Accurate and efficient conversion of AIS data into market information could be the biggest leap in shipping market efficiency and transparency in decades
Vessel tracking could reveal valuable information, not just safety, but commercial – trade route growth, regions of weak vessel supply, volumes of cargo at sea, reliability of liner schedules and vessel or company key performance indicators.
The current state of the commercial Automatic Identification System (AIS)-derived intelligence market can be split into three types of players – the AIS data providers, the cleaners of AIS data providing ‘dots on the map’ products and the high-end commercial intelligence providers.
The first hurdle in turning the AIS signals transmitted by ships into commercial intelligence is acquiring sufficient AIS data. The AIS signal is received by other ships, ground receivers and satellites. Ground receivers are situated in ports and coasts around the world, but only satellites can acquire AIS signals from vessels on trans-oceanic voyages. With the recent launch of the Spire nanosatellite network, there are at least three global AIS satellite networks – Orbcomm, ExactEarth and Spire (see table). These three companies combine their satellite data with ground-receiver satellite data to offer global coverage.
The global AIS data providers face two main challenges – data volume and latency. Orbcomm, which works with the European Maritime Safety Agency said it receives 28M AIS position messages per day. This is a huge amount of data to process and the required processing time adds to the latency issue. Latency is the time between the broadcast of the AIS signal by the vessel and the time it is received by the data provider. Orbcomm reported 86% of the 28M daily messages have an average latency of less than 60 seconds.
ExactEarth, which works with VesselTracker (acquired by oil movements specialist Genscape in 2013) quote a similar figure, but are somewhat coy about the number of transponders and the first pass detection rate. However, latency is the main comparison used in the marketing.
Spire entered the market in 2017, and claimed its low orbit nanosatellites have a lower latency of 27 seconds. These numbers may be an important selling point when pitching to the military and government safety agencies, but when it comes to building applications for commercial shipping, these time differences are marginal and easily lost in the further processing time required to produce the commercially viable intelligence.
Dots on the Map
Below the level of the raw AIS data providers are those companies providing AIS data cleaning services and producing what are known as ‘dots on the map’ products. But why is this step needed?
The problems with AIS data start on the bridge of the vessel. The AIS transceiver may have been switched off by the crew on orders from the vessel operators. This may be to conceal commercial activity, or even illegal activity. There might be bridge instrument problems with co-ordinates, producing a voyage where the vessel cuts across Africa via the Sahara.
Leaving and entering port is a fraught time on the bridge, and with some systems the AIS report is updated manually, sometimes days later. Crucially, this includes draft changes, used by many commercial application providers to nominate if a vessel is laden or in ballast.
Then there is the problem of multiple spellings, where cargo could be ‘iron ore’, ‘ore’, or blank. The destination could be ‘Rotterdam’, ‘Rdam’ or ‘Armed guards on board’ (a favourite in pirate waters). In high traffic regions, like Singapore, even though there is sophisticated circuitry involved, the AIS signals clash, ship-to-ship, creating gaps and nonsense in the raw data.
A large proportion of these AIS errors are solved using artificial intelligence (AI) algorithms that learn, via human input, to record the most likely event. Such is the effort needed to process raw AIS data that there are intermediates offering these services. AIS provider Spire will preprocess data using AI to reduce the latency, while data factory Maerospace uses proprietary algorithms to predict vessel voyages to show the latest likely positions. Without this preprocessing, the AIS data feed could have vessels with positions ranging from a few minutes old to several days old.
The aim of the algorithms is to produce a series of most likely events for each vessel, and using the (hopefully uncorrupted) AIS timestamp, string them together to form the voyage. Key decisions to be made by the algorithm are when did the voyage start? Was the ship laden, ballast or part loaded? Why did the ship stop during the voyage? (bunkers, repairs, delay arrival and so on), When and where will/did the voyage end? These crucial events are not always apparent in the AIS data and may need an expert opinion. This has to be done for all ships in the data set. It becomes clear that hours, if not days, are needed for the algorithms to run through all the AIS data.
Commercially viable intelligence
The third level of AIS data usage is to provide commercially viable intelligence. There is no single definition, but depending on who you talk to in the shipping industry, there are certain demands and expectations of what such a product should provide. Providers include: Windward, ASX Marine, Reuters Eikon, IHS Markit, Platts and the hedge fund CargoMetrics.
The products should provide the ability to drill down through the AIS data to answer specific questions, and in doing so, provide the user with a level of competitive edge quicker and more accurate than conventional means. These demands and expectations can be assessed by a product's ability to answer questions in the following scenarios:
None of these scenarios would have been remotely possible even a decade ago, but with more satellites thrown up, the decrease in processing times and ever cleverer algorithms, the time will come when commercial shipping data will be commoditised and as easily discoverable as the price of crude oil.
|Global AIS data providers with own satellite constellations|
|Satellites||18 OG2||18 Iridium Next||40 nano|
|Average number of vessels||150,000||165,000||#|
|Average latency (seconds)||60||60||27|
|Average number of messages/day||28,481,735||7,000,000||#|
|First pass detection rate||86%||83%||#|
|Source: Orbcomm, ExactEarth, Spire, June 2018|
|# Spire has yet to release this data|
*According to Mathias Olsen and Truls Rønne Kopke da Fonseca, researchers at the Norwegian School of Business, the answer is no. In their paper “Investigating the Predictive Ability of AIS-data: the Case of Arabian Gulf Tanker Rates”, they came to the conclusion that there is “weak evidence in favour of using AIS-data for forecasting purposes”. Nonetheless, this is one area that freight traders are intensively interested.