In the previous post we examined the effect of using 2D spectral energy data to determine workable weather windows in a small case study. Here it was found that it increased the workable weather windows by 40%, compared to using the conventional method (Jonswap with varying Hs). Determining the expected vessel motions based on 2D spectral energy data (energy over period and direction) therefore provides a method to determine workability of vessels more accurately.
Let’s take it back for one second. What if the data going into the model is not accurate to begin with? In the figure below we see a forecast for the significant wave height Hs as given by various metocean data suppliers. It is clear that there exists an uncertainty about the forecasted values for Hs, and for that matter the energy spectrum supplied by various metocean companies, as different companies indicate different values.
In both the industry standard method as the MO4 method the supplied data is used directly in a physical model to generate motion forecasts. This means that the uncertainty which is in the supplied data remains in the generated motion forecasts.
Potential solutions to tackle this problem could range from tweaking the ship model to comply better with the supply data, or one could try to improve the metocean supply data. The last suggestion is a concern of the meteorological studies, which is a specialization of its own. The fact that they often differ indicates that it is a complex process to forecast the expected energy at sea. Changing anything to the ship model to handle the errors might be handled by deploying very smart machine learning techniques. The problem with this approach is that it may give inconsistent results as the incoming data is strongly dependent on so many factors (location, global & local weather, etc). Apart from that, you are still working with inaccurate data.
In this blog post we take a different angle of approach. We will explain how we improve the motion forecasts by including the motion measurement data of a vessel. This method brings the real time data into the motion forecasts, which are actually virtual forecasts up until now. It seems intuitive: we should be able to learn something from the measured motion of the vessel.
With a mathematical model we combine the current MO4 forecasts with the data we obtain from measuring instruments on board. An example is given in the figure below, where the current forecast was clearly on a lower level. By combining the measurement with the forecast we get the forecast indicated by the red line, which clearly can lead to strong improvements.
This method leads to promising results for forecasting horizons of up to 12 hours. In this way it achieves greater accuracy for the short term forecasts. The fact that its performance is less for longer forecasting horizons is due to the nature of sea states generally disappearing after a number of hours and therefore the current measurements are not informative for further in the future.
With the corrected forecasts the results of the tool may be more in line with the intuition of the user overlooking the actual operation. Therefore it can increase the confidence of the user when making a decision, compared to the decoupled forecasts.
This blog post is likely not the suitable place to explain the mathematical models used to achieve this combined forecast. If you have any questions about the method or about the results, do not hesitate to reach out to us.
Author: Pieter de Wet
Pieter is graduating in the fields of both offshore engineering (TU Delft) as well as econometrics (Erasmus University Rotterdam). His research topic for offshore engineering focuses on DP operations, and for econometrics he investigated and developed various algorithms to combine measurements and forecasts.