Inside the data

Precision weather, from the hour ahead to what actually fell

Know when rain starts and stops at your exact spot. Then come back after the storm for a measured record. This guide covers how Precip builds the nowcast, live precipitation, measured history, and multi-day forecast.

Observe What is happening
Nowcast The next 60 minutes
Measure What actually fell

Better observations

Start with what the atmosphere is doing now.

We use high resolution radar and quality-controlled gauge data to observe precipitation and model accumulation on the surface of the earth. We continually refine our data to be the most accurate and useful source for monitoring rainfall, snow, and other types of precipitation.

Most weather data sources are based on forecasts, not observations. That means they are often inaccurate, and don't reflect what actually happened. Precip is based on actual observations from radar, satellites, and rain gauges.

Accumulated precipitation for the same time period as estimated by ERA5-Land (left) and Precip (right).

One connected system

Observe, predict, then check the result.

Better forecasts begin with a better picture of what is happening right now. The same observation system that measures a storm gives our nowcast its starting point.

  1. 01 Observe

    Radar sees the storm in motion while quality-controlled gauges anchor it to what reached the ground.

  2. 02 Resolve

    Scientific models align the inputs in space and time, then estimate precipitation at 1 km resolution.

  3. 03 Nowcast

    Precip's in-house model carries that high-resolution picture forward through the next 60 minutes.

  4. 04 Verify

    After the storm, additional observations refine the measured record so the forecast can be checked.

The next 60 minutes

Better inputs make a better nowcast.

Precip's in-house nowcast starts from a high-resolution view of the storm as it is happening, then tracks where precipitation will move through the next hour. It is built to answer the questions that matter at a point: when rain starts, how its intensity changes, and when it stops.

1 kmspatial resolution
5 minforecast steps
60 minforecast horizon
See how it scores against Apple, Google, and others

Nowcast benchmark

Scored against the ground, not against other forecasts.

Every nowcast, ours included, is checked against independent rain gauges and airport sensors across the US. False alarms and misses both count against a forecast. Higher is better.

85% better than Apple Weather at calling rain in the next hour
52% fewer missed rain events than Google Weather
#1 at every lead time from 10 to 60 minutes, against the top nowcast providers
Rain in the next hour

Better than the weather app that came with your phone.

Next-hour forecasts checked against 112 research grade rain gauges, hour by hour.

Any rain (over 0.1 mm)

  • Precip .412
  • Google Weather .347
  • Apple Weather .223

Heavier rain (over 1 mm)

  • Precip .369
  • Apple Weather .241
  • Google Weather .230
Critical Success Index · 19,600 station-hours · Aug 25 – Sep 2, 2026
Accuracy by lead time

Ahead of the weather industry at every minute of the hour.

Will it be raining in 10 minutes? In 60? Checked every 10 minutes against 215 airport weather sensors.

.20 .25 .30 .35 .40 .45 10 min 20 min 30 min 40 min 50 min 60 min Precip Rainbow AI AccuWeather The Weather Company Vaisala
  • Precip .437 → .364
  • Rainbow AI .412 → .342
  • AccuWeather .343 → .249
  • The Weather Company .320 → .267
  • Vaisala .288 → .225
Critical Success Index by lead time · 18 days, Aug 15 – Sep 1, 2026 · Competitor scores from the public weatherindex.ai USA series
How each forecast gets it wrong

Apple cries wolf. Google lets you get drenched.

Every wrong call over the same 19,600 station-hours.

False alarms Rain was called, none came

  • Apple Weather 1,783
  • Precip 501
  • Google Weather 191

Apple Weather called rain 1,783 times when none came, more than three times as often as Precip.

Missed rain Rain came with no warning

  • Google Weather 346
  • Precip 167
  • Apple Weather 90

Google Weather missed 346 rain events, more than twice as many as Precip.

A rain event is one hour with more than 0.1 mm at the gauge.

Two separate tests. The phone apps were scored on hourly totals at rain gauges; the industry providers on rain-or-no-rain at airport sensors. Compare within a test, not across them.

Same rules for everyone. Precip was scored from the forecast it actually published at the time, against the same ground truth as every competitor. Full methodology whitepaper coming.

How the measured record is built

Why Precip has the best historical precipitation data

Radar and gauges

The Precip system uses observational data to track what actually happened, including 180 Doppler radar stations and a quality-controlled network of rain measurement stations.

Aligned in time and space

Raw data gets processed through a series of scientific models that convert raw information into easy to understand precipitation amounts. Part of that process includes aligning data from multiple sources into the correct hourly time periods for reliable analysis.

A refined ensemble

The final stage creates a proprietary ensemble of the various model outputs to deliver the highest quality blend of low latency and highly refined data, correctly projected for geospatial analysis.

Coverage and resolution

A local answer, with the grid made clear.

Precip data currently covers the United States, Canada, and some of Mexico, with other regions becoming available upon request.

Our rainfall measurements in the US quantify rainfall for an area of roughly one half mile squared, or 160 acres. In Canada it is approximately 1 square mile or 250 hectares. Rainfall can vary even within an area that small, so our models provide an average value for the area represented.

1 km United States

Live rate, nowcast, and measured precipitation

2.5 km Canada

Measured precipitation resolution

Historical precipitation benchmark

Why observations beat forecasts.

Most sources that show historical rain totals only show the most recent forecast amount before the actual rain could be measured. Precip uses radar and ground-truth observations as rain happens to create the most accurate possible rainfall totals.

60% more accurate than a NOAA forecast
275% better rainfall detection

Precip's estimates are the best available, but no rainfall estimate is perfect. Rain gauges themselves can provide inaccurate measurements. For example, wind turbulence can cause an average error rate of 2% to 10% when using gauges to measure rainfall. Read why measuring rain with gauges is hard.

The figures below are excerpts from a 2023 benchmarking study that compared Precip to the High Resolution Rapid Refresh forecast model output and the Integrated Surface Database ground observations. The report analyzes rainfall events at various accumulation thresholds against the corresponding gauge data.

Rainfall Mean Absolute Error benchmark
Mean Absolute Error MAE measures the average magnitude of errors in a set of estimates.
Rainfall Critical Success Index benchmark
Critical Success Index CSI measures correct events against the total of correct events, false alarms, and misses.

Data provenance

Third-party data sources

Precip combines our own models and processing with authoritative third-party observations and forecast data.

  • NOAA Doppler radar and ground observation data
  • Open-Meteo Global multi-day forecast data
  • Rainbow AI Global nowcast coverage outside Precip's radar domain

Know what's next. See what actually happened.

Put precision precipitation in your pocket.

Get Precip free on iPhone, Android, and the web.

Track rain at your exact address