MeteoFrontier forecast system

FRONTIER

The self-learning aviation forecast for western Canada's TAF airports. It reads every model and every observation, learns which to trust at each airport, and proves itself against the METARs every hour.

See it at an airport → Live verification
How to use this page · tap any ? for a full explanation
  • 01 What goes in: the six weather models and the observations FRONTIER reads every hour. Each box has a ? with what that source sees and misses.
  • 02 How it thinks: the loop (blend by proven skill → learned models → forecast and TAF → verification), the hourly reality check and the physics behind it.
  • 03 Always verified: shadow testing, promotion only if better, roll-back, and the live record.
  • 04 What you get: how the airport page uses FRONTIER (probabilities, ranges, TAF, hazards, AI badge).
  • 05 Timeline: what changed and when; the same markers are on the Verification page's Progress chart.
  • FRONTIER is guidance, not an official forecast. Words are in the glossary at the bottom.

01What goes in

Everything a forecaster would look at, every hour.

A good aviation forecast starts with the full picture. FRONTIER takes in every numerical model we run and the whole observing network around each airport, then keeps taking it in as new data arrives.

Weather models

  • RDPSCanadian regional
  • HRDPSCanadian high resolution
  • RRFSNOAA rapid refresh, 13 km
  • RAPNOAA rapid refresh, hourly
  • ECMWFEuropean global
  • UKMOMet Office global

Surface observations

  • METARs and SPECIsthe truth every forecast answers to
  • Upstream stationsweather on its way in
  • Automatic stationsbetween the airports
  • Road-weather stationsvalleys, passes and highways

Remote sensing

  • Satellitecloud, fog and moisture from above
  • Radarprecipitation and its movement
  • Lightningwhere storms are active now

Upper air and aircraft

  • Soundingsthe vertical profile of the atmosphere
  • Pilot reportsturbulence, icing and cloud tops, first hand

02How it thinks

A forecast that learns from every hour it gets right or wrong.

No single model is best everywhere. FRONTIER weighs them by what they have proven at each airport, checks the result against what is happening right now, and lets a stack of learned models refine it. Every verified hour feeds back into the system.

The FRONTIER loop: inputs, a skill-weighted blend, learned models, forecast and TAF, verification; verification feeds back into the blend weights and the learned models. retrain · shadow test · promote only if better skill weights re-scored every verified hour 01Inputsmodels + observations 02Weighted blendby proven skill 03Learned modelsnowcast · specialists 04Forecast + TAFranges · probabilities 05Verificationevery hour vs METARs Hourly reality check: the newest observations re-anchor steps 02 to 04 every hour. The FRONTIER loop: inputs, a skill-weighted blend, learned models, forecast and TAF, verification; verification feeds back into the blend weights and the learned models. retrain · promote if better weights re-scored hourly 01Inputsmodels + observations 02Weighted blendby proven skill 03Learned modelsnowcast · specialists 04Forecast + TAFranges · probabilities 05Verificationevery hour vs METARs
Schematic. Each verified hour flows back in two ways: the blend re-scores its models, and the learned models are retrained and must beat the live version before they replace it.
Blend

Weighted by proven skill

Every model is weighted by how well it has recently verified at that airport and lead time. A model going through a bad patch loses its say; when it improves, it earns it back.

Reality check

Re-anchored every hour

Each hour the forecast is checked against the newest METARs and SPECIs, satellite, radar, lightning, soundings, pilot reports and upstream stations, so the next hours start from what is actually happening.

Nowcaster

Reads the whole network at once

A neural network looks at the entire airport network together for the first 6 hours: ceiling, visibility, fog, precipitation and gusts. Weather moving from one airport to the next is something it sees.

Specialists

Persistence, gusts, ceiling and visibility

A METAR persistence model follows how ceiling and visibility have been trending at the airport. Specialist models for gusts, ceiling / visibility and METAR nowcasting are trained on years of forecasts against what really happened.

Learning core

Probabilities and ranges, not one number

A learning core turns everything into calibrated probabilities and ranges, and online learners update after every verified hour, so the system keeps adjusting between bigger retrains.

TAF

MANAIR-legal automated TAFs

The forecast is written up as a TAF that follows Canada's aviation forecast rules (MANAIR), with change groups and probabilities in the right form, ready to read like the real thing.

03Always verified

Nothing goes live until it wins on live weather.

Every forecast hour FRONTIER shows is recorded exactly as it was shown and scored against what happened. New models have to earn their place on hours they never trained on.

  1. Shadow

    A new or retrained model runs beside the live forecast. It is scored on every hour but never shown.

  2. Promote

    It goes live only after beating the current forecast on fresh hours, with the uncertainty taken into account.

  3. Watch, roll back

    Live, it keeps being scored. If it falls behind, the previous version comes back.

The live record

Loading…
Wrong IFR calls avoided–vs raw RAP, same hours
vs raw Blend–wrong IFR calls avoided
IFR success index–higher is better
IFR frequency–1.00 = forecast as often as observed

Wrong IFR calls avoided, first 6 hours

Share of misses plus false alarms FRONTIER avoided compared with each raw model on exactly the same airport, issue time and lead. Rolling 7 days; above 0 means fewer mistakes.

Single days swing with the weather: one fog event can move a score a long way. The 7-day figures pool all hours of the last 7 days.

Every source, every score on the verification page

04What you get

One airport page, built for the decision.

Choose FRONTIER at any of its airports and the whole airport page runs on it: the hourly forecast, the TAF and the hazard sections, with the raw models one tap away for comparison.

  • 01
    Hourly forecast with ranges and probabilitiesCeiling, visibility, wind, gusts and weather, with the chance of IFR and the likely spread rather than a single guess.
  • 02
    Automated TAFWritten to MANAIR, updated as the forecast changes.
  • 03
    Fog, turbulence, icing and convectionHazard sections focused on the hours that matter for operations.
  • 04
    Model comparisonEvery raw model side by side, so you can see where they agree and where FRONTIER leans.
  • 05
    SoundingsThe vertical profile over the airport for the hours ahead.
  • 06
    AI badgeValues shaped by a learned model are marked, so you always know where AI had a hand.
Airport page · FRONTIER

CYQRRegina

  1. METARs
  2. TAF
  3. FRONTIER analysis
  4. Hourly forecast
  5. Fog
  6. Turbulence · icing
  7. Convection
  8. Models · soundings
Open CYQR with FRONTIER →

05Innovation timeline

Moving fast, in the open.

What has gone live and what is coming next. The same timeline marks the progress chart on the verification page.

Live

  1. Loading…

Coming next

  1. Loading…

The honest part

FRONTIER is supplemental guidance from MeteoFrontier. It is not an official aviation forecast and not an issued NAV CANADA or ECCC TAF. Always check official aeronautical weather before any flight decision.

The numbers on this page are read live from our verification record and change as the weather does. Short windows are noisy, so read them as a trend, not a promise.

Glossary: words used on this page

See the site glossary.