Weather research

How we forecast the fog that cancels flights.

Since June 2026 Havgard has run a live laboratory on North Sea visibility: every forecast is frozen when it is issued and graded against what actually happened. This is the short version.

The problem

Low visibility cancels offshore helicopter flights. The operational question is simple and expensive: will this helideck be below the limit in 6, 12 or 24 hours? A miss strands a crew; a false alarm delays a departure. Everything we build is shaped by that trade-off.

Underneath sits a scientific question: how does offshore sea fog form, and how far ahead can it be forecast from public weather data? We are answering it with real measurements, every day.

The North Sea Visibility Laboratory

Meteorological sensor mast in coastal fog
Every measured visibility report is paired with the forecast that was issued before it.

Since June 2026 we have paired the measured visibility from around 30 offshore helidecks, from the North Sea to the Barents Sea, with the forecast that was issued before it. Forecasts are frozen at issue time, so nothing is ever recomputed in hindsight, and an improvement only ships after it has proved itself on data the model has not seen.

26,000+verified observationsand growing every day
~30offshore stationsNorth Sea · Norwegian Sea · Barents
+120 hforecast horizonfrom the next hour to five days
8,500+verified forecastsserved to users and graded

What we have found

  • Offshore fog is not the fog in the textbooks. Our data overturned the standard land-based recipe, and that correction shaped everything we have built since.
  • The whole basin matters. A region-wide fog period announces itself in advance, and our forecasts read that signal.
  • Small details decide the outcome. We found and worked around systematic biases in the weather feeds that separate a good forecast from a useless one.
  • Thick fog can be told apart from light fog, so a forecast can say how bad it will be as well as whether it will happen.
  • Even sensors are not always right. We built checks that catch measurements which contradict themselves.

In our own verification, about nine in ten basin-wide fog outbreaks, the ones that actually cancel flights, were flagged six hours ahead, and the region-wide fog switch was spotted about a day in advance.

Honest by design

  • Frozen forecasts. A forecast is locked with a version stamp the moment it is issued and can never be edited afterwards.
  • Changes must earn their place. An improvement ships only after it beats the current engine on data it has not seen.
  • Every forecast says how sure it is. Confidence is part of the answer, not a footnote.
  • We check ourselves against the professionals on the same events, and we grade every forecast we serve.

Helicopter Trigger Index

HTI: where a helicopter can trigger its own lightning strike.

In winter, a helicopter flying through the wrong cloud can trigger a lightning strike itself. Our Helicopter Trigger Index maps where that risk exists, built to match the Norwegian Meteorological Institute's own operational product and checked against it.

A continuous risk map

Rendered as a smooth map with the flight corridors overlaid, not a grid of blocks.

Wide coverage

From the Barents Sea to the northern UK sector.

Made for winter

Active through the season when helicopter-triggered lightning is a real risk.

The workflow

From raw model fields to a pilot's phone.

Eight steps, every one of them logged. A chain of custody from the first data point to the final grade.

  1. Ingest

    Weather models, marine data and official forecasts are collected for every station, several times a day.

  2. Read the region

    Each forecast is placed in its regional context: what the surrounding stations and the wider basin are doing.

  3. Predict

    Our engine turns those inputs into a visibility range, a fog probability and the reasons behind the call.

  4. Freeze

    The forecast is locked with a version stamp the moment it is issued. It can never be edited afterwards.

  5. Sharpen the short term

    For the next few hours, fresh measurements sharpen the call.

  6. Serve

    Apps receive a ready-to-show forecast with a confidence level and plain-language reasons, in seven languages.

  7. Render

    The apps only display what the server says, so the engine can improve without app updates.

  8. Verify

    Every issued forecast is graded against what was measured, so the model never grades its own homework.

Måløy under a storm sky

North Sea Helicopter Radar

Measured, estimated, and honest about which is which.

A real-time offshore helicopter radar fused with the rig map. Offshore coverage is a mix of real plots and estimated positions, and the map always shows which is which: bright is real, faded is estimated.

It powers the radar in Helio and the radar module in our operator dashboards.

Working with us

Let's work on it together.

We are open to collaborating with meteorological institutes, operators and researchers. If you have a weather problem where the decision matters, we would like to hear about it.

For meteorological institutes

A live, verified testbed of around 30 offshore stations, and the infrastructure to trial new ideas prospectively. We would like to see how our approach holds up on your fields.

For operators

Station-specific verification, your own helideck sensors as extra ground truth, and the Helideck Forecast API inside your operations dashboard.

Helideck Forecast API

For researchers

Open to joint studies and co-authored papers.

Get in touch

Want this level of rigour applied to your weather problem?

Visibility, lightning, sea state, icing: the method transfers. Tell us the decision you need to make.

Request a study