Airport weather stations
About 2,200 ASOS stations at U.S. airports report temperature, humidity, wind and rain every hour.
Forecast benchmark
How accurate is Precip's 10-day forecast?
Precip's hourly 10-day forecast compared with the world's leading forecast models, ECMWF IFS HRES and Google WeatherNext 3, at thousands of weather stations across the U.S.
- Precip
- IFS HRES
- WeatherNext 3
72of 75
Best score in 72 of 75 tests
- vs IFS HRES57/60
- vs WeatherNext 375/75
- forecast runs
- 29
- hours ahead
- 240
- station checks
- 28,052,948
Headline results
Get the world's best forecast two days early
IFS HRES, from the European Centre for Medium-Range Weather Forecasts, is the model most forecasters measure against. Precip's 10-day temperature forecast is as accurate as IFS HRES's 8-day forecast, and its humidity forecast is as accurate as IFS HRES's 6-day.
Temperature20%more accurate, days 1–10
Precip3.0 °F
IFS HRES3.7 °F
Relative humidity20%more accurate, days 1–10
Precip8.3 %
IFS HRES10.4 %
Wind speed19%more accurate, days 1–10
Precip2.8 mph
IFS HRES3.4 mph
Heavy rain2.9×better on downpours, days 1–2
Precip0.134
IFS HRES0.047
Accuracy is the average error over all 10 days at about 2,200 airport weather stations. Heavy-rain skill is the accuracy score for rain over 0.16 in/hr (4 mm/hr).
Methods
How the benchmark works
Every forecast is scored on the same runs, hours and places, against what the weather actually did.
- 1
Same forecast runs
Precip, IFS HRES and WeatherNext 3 were compared on the same 29 forecast runs, twice a day from late July to early October 2026, out to 10 days ahead.
- 2
Scored as served
Precip is scored exactly as the app and API serve it. IFS HRES and WeatherNext 3 are scored on their raw output, with nothing adjusted.
- 3
Checked three ways
Each hour is compared with weather station reports, rain gauges and NOAA's gridded analysis of what happened.
Three ways to measure what happened
Climate reference gauges
About 115 research-grade NOAA stations with precise rain gauges, the strictest check on rain.
NOAA analysis grid
NOAA's 2.5 km analysis of the weather that happened, which covers every point in the U.S., not only stations.
Reading the scores
Average error
How far off the forecast was, on average, in degrees, percent, miles per hour or inches of rain.
Accuracy score
How often a forecast called rain over a rate correctly, with missed rain and false alarms counted against it. 1 is perfect.
Brier score
The error of a chance-of-rain forecast. Calling 90% on a day it rains costs little; calling 10% costs a lot.
Results · 1
Temperature, humidity and wind
Precip downscales WeatherNext 3 to a 2.5 km grid, which restores the local detail of valleys, coastlines and cities. It's more accurate than both IFS HRES and raw WeatherNext 3 in every forecast window.
Average error by day ahead
IFS HRESWeatherNext 3Precip
At day 10:IFS HRES 5.97WeatherNext 3 4.82Precip 4.72
At day 10:IFS HRES 14.3WeatherNext 3 11.5Precip 11.0
At day 10:IFS HRES 4.13WeatherNext 3 3.49Precip 3.19
At day 10:IFS HRES 5.97WeatherNext 3 4.82Precip 4.72
All scores
| Score | IFS HRES | WeatherNext 3 | Precip |
|---|---|---|---|
| Temperature (°F)4,100,164 station hours | |||
| Days 1–2 (lower is better) | 2.83 | 2.54 | 2.30+9% |
| Days 3–5 (lower is better) | 3.32 | 2.89 | 2.65+8% |
| Days 6–10 (lower is better) | 4.69 | 3.88 | 3.69+5% |
| Days 1–10 (lower is better) | 3.71 | 3.17 | 2.95+7% |
| Relative humidity (%)4,080,817 station hours | |||
| Days 1–2 (lower is better) | 8.3 | 7.3 | 7.0+4% |
| Days 3–5 (lower is better) | 9.9 | 8.3 | 7.9+5% |
| Days 6–10 (lower is better) | 12.3 | 10.1 | 9.6+5% |
| Days 1–10 (lower is better) | 10.4 | 8.7 | 8.3+5% |
| Wind speed (mph)4,077,841 station hours | |||
| Days 1–2 (lower is better) | 3.08 | 2.91 | 2.53+13% |
| Days 3–5 (lower is better) | 3.31 | 3.05 | 2.70+11% |
| Days 6–10 (lower is better) | 3.78 | 3.34 | 3.01+10% |
| Days 1–10 (lower is better) | 3.43 | 3.12 | 2.77+11% |
Average error against airport weather stations, lower is better. Bold marks the best score in each row; the percentage shows how far ahead Precip is of the next-best forecast.
