At Sofar, we deliver the most accurate marine weather, powered by the world’s largest private network of real-time ocean sensors, to improve safety and efficiency at sea.

Our sensors, known as Spotter buoys, drift freely across the ocean and make more than 1.5 million real-time observations of waves, inferred wind, sea surface temperature, and barometric pressure each day. Sofar’s team of scientists use techniques from the field of data assimilation to integrate the observations made by Spotters into our marine weather forecasts, improving accuracy by up to 50% compared to leading operational forecast centers. This advantage powers the marine weather intelligence at the core of our Wayfinder voyage optimization platform for maritime shipping.

In this blog, we will explore data assimilation further, and detail how our innovative use of its techniques drive major improvements in marine weather prediction.

What is data assimilation?

Data assimilation is a field of study that uses a combination of numerical model data and observations to produce a best guess of the state of a weather system. This current best guess of the atmosphere, waves, or ocean, is used to initialize a weather forecast created by a computer model, known as a numerical weather prediction (NWP) model.

Data assimilation follows the same logic taught in high school physics. To determine, say, the trajectory of a ball that you throw, you need to know the ball’s initial height and velocity. Similarly, to determine the best guess state of a weather system, you would ideally know the initial height, velocity, temperature, and many other properties of every parcel of air or water at an exact moment in time.

This, of course, is not possible — our planet is too large and weather is far more unpredictable than a thrown ball. The surface area of the Earth alone is roughly 510 trillion square meters, while the atmosphere extends 10 million meters high and the oceans plunge up to 10,000 meters deep. Even when we combine the billions of weather observations made on Earth each day by satellites, sensors, and other sources — all of which can be noisy and imperfect —  we still fall far short of a complete picture of our planet.

Data assimilation is a field of study that uses a combination of numerical model data and observations to produce a best guess of the state of a weather system.

There is one additional complicating factor — weather is extremely chaotic. For a ball, we can predict almost exactly where it will land as long as we are pretty close in estimating its initial height and velocity. For a chaotic system like the weather, however, any small uncertainty will grow exponentially over time; it’s as if the ball, at any moment, could reasonably be whipped away by a tornado or carried off by a bird.

Faced with weather uncertainty and insufficient observations, operational centers like Sofar and the National Oceanic and Atmospheric Administration (NOAA) use data assimilation to incorporate what observations we do have and produce a best guess of the atmosphere, ocean, or waves at a given time. The data assimilation process is iterated:

• Initially, there are no observations to go off of. A model is initialized from idealized conditions or from some estimate of climatology (i.e. the average state of the weather).
• Then, a few new observations are incorporated using a data assimilation algorithm. This brings the best guess of the weather a little bit closer to those observations.
• Then, the model produces a new forecast, new observations are made in new locations, a new best guess of the weather is made, and so on.

Data assimilation not only incorporates observations over time, but also carries forward the information from those observations into the future using the principles of physics. This fills in the gaps between observations across space and time.

How Sofar uses data assimilation to improve marine weather forecast accuracy

At Sofar, we use data assimilation to produce highly accurate marine weather intelligence. Our approach is distinguished by the millions of proprietary data points that we assimilate each day from hundreds of our Spotters around the globe, which comprise the world’s largest private network of real-time ocean sensors.

The observations made by the global Spotter network give us a more granular understanding of marine weather than conventional sources. Satellites, for example, provide indirect observations of wave height, while Spotter provides direct measurements of wave height, wave direction, wave frequency, and other important parameters, such as sea surface temperature and barometric pressure. By initializing our model with Spotter’s fine-tuned measurements of actual ocean conditions, we move our best guess of marine weather closer to reality.

‍In one peer-reviewed study, the assimilation of Spotter data into Sofar’s model was found to produce, “end-to-end improvements in forecast skill of significant wave height of 38%, and up to 45% for other bulk parameters,” when compared to the skill of Sofar’s model without data assimilation.

To better understand how the assimilation of Spotter observations improves the accuracy of Sofar’s models, let’s consider a real-world example:

• Imagine that a Spotter drifting in the middle of the Pacific Ocean encounters a large swell wave. The Spotter observes that the wave height is 10 feet, which is larger than the wave height that Sofar’s forecast had predicted.
• Data assimilation increases the wave height in the model, bringing the forecast closer to the observation.
• Over time, the real swell wave will propagate across the ocean. When the wave encounters another free-drifting Spotter, that Spotter may indicate that the wave height is actually 8.5 feet, a bit lower than what the model indicates. Data assimilation adjusts the model accordingly, decreasing the forecasted wave height. 
• This process occurs simultaneously for all waves propagating across the ocean at all scales. Hourly, every correction to Sofar’s operational forecast system is made simultaneously.

In addition to Spotter observations, Sofar also incorporates other data sources to further increase forecast accuracy. This includes observations from satellite platforms — such as remote sensing altimeter measurements of sea surface height from Jason 3, Sentinel-6a, and Saral/Altika — as well as data from other operational forecasting centers, like NOAA and the European Centre for Medium-Range Weather Forecasts (ECMWF). Sofar’s approach guarantees that our forecasts combine the best publicly available data with proprietary observations from the Spotter network to produce uniquely accurate marine weather intelligence.

Real-world impact: How Sofar’s use of data assimilation helps Wayfinder better optimize voyages for maritime shipping

Sofar’s use of data assimilation to deliver the most accurate marine weather has a direct impact on the utility of our Wayfinder platform.

Wayfinder provides maritime shipping vessels with dynamic voyage optimization to maximize the safety, efficiency and profitability of every voyage. A key input powering the daily RPM and route guidance that Wayfinder sends to vessels is Sofar’s marine weather intelligence; in order to instruct a ship where and how fast to sail, it is essential to know the latest weather, along with market and vessel performance data.

By assimilating observations from the global Spotter network, Sofar’s forecasts supply Wayfinder with an unprecedented real-time view of marine weather conditions along a vessel’s route. For Captains and Operators using the platform, this real-time weather visibility translates to more efficient sailing and increased savings in time, fuel, and emissions.

“Sofar Ocean’s real-time sensor-derived ocean data allows our onboard and shoreside teams to dynamically optimize speed and routing decisions with safety, cost, and emissions benefits,” said Jonathan Dowsett, Director of Fleet Performance at Star Bulk.

Looking ahead: How Sofar will continue to advance the state-of-the-art in data assimilation

At Sofar, our team is constantly refining, enhancing, and expanding our use of data assimilation. We are developing new techniques to help correct wave predictions at distances as great as hundreds of miles from a Spotter; we are rolling out a forecasting system initialized with all surface observations made by Spotter — waves, winds, sea surface temperature, and barometric pressure; and we are using Bristlemouth to add new sensors to Spotter to expand the types of data it can collect. As we continue to redefine what is possible in marine weather intelligence, data assimilation will be an essential tool keeping us at the cutting edge.

Interested in learning more about our data assimilation approach? Please contact our Sales team.