Weather.Co · Data sources

Engineering note · published 2026-06-10

Where Weather.Co gets its data — and why it now comes from five providers, not one.

Weather.Co launched on a single weather provider. As the app grew and the people relying on it included pilots making go/no-go calls, leaning on one source started to feel reckless. This page documents the new multi-source architecture, the providers behind it, and what (very little) you'll notice as a user.

5 forecast providers 1 aviation feed (unchanged) Automatic failover Richer forecasts, not just safer

01 Why this changed

One provider was a single point of failure.

Weather.Co has used Open-Meteo as its primary forecast source since launch. It's a genuinely great service: open data, well-maintained, generous free tier, fast. Nothing about that has changed and Open-Meteo remains a trusted source in the new architecture.

What changed is the calculus around relying on a single provider. Open-Meteo is run by a small team. So is OpenWeatherMap. So are most weather APIs. Terms shift, free tiers tighten, services occasionally have a bad afternoon. For a casual weather app that's fine. For an app where a pilot uses our forecast to decide whether to launch a Cessna, single-vendor dependency is the kind of risk you fix before it bites you, not after.

The fix is straightforward: more sources, with automatic failover. Weather.Co now draws from five forecast providers, prefers the highest-quality source available for each request, and falls back to the next-best when one is slow or unreachable. The app code doesn't change. The pilot-facing UI doesn't change. The dependency graph just got a lot more resilient.

02 The five forecast providers

Each one earns its place.

Weather.Co never shows you a "provider picker." The right source for a given request depends on where you are, what you're asking for, and what's reachable in the moment. Here's the lineup.

Apple WeatherKit weatherkit.apple.com

Primary for iOS users. Backed by Apple's acquisition of Dark Sky in 2020, now integrated as a first-party iOS framework. Uses Apple's own forecast models with privileged on-device access, which means lower latency and no API key visible to the user. High-quality short-range forecasts and the smoothest mobile integration of any source on this list.

WorldwideFirst-party iOS500K calls/mo free

PRIMARY (iOS)

Open-Meteo open-meteo.com

The original primary, still trusted. Open-source weather API aggregating multiple national weather models (ECMWF, ICON, GFS, more) and offering excellent hourly + 10-day forecasts. Where Weather.Co started, and where it returns whenever Apple WeatherKit isn't appropriate or isn't reachable.

WorldwideOpen dataNo key required

TRUSTED PRIMARY

National Weather Service api.weather.gov

Gold standard for US locations. The same data the FAA uses. Free, federal, no API key, no commercial terms to renegotiate. Used as a cross-check on US forecasts and as a failover when commercial providers aren't reachable. There's no better source for the United States than NWS, full stop.

US onlyPublic domainFederal data

FAILOVER · US

OpenWeatherMap openweathermap.org

Worldwide failover. One of the longest-running commercial weather APIs, broad international coverage, well-documented. Used when the primary sources are unreachable for non-US locations. Different data lineage from Open-Meteo, which is the whole point: a backup needs to fail at different times than the primary.

WorldwideAPI keyGenerous free tier

FAILOVER · WORLDWIDE

WeatherAPI.com weatherapi.com

Third-line worldwide. Independent commercial weather API with a different model mix again. Sits at the back of the chain to catch the rare scenario where Apple WeatherKit, Open-Meteo, and OpenWeatherMap are all unreachable for the same request. In practice this almost never happens, but "almost never" is exactly the situation a backup chain is built for.

WorldwideAPI keyFree tier

FAILOVER · LAST RESORT

03 How failover works

The right source, automatically.

For each forecast request the app tries providers in priority order and uses the first one that returns a healthy response inside the latency budget. If a provider is slow, errors out, or returns obviously bad data, the next one in the chain takes over. The pilot sees a forecast. They don't see which API delivered it.

1 Apple WeatherKit(if on iOS, primary for the geo)

↓ if unreachable or non-iOS

2 Open-Meteo(worldwide, open data)

↓ if unreachable

3 National Weather Service(US locations only)

↓ outside US or if unreachable

4 OpenWeatherMap(worldwide)

↓ if unreachable

5 WeatherAPI.com(last resort)

Cache headers and edge caching mean a single live forecast is usually being served from one provider for an entire region for a few minutes at a time, so this chain runs cold maybe a few times per minute at peak. Cost of running five providers instead of one: negligible at our scale, because the failovers are rare and most requests are served from cache.

