NOAA rain gauges
Each station's gauge measures the rain that fell in each hour. This check doesn't depend on radar.
Nowcast benchmark
How accurate is Precip's rain forecast?
Precip's next-hour rain forecast compared with Google, Apple, Tomorrow.io and Rainbow.ai at NOAA weather stations across the U.S., using each provider's forecasts exactly as they were issued.
- Precip
- Apple
- Tomorrow.io
- Rainbow.ai
37of 38
Best score in 37 of 38 tests
- vs Google & Apple19/19
- vs Tomorrow.io & Rainbow.ai18/19
- NOAA rain gauges
- 110
- hours scored
- 475
- 5-minute checks
- 635,743
Headline results
Precip dominates the scores for medium to heavy rain
Medium to heavy rain is the rain that disrupts plans, and it's the hardest to forecast. Precip is substantially better here.
vs Google+92%higher accuracy score
Precip0.309
Google0.161
vs Apple+121%higher accuracy score
Precip0.309
Apple0.140
vs Tomorrow.io+111%higher accuracy score
Precip0.259
Tomorrow.io0.123
vs Rainbow.ai+30%higher accuracy score
Precip0.259
Rainbow.ai0.200
The accuracy score is correct rain calls ÷ (correct calls + missed rain + false alarms) when the rain was over 8mm/hour. A score of 1 is perfect.
Methods
How the benchmark works
Most weather apps don't publish their past forecasts, so the only fair test is to record them as they happen.
- 1
Fetch each hour
Every hour, each provider's next-hour forecast is fetched from its public API at NOAA weather stations across the U.S.
- 2
Save as issued
Forecasts are saved exactly as issued, including Precip's, so nothing is rerun or adjusted after the fact.
- 3
Compare to gauges
Every provider is scored on the same hours and places, against the rain that fell at NOAA weather stations across the U.S.
Three ways to measure the rain that fell
Live NOAA radar
NOAA's real-time radar estimate of the rain at each station, every five minutes through the hour.
Gauge-adjusted radar
NOAA's official hourly radar estimate after it has been corrected with rain gauges.
Reading the scores
Accuracy score
How often a forecast called rain correctly, with missed rain and false alarms counted against it: correct calls ÷ (correct calls + missed rain + false alarms). 1 is perfect.
Average error
How far off the forecast amount was, on average, in inches.
Head to head · 1
Precip vs Google and Apple
475 hours at 110 NOAA weather stations, Aug 31, 2026 to Oct 12, 2026.
Accuracy over the hour ahead
Accuracy score at each five-minute step of the forecast, checked against NOAA radar at the stations. Each chart is a different rain rate. Hover or tap a chart for values.
GoogleApplePrecip
At 60 min:Google 0.52Apple 0.39Precip 0.61
At 60 min:Google 0.38Apple 0.33Precip 0.50
At 60 min:Google 0.31Apple 0.27Precip 0.40
At 60 min:Google 0.17Apple 0.16Precip 0.23
At 60 min:Google 0.09Apple 0.05Precip 0.08
At 60 min:Google 0.00Apple 0.00Precip 0.11
At 60 min:Google 0.52Apple 0.39Precip 0.61
All scores
Accuracy scores run from 0 to 1, higher is better. The five-minute rows average every step of the hour ahead, so a provider can lead here while trailing near the end of the hour in the charts. The hourly rows score each forecast's total for the hour.
| Score | Apple | Precip | |
|---|---|---|---|
| Against NOAA radar, every 5 minutes635,743 five-minute checks at 112 stations | |||
| Any rain | 0.540 | 0.451 | 0.683+26% |
| Over 0.04 in/hr (1 mm/hr) | 0.396 | 0.408 | 0.586+44% |
| Over 0.08 in/hr (2 mm/hr) | 0.328 | 0.360 | 0.493+37% |
| Over 0.16 in/hr (4 mm/hr) | 0.187 | 0.247 | 0.357+45% |
| Over 0.31 in/hr (8 mm/hr) | 0.106 | 0.139 | 0.247+78% |
| Over 0.63 in/hr (16 mm/hr) | 0.000 | 0.094 | 0.262+179% |
| Average error (in/hr) (lower is better) | 0.0046 | 0.0053 | 0.0033+30% |
| Against NOAA rain gauges, hourly49,991 gauge hours at 110 stations | |||
| Any rain | 0.478 | 0.388 | 0.549+15% |
| Over 0.04 in (1 mm) | 0.417 | 0.387 | 0.529+27% |
| Over 0.08 in (2 mm) | 0.366 | 0.377 | 0.491+30% |
| Over 0.16 in (4 mm) | 0.233 | 0.268 | 0.335+25% |
| Over 0.31 in (8 mm) | 0.161 | 0.140 | 0.309+92% |
| Average error (in) (lower is better) | 0.0042 | 0.0051 | 0.0032+24% |
| Against gauge-adjusted radar, hourly48,405 hourly checks at 112 stations | |||
| Any rain | 0.525 | 0.436 | 0.662+26% |
| Over 0.04 in (1 mm) | 0.419 | 0.432 | 0.576+33% |
| Over 0.08 in (2 mm) | 0.357 | 0.407 | 0.517+27% |
| Over 0.16 in (4 mm) | 0.270 | 0.296 | 0.386+30% |
| Over 0.31 in (8 mm) | 0.172 | 0.149 | 0.349+103% |
| Average error (in) (lower is better) | 0.0039 | 0.0046 | 0.0028+28% |
Precip, Google and Apple, scored on the same hours. Bold marks the best score in each row; the percentage shows how far ahead Precip is of the next-best provider.
