If your area has already been created, use the Areas menu to find your area and make sure you have "Enable Entries" selected. If it is selected, the option should be changed to "Disable Entries".
The first time you choose to generate entry events, it will take 1-2 hours for entry events to start appearing on the map. Once these entry events start generating, they will continue to be updated on an hourly basis.
Historical entries into an area are not currently available. Only entries from when you started generating events are available.
No. Entry events are generated the first time Skylight detects a vessel inside an area of interest (AOI), but it is not the last known position of that vessel. Vessels have probably traveled since generating that entry event. Use Tracks View to check where the vessel traveled before and after the Entry event.
Unfortunately at present, you cannot create AOIs from the other layers in the platform (e.g. EEZs layer, 12 NM layer, bathymetry). That means if you want to monitor your EEZ, you should trace the EEZ boundary using the drawing tool or upload a KMZ file from the web.
Note that there are limitations to creating AOIs:
A vessel will generate Entry events whenever it enters any area that has been enabled to monitor Entry events. If you have overlapping areas enabled for entries, you might experience the same vessel entry position recorded in multiple Entry events.
No, only the first entry into an area triggers an Entry event. Transmissions captured in the area after the initial entry will be captured in the Tracks View of that Entry event. A new Entry Alert will be triggered if:
A Standard Rendezvous indicates when two AIS-transmitting vessels have come within a short distance together, which is helpful for (but not exclusive to) analyzing potential, at-sea transshipment events.
We use a rules-based method to determine whether two vessels are participating in a rendezvous. Two vessels have to come within a short distance of each other for at least 30 minutes and travel at approximately the same speed and heading to be considered a rendezvous. When we say “short distance”, vessels should be within 250 meters of each other within 3 minutes of the AIS position reports.There are some additional parameters we apply to try and exclude activities like vessels carrying fishing equipment outfitted with AIS.
A Dark Rendezvous indicates that an AIS-transmitting vessel may have come close together with a vessel that is not transmitting AIS, which is helpful for (but not exclusive to) analyzing potential, at-sea transshipment events.
Unlike Standard Rendezvous events, we use machine learning to determine whether a vessel may be participating in a rendezvous with a dark (non-AIS transmitting) vessel. Vessels loitering in the high seas have anomalous tracks that can often be identified by the human eye. Given that, we had subject matter experts look at randomized vessel tracks and draw boundaries around tracks where they believed a rendezvous occurred with another vessel not transmitting AIS. We fed those annotated tracks to a machine learning model and the model learned the characteristics of tracks where rendezvouses occurred and did not occur. Skylight now applies that model to a subset of incoming tracks to determine where Dark Rendezvous may be happening in real-time.
Anything dependent on AIS alone cannot be used to confirm that a transshipment occurred. Without imagery or eyes on the water, we can never confirm that a transshipment occurred. However, we can build confidence by checking various factors to support the use of a patrol or flight.
First, a Standard Rendezvous carries more confidence than a Dark Rendezvous because there are two transmitters consistently transmitting rather than one. The Dark Rendezvous model should reflect what expert analysts would assume to be a Dark Rendezvous based on AIS tracks alone (i.e. automate the analysis of suspicious tracks for the analyst workflow). Presently, the Dark Rendezvous model matches what those subject matter experts would have picked up at a precision rate of ~42%, but for fishing vessels this precision rate is significantly lower.
Yes. Because detecting Dark Rendezvous is dependent on machine learning, other types of activity may be falsely labeled as a Dark Rendezvous. In general, Dark Rendezvous events are interesting because vessels are slowing down and making maneuvers in the water when normally they would travel from Point A to Point B. When analyzing Dark Rendezvous events, consider sea state, local structures (e.g. oil platforms, pipelines), and port lanes while analyzing whether an event was a transshipment.
If you see a series of Dark Rendezvous events next to each other, it may be because a vessel is meeting multiple vessels, one vessel for a long period of time, or no vessels at all but loitering similarly for some other reason. Multiple Dark Rendezvous events were generated because the vessel is still displaying behavior indicative of a Dark Rendezvous. We limit the max period of time a single Dark Rendezvous can be occurring based on vessel type.
Transshipment confidence is a proxy for how likely it is that goods were transshipped in rendezvous events. Transshipment confidence for Dark Rendezvous events considers duration and vessel type, while transshipment confidence for Standard Rendezvous events considers duration, vessel types, and flags. Duration is the most powerful factor; in general the longer the rendezvous event was detected, the higher the confidence that a transshipment occurred.
‘High’ confidence transshipments do not necessarily make those events more important or suspicious. For example drug trades, which can happen in minutes, will likely show up in the platform as ‘low’ confidence transshipments while fish transshipments, which take hours, will often show up as ‘high’ confidence transshipment events. Use transshipment confidence to filter on the types of trades/goods that are of most-interest to you, particularly in areas where there are too many Rendezvous events being generated to analyze them all.
For Standard Rendezvous events, we had subject matter experts score how likely it was for certain vessel combinations to conduct transshipments. Their primary decision factors were vessel types, the duration of the rendezvous, and the flag of the two vessels. These scores were then fed into a machine learning model which applies their thought process to all incoming Standard Rendezvous events on the Skylight platform.
For Dark Rendezvous events, we calculate transshipment confidence based on vessel type and duration alone. This is with a rules-based approach rather than machine learning.