04 Aviation data is different

METAR · TAF · AIRMETs · NOTAMs

Aviation-specific data still comes from one place, and always will.

All aviation data in Weather.Co — current observations (METAR), terminal forecasts (TAF), AIRMETs, SIGMETs, NOTAMs, PIREPs — comes directly from aviationweather.gov, run by the FAA's Aviation Weather Center. There is no second source for this data because there is no second source worth using: every alternative is reselling or repackaging the same NOAA feed anyway.

This page is about general-purpose forecast data (temperature, wind, precipitation, hourly + 10-day). For pilots, that's the "what will the weather be like tomorrow at my destination" layer. The METAR you see overlaid on the map is, and always has been, from FAA.

05 The unexpected upside

It turns out diversifying didn't just buy us safety. It made the forecast better.

This work was sold to ourselves as defensive engineering. Insurance against a provider having a bad afternoon. What we got instead was a forecast layer that's measurably richer than what any single source can produce on its own — and the resilience turned out to be the bonus.

Model consensus is a confidence signal

When Apple WeatherKit, Open-Meteo, and NWS all land on a 60% chance of rain tomorrow, that's a much stronger forecast than any one source saying 60% by itself. When they disagree — say, one says 30% and two say 70% — Weather.Co can surface that as low-confidence rather than pick a single number that's probably wrong. For a pilot making a go/no-go call, knowing the forecast is uncertain is often more useful than the false precision of a single point estimate.

Each provider has a specialty we can lean on

Apple WeatherKit (inherited from Dark Sky) gives minute-by-minute precipitation timing in supported regions — almost nobody else does this well. NWS publishes hourly point forecasts with the forecaster's own discussion attached, which can include nuance no raw model output captures. Open-Meteo internally aggregates multiple national models (ECMWF, ICON, GFS) and surfaces an ensemble. Pulling the best layer from the best source beats taking everything from one provider that's just average at all of it.

Spatial resolution stacks

A 13km grid from one model and a 2.5km grid from another are different tools. Combined, they let us show high-detail microclimate at a grass strip in a valley while still drawing the broad picture along the route. A single-source app has to pick one resolution. We don't.

Each forecast horizon plays to a source's strengths

Short-range nowcasting (next 60 minutes) is where Apple WeatherKit and high-resolution rapid-refresh models shine. Mid-range (3–7 days) is where ECMWF (via Open-Meteo) tends to win. Long-range (10+ days) is where commercial aggregators have invested the most R&D. Asking each provider to do its best job instead of all three is just better forecasting.

The take-home: failover was the design goal. The forecast getting tighter, the new confidence indicator, the better microclimate, the minute-precision precipitation — those were all happy accidents we discovered after the work was done. Different providers fail differently, and it turns out they also succeed differently. That's information you only get when you talk to all of them.

06 What changes for you

You should actually feel a few of these.

The original plan said "you won't notice anything." Then we shipped it and noticed quite a lot, in a good direction. If you've been a Weather.Co user for a while, here's what to expect:

Faster, tighter forecasts on iOS. Apple WeatherKit responds locally on-device, often in well under a second. The first load of the hourly view should feel snappier than it did.

Minute-by-minute precipitation timing in supported regions (US, much of Europe, parts of Asia) — the same level of "rain in 12 minutes, ending at :47" precision Dark Sky was loved for, now back via Apple WeatherKit.

A confidence indicator when our sources disagree on a forecast. If they all agree, you see a clean number. If they spread, you see the spread. For pilots, this is a decision aid — uncertainty is itself information.

Fewer "data unavailable" moments when a provider has a hiccup, because the next one in the chain just takes over.

Everything else — the UI, the layouts, the calculators, the pricing — is unchanged. Still free. Still free forever for CFIs and flight schools. If something does break in a way that feels new, please tell us so we can chase it: in-app feedback button, or email Ethan directly.

Architecture: Priority-ordered failover across 5 forecast providers
Aviation data: aviationweather.gov (FAA) — unchanged
Edge caching: Cloudflare Workers, 60s on forecast endpoints
User-visible upgrades: Minute precip · confidence indicator · faster first load
Pricing impact: None. Weather.Co is still free.
CFI / flight school pricing: Still free, still forever.