The ‘route’ of all evil or why grib files are barking mad

In 1999 I was racing my yacht Barking Mad, and at the same time I was deeply immersed in the technology sector. The internet and email were just beginning their transition from curiosity to necessity, and it was already clear, at least to me, that digital innovation was going to fundamentally change navigation, particularly offshore racing. The Royal Ocean Racing Club had long taken a progressive view on the adoption of new ideas within its Special Regulations, ensuring it remained at the forefront of offshore, more precisely, ocean racing. Having recently joined the committee, I saw first-hand that the question of when, and how, to adopt emerging technology was not theoretical — it was ongoing and often discussed.

Decca had been allowed in offshore racing through the late 1970s and into the 1980s, but the reality was that the equipment was still relatively bulky and expensive, and only really began to find its way onto race yachts in a meaningful way by the mid-1980s. Even then, while it could be extremely effective within its coverage areas, particularly around the UK, it was far from perfect, with propagation errors, lane ambiguity, and signal inconsistencies all part of the navigational landscape. It was a useful tool, but not one you could ever entirely rely on without question.

GPS, by contrast, was just beginning to emerge during the late 1980s, but it wasn’t until the early 1990s that it became a realistic and then rapidly dominant solution for offshore racing. Even then, for military reasons, the civilian signal was deliberately degraded through what was known as Selective Availability, meaning the position you received was intentionally offset. Despite that, the trajectory was obvious.

By the time companies like Garmin began producing relatively affordable handheld units in the early 1990s, we had one onboard Barking Mad, and you could already sense that navigation was moving into an entirely different phase. As the technology became smaller, more robust, and more accessible, I decided to take the next step and put a laptop — what we would now call a notebook — on the boat.

That might not sound remarkable now, but at the time it was a little crazy. So I had to make it useful. The Sony laptop, the smallest I could find, was connected via a waterproof cable using an RS232 interface to a ruggedised military display originally designed for armoured vehicles. It was handheld and waterproof, basically and realistically bulletproof, which meant we could actually get usable information into the cockpit. By any sensible standard it was completely over-engineered, but it felt like the future and crucially, it worked.

The objective was simple: bring navigational data and tactical decision-making into the cockpit using what was, at the time, genuinely cutting-edge offshore routing software, Seatrack, developed by Peter Scofield, a pioneer in this space and, importantly, a very capable yachtsman and a good friend.

The process required to make it useful offshore, however, was anything but simple.

Weather data came in the form of GRIB files, typically sourced from systems aligned with the World Meteorological Organisation, but in reality these had to be downloaded before the race, often via dial-up connections, as bandwidth was extremely limited and offshore access was, for most of us, impractical. The files were small, compressed, and by necessity relatively low resolution.

It is worth remembering that GRIB files were never designed for sailors. They were created for meteorological exchange between forecasting centres, aviation systems, and military applications. We were, in effect, borrowing them and trying to interpret them in an offshore racing context they were never intended for.

Once downloaded, usually before leaving the dock or back at the office, the file would be fed into Seatrack. Alongside it, I would input boat polars that I had built up myself from recorded sailing data. Even then, Seatrack had the ability to refine and update those polars dynamically, which at the time felt extraordinary. The laptop was connected into the onboard instruments via the NMEA system using an RS232 interface, pulling data from the B&G system, so in many respects it functioned in a similar way to a modern integrated setup, just significantly slower and far less refined.

The system would then combine the GRIB data with tidal information and calculate an optimal route. The underlying mathematics, the routing algorithm itself, was not fundamentally different from what is used today. It evaluated multiple possible paths, balancing wind, tide, and boat performance, and selected the fastest theoretical option.

Even then, it was iterative. You could recalculate as often as you liked, although in reality there was little point doing so more frequently than the resolution of the GRIB itself, which by modern standards was extremely coarse. And of course, once you were offshore, you couldn’t realistically update the data, so wind inputs were often adjusted manually using forecasts from the Met Office broadcast over longwave radio — a proper “Sailing By” moment.

So yes, we had a routing solution. Just.

And if I’m honest, it was great fun.

But I never fully trusted it.

Alongside the software, I would still sit down with a chart, a pencil, a rubber, and the nautical almanac, working through the route in a more traditional way. Not because I didn’t believe in the technology, but because I understood even then that the output was only ever as good as the assumptions behind it.

But for all the extraordinary progress, one thing has remained constant: the technology may have evolved, but uncertainty hasn’t gone anywhere. It’s just been repackaged.

HAVE WE REPLACED INSTINCT WITH PRECISION AND MISTAKEN IT FOR UNDERSTANDING?

So even today there is a moment in almost every offshore race where the tone and chat on board subtly shifts, normally as you lose mobile phone signal.

Someone is below, studying a screen filled with wind arrows, stepping forward through the forecast, watching the pattern evolve as if the race has already been decided somewhere inside the machine. On deck, just before they lose mobile signals, all kinds of strategies are being looked at from an array of weather and routing apps.