| Score | IFS HRES | WeatherNext 3 | Precip |
|---|---|---|---|
| Temperature (°F)2,417,664,576 grid-point hours | |||
| Days 1–2 (lower is better) | 2.77 | 2.45 | 2.00+18% |
| Days 3–5 (lower is better) | 3.31 | 2.81 | 2.39+15% |
| Days 6–10 (lower is better) | 4.81 | 3.90 | 3.60+8% |
| Days 1–10 (lower is better) | 3.74 | 3.13 | 2.75+12% |
| Relative humidity (%)2,417,664,576 grid-point hours | |||
| Days 1–2 (lower is better) | 7.8 | 6.4 | 5.8+9% |
| Days 3–5 (lower is better) | 9.4 | 7.5 | 6.8+9% |
| Days 6–10 (lower is better) | 12.5 | 9.7 | 9.0+7% |
| Days 1–10 (lower is better) | 10.2 | 8.1 | 7.4+9% |
| Wind speed (mph)2,417,664,576 grid-point hours | |||
| Days 1–2 (lower is better) | 2.90 | 2.64 | 2.13+19% |
| Days 3–5 (lower is better) | 3.15 | 2.79 | 2.32+17% |
| Days 6–10 (lower is better) | 3.65 | 3.12 | 2.71+13% |
| Days 1–10 (lower is better) | 3.28 | 2.88 | 2.42+16% |
Average error against noaa analysis grid, lower is better. Bold marks the best score in each row; the percentage shows how far ahead Precip is of the next-best forecast.
Results · 2
Chance of rain
The chance of at least a trace of rain in each hour. Precip blends WeatherNext 3, the ECMWF ensemble and, for the first two days, NOAA's HRRR. Its chance of rain beats WeatherNext 3's in every window, with 34% lower error in days 1–2 at climate reference gauges. IFS HRES is a single run that can only say yes or no, so it isn't scored here.
Probability error by day ahead
WeatherNext 3Precip
At day 10:WeatherNext 3 0.0470Precip 0.0370
At day 10:WeatherNext 3 0.0518Precip 0.0436
At day 10:WeatherNext 3 0.0591Precip 0.0536
At day 10:WeatherNext 3 0.0470Precip 0.0370
All scores
| Score | WeatherNext 3 | Precip |
|---|---|---|
| Brier score4,212,327 station hours | ||
| Days 1–2 (lower is better) | 0.0463 | 0.0285+38% |
| Days 3–5 (lower is better) | 0.0471 | 0.0334+29% |
| Days 6–10 (lower is better) | 0.0485 | 0.0376+22% |
| Days 1–10 (lower is better) | 0.0474 | 0.0337+29% |
| Days 1–10, skill over climate | -0.361 | 0.032 |
Chance of rain against airport weather stations. Brier score is the error of a probability, lower is better; skill over climate is higher-is-better, and 0 means no better than always forecasting the usual chance.
| Score | WeatherNext 3 | Precip |
|---|---|---|
| Brier score201,995 gauge hours | ||
| Days 1–2 (lower is better) | 0.0432 | 0.0284+34% |
| Days 3–5 (lower is better) | 0.0462 | 0.0354+23% |
| Days 6–10 (lower is better) | 0.0522 | 0.0434+17% |
| Days 1–10 (lower is better) | 0.0477 | 0.0367+23% |
| Days 1–10, skill over climate | -0.121 | 0.138 |
Chance of rain against climate reference gauges. Brier score is the error of a probability, lower is better; skill over climate is higher-is-better, and 0 means no better than always forecasting the usual chance.