Head to head · 2
Precip vs Tomorrow.io and Rainbow.ai
325 hours at 101 NOAA weather stations, Sep 28, 2026 to Oct 12, 2026.
Accuracy over the hour ahead
Accuracy score at each five-minute step of the forecast, checked against NOAA radar at the stations. Each chart is a different rain rate. Hover or tap a chart for values.
Tomorrow.ioRainbow.aiPrecip
At 60 min:Tomorrow.io 0.49Rainbow.ai 0.58Precip 0.61
At 60 min:Tomorrow.io 0.39Rainbow.ai 0.47Precip 0.53
At 60 min:Tomorrow.io 0.26Rainbow.ai 0.36Precip 0.40
At 60 min:Tomorrow.io 0.12Rainbow.ai 0.21Precip 0.18
At 60 min:Tomorrow.io 0.06Rainbow.ai 0.10Precip 0.04
At 60 min:Tomorrow.io 0.00Rainbow.ai 0.06Precip 0.07
At 60 min:Tomorrow.io 0.49Rainbow.ai 0.58Precip 0.61
All scores
Accuracy scores run from 0 to 1, higher is better. The five-minute rows average every step of the hour ahead, so a provider can lead here while trailing near the end of the hour in the charts. The hourly rows score each forecast's total for the hour.
| Score | Tomorrow.io | Rainbow.ai | Precip |
|---|---|---|---|
| Against NOAA radar, every 5 minutes111,378 five-minute checks at 105 stations | |||
| Any rain | 0.541 | 0.655 | 0.696+6% |
| Over 0.04 in/hr (1 mm/hr) | 0.438 | 0.529 | 0.601+14% |
| Over 0.08 in/hr (2 mm/hr) | 0.305 | 0.422 | 0.491+16% |
| Over 0.16 in/hr (4 mm/hr) | 0.175 | 0.266 | 0.340+28% |
| Over 0.31 in/hr (8 mm/hr) | 0.094 | 0.166 | 0.231+39% |
| Over 0.63 in/hr (16 mm/hr) | 0.050 | 0.056 | 0.245+338% |
| Average error (in/hr) (lower is better) | 0.0114 | 0.0091 | 0.0070+23% |
| Against NOAA rain gauges, hourly8,444 gauge hours at 101 stations | |||
| Any rain | 0.543 | 0.579 | 0.571−1% |
| Over 0.04 in (1 mm) | 0.488 | 0.552 | 0.571+3% |
| Over 0.08 in (2 mm) | 0.398 | 0.476 | 0.498+5% |
| Over 0.16 in (4 mm) | 0.223 | 0.329 | 0.346+5% |
| Over 0.31 in (8 mm) | 0.123 | 0.200 | 0.259+30% |
| Average error (in) (lower is better) | 0.0099 | 0.0085 | 0.0072+15% |
| Against gauge-adjusted radar, hourly8,053 hourly checks at 102 stations | |||
| Any rain | 0.590 | 0.653 | 0.663+2% |
| Over 0.04 in (1 mm) | 0.516 | 0.523 | 0.592+13% |
| Over 0.08 in (2 mm) | 0.376 | 0.451 | 0.488+8% |
| Over 0.16 in (4 mm) | 0.250 | 0.366 | 0.402+10% |
| Over 0.31 in (8 mm) | 0.151 | 0.212 | 0.275+30% |
| Average error (in) (lower is better) | 0.0080 | 0.0069 | 0.0056+19% |
Precip, Tomorrow.io and Rainbow.ai, scored on the same hours. Bold marks the best score in each row; the percentage shows how far ahead Precip is of the next-best provider.
Open models
Research models, rerun on 2024 storms
16 storms from 2024, a year Precip's model never trained on. This test is fixed and doesn't update.
Open research models can be rerun on past weather, which commercial apps can't. Precip, three pySTEPS models and NVIDIA's StormScope were run on the same storms and checked against NOAA radar across a 636 × 636 mile map.
Accuracy, over 0.31 in/hr (8 mm/hr), every 5 minutes+42%higher than the best other model, StormScope
Precip0.339
StormScope0.239
Accuracy, over 0.31 in (8 mm), hourly+39%higher than the best other model, pySTEPS extrapolation
Precip0.431
pySTEPS extrapolation0.309
Average error, every 5 minutes+23%lower error than the best other model, pySTEPS S-PROG
Precip0.011 in/hr
pySTEPS S-PROG0.015 in/hr
The full report
Read the whitepaper
The whitepaper has the full method, every score, and the research-model test, written for meteorologists and researchers. For questions or commercial use, get in touch.
Know when rain will start and stop
See the next hour of rain at your location in the Precip app.