Successful maritime operations depend on accurate data regarding the actual position and route of the ship.
Managing a vessel from shore with erroneous position data can cost shipping companies a large amount.
Unfortunately, AIS is frequently unreliable, and Nautilus Labs has seen that even sensor-direct GPS data can be deficient depending sensor integrity and the ship in question. Position error can further be introduced throughout the data collection, retrieval, and storage process.
Nautilus Labs aims for its clients to trust all the data displayed to ensure confidence in the insights our platform provides. To do this for position data, the company implements proprietary techniques such as dead reckoning to detect, measure, and, if possible, correct for this kind of systematic error, helping clients answer the elemental question of where a vessel actually was at any point in time.
By combining accurate ship position data with weather data, owner-operators can take concrete actions to optimise voyage economics in real-time.
No physical sensor is free of error and noise, including on-board GPS receivers. Several on-ship subsystems can sit between a GPS receiver and a data logger, and they can introduce error and noise along the way.
At Nautilus Labs, hundreds of thousands of hours of GPS data has been collected, observing many different kinds of anomalies.
Sometimes independent latitude and longitude values can update at different frequencies. At other times, the GPS sensor interface does not indicate null values, so when the GPS source is powered off, zero values are not easily distinguishable from no data.
Sensor data collection and storage can also introduce error by not treating GPS data as a circular value in the 360° system when performing calculations. This causes anomalies when a ship crosses the 180° meridian – sometimes the ship will briefly appear on the opposite side of the Earth, at the prime meridian.
In many cases, the statistical impact of sensor noise can be handled with basic smoothing and aggregation. However, when it comes to certain kinds of data, even a small amount of noise can introduce a significant amount of visual disturbance for a ship shown on a map.
All too often, Nautilus Labs sees clients dealing with low resolution, low-frequency GPS data that also has noise. A fundamental question that clients ask is: “Where is my ship?”
When designing new systems, several practices can help mitigate these issues. Sensor collection systems should be able to distinguish an absent GPS value from a present GPS value.
Structured data collection could also support the use of unambiguous coordinate systems (such as n-vector geodesy, that avoid the pitfalls of the ambiguous 360° system).
Collection of sensor metadata could allow for the monitoring of the health and status of a physical sensor.
Nautilus Labs strives for this kind of excellence and is always raising the bar in terms of standards. However, at the same time, the company recognises that in order to deliver valuable solutions quickly to clients, it is not possible to fix every sensor-related issue at once.
To provide the best experience, Nautilus Labs must adapt to handle the downsides of current systems, using software to do it.
One way to solve this kind of systematic GPS error is to leverage dead reckoning and sensor fusion.
In a vacuum, a GPS receiver is extremely difficult to correct, but in the presence of a rich array of sensor values, including reliable weather data, ship heading and speed, previous position, and noon data points, dynamic models of a ship’s movement in the ocean can be created independent of GPS data.
Dead reckoning is one of the oldest forms of maritime navigation. Before on-board GPS and ECDIS systems, captains leveraged dead reckoning to track their course.
By using a known starting position, and the duration spent at different speeds and headings, an approximation of current position can be calculated.
The resulting ‘projected’ position can then serve as input to calculate the proceeding projected position. Of course, this technique is susceptible to accumulated error; the larger the difference in time between the known starting position and the current estimated position, the more error incurred and the greater the deviation from the true course.
The model can be improved by introducing other external factors such as wind and current speed into distance calculations, but accumulated error will always be present.
Part of the reason systems such as GPS are so popular is because they are immune to accumulated error. However, as described above, GPS data collection has its own issues.
Sensor fusion comes into play by allowing the empirical dead reckoning technique to be combined with the high-accuracy GPS data sources such as sensor data and noon reported values.
By selectively incorporating a known position as an input to a dead reckoning model, accumulated errors can be bounded and made close to negligible. In turn, the dead reckoned values can accurately fill in detail of low frequency or erroneous GPS data.
This is one of the big value propositions of comprehensive high-frequency data collection: by painting a complete picture of a ship and its environment, the accuracy and sophistication of insights can be improved.
While answering the question ‘where is my ship?’ is important, it’s only the start of the kinds of problems that can be solved by combining integrated high-frequency data, noon, and third-party weather collection with sensor fusion.
Nautilus Labs sees incredible value in bringing these same techniques to applications beyond GPS, enabling a class of retrofitted vessels to operate competitively in a rapidly changing maritime industry.