I tested six weather APIs at 12 airports over one week, comparing their forecasts with the recorded weather. Here’s how they did and what to check before choosing one.
of its next-day temperature forecasts were within 3.6°F of the recorded temperature.
Forecasts were off by 1.85°F on average.
Forecasts within 3.6°F Higher is better
12 airports · September 21–27, 2026.
See results by airportPirate Weather’s next-day temperature forecasts were within 3.6°F of the recorded temperature 88.3% of the time. Open-Meteo was within the same range 72.1% of the time, on the same 4,000 forecast hours.
Compare temperature ranges →At JFK and LAX, Visual Crossing had the lowest next-day temperature error: 1.41°F, compared with 1.89°F for Pirate Weather.
Compare all six at JFK + LAX →Visual Crossing was within 4.47 mph on 84.9% of next-day wind forecasts, compared with 81.4% for Pirate Weather. Their average errors were almost tied.
Compare wind forecasts →Check the average error or how often forecasts were within a set range. Pick an airport, measurement, and lead time to see which APIs were closest.
Temperature · 12–36 hours ahead · 4,000 matched forecast hours per API · 2,144 airport observations.
Bars show mean absolute error (MAE), the average difference from the recorded weather. Lower is better.
Pirate Weather had the lowest average error, at 1.85°F. Apple Weather was next at 2.19°F.
Each API is scored on the same 4,000 forecasts, matched from 4,032 possible comparisons.
| Provider | Mean error (MAE) | RMSE | Bias | Within 3.6°F | 90% within (°F) |
|---|---|---|---|---|---|
| Pirate Weather | 1.85 | 2.45 | +0.12 | 88.3% | 3.96 |
| Apple Weather (WeatherKit) | 2.19 | 2.82 | -0.18 | 81.9% | 4.43 |
| Visual Crossing | 2.21 | 2.92 | -0.61 | 83.7% | 4.68 |
| MET Norway | 2.43 | 3.15 | -0.74 | 77.6% | 5.22 |
| Open-Meteo | 2.68 | 3.41 | -1.08 | 72.1% | 5.58 |
“Within” counts forecasts whose absolute error was at most 2°C for temperature or 2 m/s for wind, converted to your units. “90% within” is the error limit that includes at least 90% of the matched forecasts. RMSE gives more weight to larger errors. A negative bias means cooler or less windy than recorded; a positive bias means warmer or windier.
Each row compares next-day forecasts with the same airport observations. Pirate Weather had the lowest temperature error at seven airports. Three other providers came out ahead at the remaining five. Choose an airport above to see its scores.
Pirate Weather had the lowest average temperature error in all three time ranges. Each range compares the same forecast hours across all five providers.
Apple Weather had the lowest error for 24-hour rain totals at JFK: 0.108 in on average. That’s 26.6% lower error than MET Norway, the next closest source. Apple also had the lowest error when only rainy days were counted.
We compared the same seven measured 24-hour periods at JFK, including four wet days and three dry days. Each API had 14 saved forecasts, made about 12 or 24 hours before the period began. These are daily totals, a different measure from the hourly Heathrow comparison below.
The reference was NOAA’s ASOS rain reports. Daily amounts were checked against complete hourly reports. The first and last forecast hours were adjusted to match the measurement period, assuming rain was evenly spread within those two hours. Missing amounts weren’t counted as dry weather. NWS’s saved forecasts contained rain chances but no amounts.
Visual Crossing had the lowest error during rainy hours at Heathrow. All five sources averaged about 0.044 in of error per rainy hour.
| Source | Rainy hours | Including dry hours |
|---|---|---|
| Visual Crossing | 0.044 in | 0.005 in |
| Pirate Weather | 0.044 in | 0.005 in |
| Open-Meteo | 0.044 in | 0.005 in |
| MET Norway | 0.044 in | 0.005 in |
| Apple Weather | 0.044 in | 0.005 in |
Dry hours make the overall errors look much smaller. Visual Crossing and MET Norway tied for the lowest overall error. All five sources underestimated the rain during the matched rainy hours.
We checked saved hourly amounts against NOAA’s measured precipitation at Heathrow, matching the same one-hour periods for every source. The comparison includes 139 recorded hours, 16 with measurable rain. Missing and trace amounts were left out. Amounts are liquid-equivalent precipitation, not rain chances or minute-by-minute timing.
NWS’s saved forecasts contained rain chances but no amounts, so it isn’t included here.
Compare forecast accuracy, features, pricing, and terms to choose an API for your app.
Global forecasts with a Dark Sky-compatible API
Had the lowest average next-day temperature error across all 12 airports: 1.85°F. Its Dark Sky-compatible format may make it easier to adapt an existing integration.
Check results for your region, how far ahead the hourly forecast goes, and usage limits.
Plans and documentation ↗Forecasts and historical weather in one API
Had the lowest next-day temperature error across JFK and LAX: 1.41°F. Its Timeline API combines forecasts and weather history in one endpoint.
Check how your queries are billed.
Plans and documentation ↗Native Apple tools and a REST API
Averaged 2.19°F next-day temperature error across all 12 airports, close to Visual Crossing. You can use Apple’s frameworks or call WeatherKit’s REST API from other platforms.
Requires Apple Developer Program membership. Check request signing, usage limits, and attribution requirements.
Access and documentation ↗Public forecasts without a paid API key
Had the lowest next-day temperature error at Toronto: 1.46°F. Forecasts are available through its Locationforecast API.
Identify your app, follow the cache headers, and include attribution. Some later forecasts are spaced three hours apart; I compared them as provided.
Usage guidelines ↗An API with several weather models to choose from
Had the lowest next-day temperature error at Heathrow: 1.74°F. The comparison used the default forecast endpoint.
The open-access service is for non-commercial use. For a business app, check commercial plans and attribution requirements.
Plans and model documentation ↗Free U.S. forecasts, alerts, and observations
Averaged 2.00°F next-day temperature error across JFK and LAX. Its API also includes U.S. alerts and observations.
I tested NWS at the two U.S. airports. Select the JFK + LAX comparison above to compare all six APIs on the same forecasts.
API documentation ↗The median is the middle response time. About 95% of collected requests finished within the time in the last column.
| Provider | Forecasts collected | Median | 95th percentile |
|---|---|---|---|
| Visual Crossing | 168 | 0.098 s | 0.553 s |
| Pirate Weather | 168 | 0.178 s | 3.022 s |
| Open-Meteo | 167 | 0.216 s | 1.758 s |
| MET Norway | 168 | 0.416 s | 0.713 s |
| Apple Weather (WeatherKit) | 168 | 0.113 s | 0.267 s |
| NWS | 28 | 0.735 s | 7.290 s |
We checked saved temperature and wind forecasts against NOAA airport observations, using the same forecast hours for every API. “Next day” means 12–36 hours ahead.