Data, data, data and even more data is often the main direction of shipping analysis outfits to become even more accurate in their forecasts. Generally speaking, there are 3 dimensions one moves along to improve models: more accurate data, higher frequency (more speed) and new/alternative data. Though it must be noted that causation trumps all of the 3 dimensions above.

Shipping analysis has and still is heavily reliant on trade statistics reported by national statistical agencies. The statistics have mainly one flaw; the long time lags before they eventually arrive. Lags vary from country to country. The quicker national agencies can deliver the trade statistics within a month, while it is not uncommon to wait 3-4 months for the data to arrive from some countries. Circumventing these lags, some companies started tracking vessels physically for various segments. This was definitely a manpower intensive job. The introduction of alternative data, such as AIS data, was therefore seen as a welcome solution to the shipping industry (though of course its original intended use was not shipping analysis).

An AIS dataset which is properly cleaned and massaged has of course the potential to deliver along all 3 dimensions outlined above.  Instead of country of import/export given by the national trade statistics, AIS can deliver the actual import/export by port/terminal/berth. On the frequency scale, AIS signals are streamed on a near minute basis, enabling the possibility of real-time data. Further, voyages can be assessed, with possibility of measuring speed, waiting, congestion, port-time and other productivity factors. The potential for analysis is therefore endless and it can be conducted in real-time.

The catch, which is well known, is that the AIS datasets are far from complete. AIS signals have a tendency to disappear, particularly when approaching ports. Further, there seems to be a number of ways to manually affect the transmission of the signals. The automatic position, speed and heading data is generally reliable when received. However, the manual input regarding destination and, also draft is very often unreliable.

We (Maritime Optima) have conducted a segment study for LNG based on the AIS data we have collected since end of 2019. All LNG berths, terminals and ports were mapped. Initial logic was put in place to determine voyages (start and end points), including factors such as waiting time, speed and port days. The voyages have been aggregated and some of the results may be seen below:

Export of LNG:

Import of LNG:

The main conclusion is that, with fairly simple logic regarding arrivals/departures it is possible to construct voyages for the vessels. When aggregating the data we find that the AIS based data can have a high correlation with customs statistics (as collected by JODI). Examples of high correlations, are exports of LNG from USA and imports of LNG to France (top left hand graphs). However, there are also many export/import countries which only show a loose correlation with the customs statistics (see bottom right hand graphs).

It is apparent that the AIS data will need further processing, in order to reveal the data’s secrets. The next step we will perform is to build a trajectory/voyage database and with ML train models to predict more accurately the voyages (See: “Vessel destination forecasting based on historical AIS data”, by Morten Omholt-Jensen). This approach will bring more accuracy to the movement of the vessels and hence give a clearer picture of the aggregated volumes of goods moved between ports.