These are vessels that were in the vicinity of the Dark Rendezvous and could have feasibly turned off their AIS and participated in a dark transshipment with the main vessel. Each partner is listed with a “low”, “medium”, or “high” demarcation based on how feasible it was for that partner to have reached the Dark Rendezvous location.
Note that we only list 3 partners though there may be more than 3 in the vicinity. The partner may also not be any of those vessels and instead a vessel carrying VMS or no communications at all. There is also always the possibility there was no partner and this vessel maneuvered in a similar way to a rendezvous for some other reason. We suggest analysts consider the potential partner list alongside these other possibilities.
We do not generate Standard Rendezvous events within 10km of shore and Dark Rendezvous events within 100km of shore. Close to shore, vessels frequently travel slowly and close together without necessarily having a rendezvous. To reduce false positives, Skylight limits reporting standard and dark rendezvous close to shore. This means that there may be rendezvous activities happening close to shore, though they are not visible on Skylight.
The fishing gear classifier model runs once daily so there may be instances when new gear begins transmitting AIS before it has been classified as a non-vessel. In this rare instance Skylight may inform you of a rendezvous with this gear and later recognize it was a gear.
Skylight was built with the intention of detecting rendezvous between vessels, so we try not to surface rendezvous events with vessels and oil platforms. Skylight identifies likely platform locations via a cluster analysis process. In a few cases, users may still observe what they believe to be standard rendezvous with oil platforms because the platform was not recognized by Skylight’s clustering analysis process.
Yes, you can see the tracks for vessel(s) involved in an event by opening the event card and using the “Show Tracks” button. This will show vessel(s) tracks 48 hours before and 48 hours after that event. The intent of Tracks view is to help analyze events and determine if they are suspicious, not monitor vessels. As such, Skylight does not show tracks for vessel(s) beyond 48 hours of that event.
You cannot view historical tracks for vessels outside the context of events. If you search for a vessel in the Skylight search bar, you will find its activity history in list format (e.g. last known position, port visits, historical events), but you cannot view historical tracks.
In the ‘Tracks View’, tracks have been summarized so that only the points when the vessel changed direction or speed significantly are shown on the map. Vessel position transmissions in between those ‘change points’ are summarized as straight lines. This lightweight, summarized visualization is what Skylight machine learning models use to identify events, which is helpful for analysts looking to understand why machine-learning based events are generated (e.g. Dark Rendezvous events).
However, it also means that if you compare tracks in Skylight to other tools (e.g. SeaVision, MarineTraffic), you will notice some differences. You may also observe some cases where an event location does not align directly with the track because of how the tracks are generated. If you encounter an instance when the event indicator is not displayed on the track, please send us a screenshot of the tracks, with as much detail as possible about the event to email@example.com. We are always looking for ways to improve and having examples where the event and tracks do not align is very helpful.
Skylight has a machine-learning model trained to recognize characteristics of AIS-marked fishing gear. When an event is generated by an MMSI number that the model suggests is fishing gear, the event is excluded from appearing on the map. This applies to Entry, Rendezvous, and Dark Rendezvous events. For example, we intentionally avoid showing meetings between equipment and vessels as Standard Rendezvous events.
In some cases, it can be difficult for the model to distinguish between a vessel and a buoy. You may see, for instance, Rendezvous events involving an MMSI that is actually equipment. If you encounter an instance where you are really confident that an MMSI participating in an event is from equipment rather than a vessel, please send a screenshot of the event to firstname.lastname@example.org. We are always looking for ways to improve the model and having examples where you believe the model was wrong is very helpful.
Vessels can report themselves as the ambiguous type “Ship” in their AIS transponder by using codes 90-99. We show a vessel type as “Unknown” when the vessel met one of these conditions:
Note that “Unknown” vessel type is different from “Unknown Fishing” and “Other Fishing”, which are different types you can filter events with.
Skylight uses the "Nationality" filter to make it easier for analysts to search for events where the involved vessels have flags or ownership reside in specific countries. The nationality filter searches for a match in at least one of three fields visible in event details cards:
Skylight does not have national registries/lists from flag states.
Skylight uses Synthetic Aperture Radar or Satellite Radar (SAR) to identify vessels who are not transmitting on AIS. As long as the vessel meets the minimum resolution of the collection, it will be detected. If a vessel is transmitting AIS at the same time as the SAR collection, then it appears on Skylight as a correlated Vessel Detection. If the vessel is not transmitting AIS during the time of the SAR collection, it appears on Skylight as a Dark Vessel Detection. Vessels on the high seas who are not carrying AIS are generally suspicious.
However, note that AIS carriage is dependent on tonnage and voyage plans. If a country uses VMS, we suggest analysts check whether the Dark Vessel Detection can be found on VMS before deeming it a contact of interest.
Satellite radar is collected by several different providers and companies. Unlike satellite and terrestrial receivers for AIS which can capture transmissions across the globe at most times of the day consistently, satellite receivers for radar are available on a much less frequent basis and the cost of each collection is much higher.
For agencies using Skylight, that means Skylight does not currently conduct ongoing monitoring for any areas. All areas where we are collecting SAR had pre-planning and a time-limited, specific objective/mission. Skylight pays for satellite radar collections on behalf of agencies via contract with MAXAR Technologies and MDA, which manages the RADARSAT-2 constellation.
To place an imagery request, please log in to your Skylight account. In the "My Skylight" menu, choose "Request Satellite Radar Collection" and fill out the Imagery Feedback and Request form.