Interestingly what you are looking at is not actually the weather you are about to sail into. It is just a model’s interpretation of how the atmosphere might behave, built on incomplete observations, processed through a finite grid, and projected forward under assumptions that are necessarily imperfect. Could a finger in the air perhaps give you more real-time info?

And once you see that clearly, the role of the GRIB and the routing it creates changes.

It stops being something you follow.

And becomes something you question.

WHAT YOU’RE ACTUALLY LOOKING AT

At its core, a GRIB file is not something a sailor reads in any direct sense. It is a compressed binary dataset of gridded atmospheric values — wind, pressure, temperature — produced by models such as ECMWF, NOAA, and the Met Office.

On its own, it is unusable. Of course you need an app to run them on a PC, Mac, iPad running on Android, iOS or Microsoft platforms; new route programs such as PredictWind offer even easier access, but we’ll discuss this further later.

What you see on screen is a visualisation of that data — wind arrows, colour fields, smooth gradients — rendered by software and in the last 12 months or less enhanced by AI.

That visualisation matters a lot because although it says the model is updated every four or six hours, in reality the real information from ships, weather buoys and met stations around the world is already hours out of date, so everything you see is from a data statistical model burning and turning continuously.

Because the atmosphere is continuous, and the model is not.

The model also exists on a grid, often 10–25 km in spacing, and everything between those points is interpolated as well. The smooth flow you see is not what the model “knows”; it is what the software helps you believe.

Even before the forecast begins, the model constructs its starting point through data assimilation, blending observations with prior forecasts — many years of data and growing daily — to produce what is deliberately described as the best estimate.

Not reality.

An estimate.

WHERE THE MODELS START TO GUESS

Out at sea, observations are sparse. Satellites dominate. Buoys and ships provide only fragments and all this has to be fed into the different companies and government agencies who provide this information, not for sailors but for commercial ships, aviation, farmers and energy providers. We are just an add-on user.

So what you are looking at offshore is not measured wind.

It is a calculated field built on incomplete data.

That leads to a simple but important truth:

The further offshore you are, the more the model is guessing.

WHEN MODELS DISAGREE — THAT’S WHERE THE RACE BEGINS

Most navigators look at several models — ECMWF, GFS, UKMO, ICON, ARPEGE and others.

The instinct is to decide which one is right.

But the better question is:

Why are they different?

Because when models diverge, they are exposing uncertainty in the atmosphere.

And that is where the opportunity lies.

Races are rarely won in the parts of the forecast everyone agrees on.

They are won in the parts they don’t.

WHY GRIBS GET IT WRONG — FROM A SAILOR'S PERSPECTIVE

Every offshore sailor has seen it.

The GRIB says 12 knots.

You get 25.

The instinct is to say the model is wrong.

It might not be, for many reasons. It is simply operating at a scale that does not resolve squalls, rapid fronts, local acceleration or sea-state interaction. They are better at getting it right at aircraft height, but at the surface or 0–100m where yacht racing happens, land, cliffs and fronts all get in the way.

The model gives you accurate structure at a macro view perhaps.

Reality provides the detail and experience adds the secret sauce.

Over the years this has led to all kinds of fixes, some of which became the norm.

ADDING A PERCENT TO THE WIND SPEED

The idea of adding 5–20% to GRIB wind speeds is common, but it is not a rule. It is an old fix that was a rough blend of masthead versus surface wind difference, model underestimation offshore and practical safety margin.

Routing tools such as Expedition, Adrena, and PredictWind allow this to be handled properly by adjusting your polars as well as the GRIB, but not all routing software does it the same way. All the models have improved dramatically as the amount of data increases day by day, so every event can be mapped back to a similar event to predict what might happen.

So the real aim is not to just inflate the forecast.

It is to align the model with reality as your boat experiences it.

WHERE ROUTING SOFTWARE FITS — AND WHERE IT MISLEADS

A GRIB file only becomes a useful tactical tool when combined with polars, tide, and constraints inside routing software. Much the same as traditional navigation to find your optimum course made good.

The software then produces something far more seductive than a pencil mark on a chart: you get a digital course on screen you should follow that gets you to the mark the quickest.

An optimum route.

Clean. Precise. Logical.

And dangerously convincing.

Because at that moment, many crews stop interpreting weather.

And start following instructions.

But routing is not a decision.

It is a suggestion based on assumptions.

Then you run the routing for all the relevant models in your area and suddenly you get three or four or even more beautifully colourful optimal courses.

So you are back to square one — and I suppose that’s the biggest dilemma.

THE REAL SKILL OFFSHORE

The best sailors do not trust a single model.

They do not follow routing blindly.

They compare, question, and relate what they see on screen to what they feel on deck.

And most importantly:

They recognise when the model is wrong before everyone else does and they act.

FINAL THOUGHT

So a GRIB file is a simplified mathematical opinion of the future atmosphere.

It is not reality. It is not strategy. And it is not an answer.

The tools have improved.

The problem has not.

Races are still won by people making better decisions with imperfect information.

Sometimes being ‘Barking Mad’ still helps!

This is just an intro into further articles where I look at all the latest software, test them offshore and draw further conclusions.