Why “our data is 50 percent more accurate” does not mean anything in ocean modelling
In this article
What ‘accuracy’ really means in hydrodynamic modelling
How accurate are hydrodynamic or metocean models? It’s a question Tidetech hears from time to time from all sorts of people — sailors planning a race, ports managing vessel traffic, voyage optimisation providers… It’s also a question that doesn’t always have a simple answer.
In this article, Tidetech’s chief scientist Roger Proctor explains why model accuracy can’t be reduced to a single number or percentage, why more data doesn’t automatically mean better predictions, and what actually matters when assessing the quality of metocean information.
He explains that any comparison of model predictions is not as simple as claiming Tidetech’s accuracy is percentage points greater than hydrographic office or competitor data.
Be wary of providers that make that sort of claim — “our data is 50 percent better than this other data” — as it is potentially a demonstration of how little they understand metocean data collection and modelling, Proctor says.
Many factors influence metocean modelling at a global or local scale, and the often relatively sparse empirical measurements available mean there is no universal answer to the question of model accuracy.

“Our clients are usually trying to compare our results against basic hydrographic data such as tide tables or data from the weather provider they are contracted to,” Proctor said.
“We take the time to demonstrate the quality of our information for the comparisons that are important to them, and we explain why what they sometimes think is good data, isn’t. Sensor errors, data gaps, and inconsistent readings can significantly reduce accuracy in ways that are not made obvious to end-users.”
More data doesn’t necessarily mean better models
Tidetech uses a range of global and regional ocean and weather models from various sources and additionally constructs hydrodynamic models of tidal elevation and surface currents using open source software. The models are validated using available observations.
“The quality of the data used to build a model is critical to any measure of accuracy that can be provided about its results,” Proctor said.
Here are some examples:
In the case of surface currents, validation against current meter readings is preferable, especially if these measurements are for a month or longer. However, these data are not always available, and sometimes it’s clear that much shorter time series have been used (sometimes a few days) which can introduce a distorted impression of current variability.
Water level predictions can be validated against tidal elevations obtained from analysis of tide gauge data. At least 12 months of observations is required to accurately predict future tidal elevations. Often only major ports have this information. For many others, time series are much shorter, so longer-term variability is missing or inferred from major ports. Meteorological effects such as storm surges or wind-driven currents are often not satisfactorily identified in these shorter time series.
Taking a sea surface temperature measurement generated from satellite data and comparing it to a reading from a floating buoy might reveal different values because the geographical coordinates are in error or because the sensor on the buoy will be calibrated slightly differently to the way the satellite is seeing what sea surface temperature is.
“If you attempt to assimilate both measurements into your model, because of the difference, numerical instabilities could arise to degrade the solution,” Proctor said.
Tidetech evaluates each source of data when establishing its models to confirm whether new data will add accuracy to the model or not. To understand the mechanisms that can create errors, Tidetech uses a hierarchy of techniques, including analysing correlations, model bias and of the distribution of model errors both spatially and temporally.
“Limited observations are usually available for supporting the development of high resolution models of coastal regions, so our preference is to use these in validation, rather than assimilation, to give confidence in the modelled solution,” Proctor said.
Understanding metocean processes over space and time
Local area models typically cover a pre-defined domain and can be resolved to grid levels of 50 or 100 metres.
Metocean data is used to describe the hydrodynamic character of a gridded area and to determine how that area reacts over a given timespan. A technique called boundary forcing ensures that data from beyond the focus area such as currents and atmospheric pressure is used to ensure the model domain is not treated as an isolated part of an obviously connected surrounding ocean.
“The purpose of assimilating observations into a model is to improve the representation of processes by reducing errors, measured against an independent set of observations. This requires some understanding of the radius of influence of an observation and an appreciation of whether or not addition will make some improvement. By introducing ‘ghost’ observations into simulation experiments, it is also possible to show where and what kind of additional observations would prove most fruitful,” said Proctor.
What modelling expertise means for those putting it in practice
A perfect illustration of what the technical details of Tidetech’s work mean in practice can be seen when examining Deepsea Technologies market-leading weather routing. Even the more advanced providers will run their weather predictions through a relatively basic, non-specialised vessel model, with the aim of providing a voyage speed plan that saves fuel. However, because the model is only 80 percent accurate, it doesn’t capture the vessel’s real performance in enough detail to actually save fuel.
DeepSea Technologies uses the industry’s most advanced AI to overcome this limitation for its ship and voyage optimisation tools. The company models vessel behaviour at sea with 99 percent accuracy to help their customers cut fuel costs by up to 10 percent.
“Obviously metocean data is key. We need good reliable data, good coverage, and good granularity, to understand vessel behaviour under different conditions,” said Stavros Paschalakis, Chief Technical Officer at DeepSea Technologies.
“If you don’t have that, the AI models are going to learn an average overall weather-agnostic state which is not useful enough to meet the needs of the industry today,”
Yacht racing is another application where Tidetech’s expertise offers advantages. Navigator Andy Green attributed his success in a recent Admiral’s Cup race to the accuracy of Tidetech’s modelling. During the event, he studied three competing tidal models for the Solent and English Channel. Each was tested against the yacht’s measured set and drift, checked against satellite imagery, and observed at times of maximum tidal complexity.
“We kept coming back to Tidetech,” he said.
“It was consistently the closest to reality — within about five degrees of the yacht’s set and within 0.1 knots of drift.”
Safety depends on a navigator’s ability to understand weather modelling and how their boat will respond to the conditions.
“Sometimes the models are spot‑on, other times they’re wildly different,” Green said.
“Your job is to know which tools you can trust at any given moment.”
Having the confidence to lead
From racing yachts to cruise liners, port authorities to hydrographic offices, Tidetech’s predictions are now used around the world. The company hosts a team of world-leading scientists and IT specialists and has partnerships in place which ensure access to the most accurate data and local experience.
“We make every effort to ensure we get a solution that we are willing to sell to customers, and we’ve got skilled modellers who know how to pull it all together,” said Proctor.
When ocean data matters, Tidetech is ready to deliver with experts who understand the accuracy of their models.
Related blog articles
Why “our data is 50 percent more accurate” does not mean anything in ocean modelling
How accurate is a hydrodynamic model? Tidetech's chief scientist Roger Proctor explains why the answer is never a single number, and what actually matters when judging model quality.
What S-100 means for ECDIS: from static charts to live navigational displays
S-100 is set to turn ECDIS from a static chart into a live, layered picture of the sea ahead. We explore how better bathymetry, water levels and current data will sharpen decisions on the bridge and in port.
Reading the current – our takeaways from the Australian Commercial Marine Conference
Two speakers stood out at last month's Australian Commercial Marine Conference — an economist with a frank read on Australia's fuel security, and an industry chief with the numbers to finally put commercial marine's economic case to government. Here's what caught our attention.





