| Score | WeatherNext 3 | Precip |
|---|---|---|
| Brier score2,417,664,558 grid-point hours | ||
| Days 1–2 (lower is better) | 0.0407 | 0.0319+22% |
| Days 3–5 (lower is better) | 0.0465 | 0.0406+13% |
| Days 6–10 (lower is better) | 0.0558 | 0.0505+9% |
| Days 1–10 (lower is better) | 0.0486 | 0.0422+13% |
| Days 1–10, skill over climate | 0.075 | 0.1972.6× |
Chance of rain against noaa analysis grid. Brier score is the error of a probability, lower is better; skill over climate is higher-is-better, and 0 means no better than always forecasting the usual chance.
Results · 3
Rain
Averaging an ensemble smears a storm's rain thinly over every place it might pass. Precip's blend keeps both where rain will fall and how hard, so it catches far more rain and downpours.
Accuracy score by day ahead
IFS HRESWeatherNext 3Precip
At day 10:IFS HRES 0.06WeatherNext 3 0.07Precip 0.08
At day 10:IFS HRES 0.02WeatherNext 3 0.00Precip 0.02
At day 10:IFS HRES 0.00WeatherNext 3 0.00Precip 0.00
At day 10:IFS HRES 0.06WeatherNext 3 0.07Precip 0.08
All scores
| Score | IFS HRES | WeatherNext 3 | Precip |
|---|---|---|---|
| Any rain15,794,126 station hours | |||
| Days 1–2 | 0.160 | 0.149 | 0.194+21% |
| Days 3–5 | 0.119 | 0.119 | 0.142+19% |
| Days 6–10 | 0.071 | 0.075 | 0.088+17% |
| Days 1–10 | 0.099 | 0.096 | 0.118+19% |
| Over 0.04 in/hr (1 mm/hr)Accuracy score, higher is better | |||
| Days 1–2 | 0.131 | 0.191 | 0.215+13% |
| Days 3–5 | 0.076 | 0.132 | 0.136+3% |
| Days 6–10 | 0.039 | 0.029 | 0.053+36% |
| Days 1–10 | 0.067 | 0.107 | 0.114+7% |
| Over 0.16 in/hr (4 mm/hr)Accuracy score, higher is better | |||
| Days 1–2 | 0.047 | 0.044 | 0.1342.9× |
| Days 3–5 | 0.020 | 0.009 | 0.0482.4× |
| Days 6–10 | 0.006 | 0.000 | 0.004−33% |
| Days 1–10 | 0.020 | 0.012 | 0.0532.6× |
Hourly rain accuracy score against airport weather stations, higher is better. Bold marks the best score in each row; the percentage or multiple shows how far ahead Precip is of the next-best forecast.
| Score | IFS HRES | WeatherNext 3 | Precip |
|---|---|---|---|
| Any rain757,314 gauge hours | |||
| Days 1–2 | 0.203 | 0.191 | 0.244+20% |
| Days 3–5 | 0.153 | 0.158 | 0.189+20% |
| Days 6–10 | 0.092 | 0.100 | 0.116+16% |
| Days 1–10 | 0.126 | 0.128 | 0.155+21% |
| Over 0.04 in/hr (1 mm/hr)Accuracy score, higher is better | |||
| Days 1–2 | 0.156 | 0.241 | 0.264+10% |
| Days 3–5 | 0.099 | 0.158 | 0.171+8% |
| Days 6–10 | 0.044 | 0.029 | 0.063+43% |
| Days 1–10 | 0.080 | 0.122 | 0.135+11% |
| Over 0.16 in/hr (4 mm/hr)Accuracy score, higher is better | |||
| Days 1–2 | 0.048 | 0.033 | 0.1423.0× |
| Days 3–5 | 0.020 | 0.011 | 0.0582.9× |
| Days 6–10 | 0.005 | 0.000 | 0.002−60% |
| Days 1–10 | 0.019 | 0.010 | 0.0522.7× |
Hourly rain accuracy score against climate reference gauges, higher is better. Bold marks the best score in each row; the percentage or multiple shows how far ahead Precip is of the next-best forecast.
| Score | IFS HRES | WeatherNext 3 | Precip |
|---|---|---|---|
| Any rain9,066,237,606 grid-point hours | |||
| Days 1–2 | 0.241 | 0.257 | 0.324+26% |
| Days 3–5 | 0.187 | 0.205 | 0.235+15% |
| Days 6–10 | 0.108 | 0.123 | 0.137+11% |
| Days 1–10 | 0.151 | 0.162 | 0.191+18% |
| Over 0.04 in/hr (1 mm/hr)Accuracy score, higher is better | |||
| Days 1–2 | 0.167 | 0.242 | 0.286+18% |
| Days 3–5 | 0.103 | 0.152 | 0.175+15% |
| Days 6–10 | 0.049 | 0.028 | 0.062+27% |
| Days 1–10 | 0.085 | 0.117 | 0.141+21% |
| Over 0.16 in/hr (4 mm/hr)Accuracy score, higher is better | |||
| Days 1–2 | 0.052 | 0.036 | 0.1553.0× |
| Days 3–5 | 0.024 | 0.009 | 0.047+96% |
| Days 6–10 | 0.006 | 0.000 | 0.004−33% |
| Days 1–10 | 0.022 | 0.010 | 0.0532.4× |
Hourly rain accuracy score against noaa analysis grid, higher is better. Bold marks the best score in each row; the percentage or multiple shows how far ahead Precip is of the next-best forecast.
Common questions
Frequently asked questions
How does Precip's 10-day forecast compare with ECMWF and Google?
Across 29 forecast runs checked at about 2,200 U.S. airport weather stations, 115 NOAA climate reference gauges, and NOAA's 2.5 km analysis grid, Precip had the best score in 72 of 75 tests against ECMWF IFS HRES and raw Google WeatherNext 3 (including all 75 of 75 against WeatherNext 3). Over all 10 days at airport stations, Precip's average error is 20% lower for temperature, 20% lower for relative humidity, and 19% lower for wind speed than IFS HRES.
Why is Precip more accurate than the raw models?
Global models run on coarse 9 to 10 km grids that smooth out valleys, coastlines, and cities, and averaging an ensemble smears heavy rain into light drizzle. Over the U.S., Precip sharpens WeatherNext 3 down to a 2.5 km terrain grid for temperature, humidity, and wind, and blends WeatherNext 3, the ECMWF ensemble, and NOAA's 3 km HRRR model for rain.
What does 'getting the forecast two days early' mean?
Forecast errors grow with each day ahead. At U.S. airport stations, Precip's temperature error at day 10 is the same as ECMWF IFS HRES's error before day 8, and its day-10 humidity error matches IFS HRES before day 6. You get the accuracy of a 8-day ECMWF forecast two full days earlier.
How much better is Precip on downpours and the chance of rain?
In the first two days, Precip's accuracy score for heavy rain over 0.16 in/hr (4 mm/hr) is 2.9× that of IFS HRES at airport stations and 3.0× at NOAA climate reference gauges (4.3× raw WeatherNext 3). For the hourly chance of rain, Precip has a lower Brier score than WeatherNext 3 in every window, with 34% lower error in days 1–2 at climate reference gauges.
Were Precip's models tested on data they trained on?
No. Every learned step was scored on forecast runs it never saw during training. The 2.5 km downscaling models were trained in two halves split at September 1, 2026 so each half was scored on the other, and the rain blend was fitted on eight separate runs that aren't part of the 29 runs scored here.
How does Precip do on next-hour rain alerts?
For the next 60 minutes, Precip runs its own 1 km radar nowcast updated every 5 minutes. In our live benchmark against Apple Weather, Google Weather, Tomorrow.io, and Rainbow.ai at NOAA rain gauges, Precip wins 37 of 38 tests and ranks #1 at every 5-minute step of the hour ahead. See the next-hour rain benchmark →
The full report
Read the whitepaper
The whitepaper has the full method, every score against all three references, and how Precip's forecast is built, written for meteorologists and researchers. Also see our live 60-minute rain nowcast benchmark comparing Precip with Apple Weather, Google Weather, Tomorrow.io, and Rainbow.ai.
Plan your week with a better forecast
See Precip's hourly 10-day forecast for your location in the Precip app.