# How Forecasts Are Made

A miniature landscape showing how observations, data assimilation, a numerical model and a 51\-member ensemble turn into a weather forecast

By Kylie

Topic: Earth &amp; Geography

[Open the interactive Jig](https://ask.jig.com/j/5yqFdTqLXLm5)

## Transcript

### 1. Watching the sky

A forecast starts by measuring the air as it is right now\. On this miniature coast, a weather station records temperature, humidity, pressure and wind\. A buoy and a ship measure over the sea, a radar scans for rain and a satellite watches from far above\. Each blinking dot is one observation on its way to the forecasters\. Drag to turn the model\.

Forecast centres such as ECMWF take in about 60 million observations every 6 hours, from land stations, ships, buoys, aircraft, balloons and satellites\.

### 2. Launch a weather balloon

The ground is only the bottom of the story, so the weather also has to be measured higher up\. Twice a day a latex balloon lifts a small radiosonde about 30 km up\. On the way it radios back temperature, humidity and pressure, and its drift shows the wind\. The balloon swells as the air thins until it bursts, and then a parachute brings the sonde down\. Launch one and watch the dots cool as it climbs\.

About 90 US sites launch together at 00Z and 12Z\. A balloon about 1\.5 m wide at launch grows to 6–7\.5 m before it bursts at 27–30 km, around two hours later\.

### 3. Radar and satellites

Instruments that scan from a distance fill the gaps between stations\. A Doppler radar sends out pulses and listens for echoes from raindrops, which shows where rain is falling and how it moves\. Geostationary satellites photograph the clouds every few minutes\. Switch between the two to see what each one can measure\.

A NEXRAD dish is 8\.5 m across inside a 12 m dome, with a beam about 1° wide\. GOES satellites orbit 35,786 km above the equator, so they watch the same half of the planet all the time\.

### 4. Blend into a starting picture

Computers do not use the observations as they come in\. In data assimilation, each measurement corrects a first guess, which is the last forecast moved on to now\. The result is a complete 3D grid of boxes over the whole planet, each holding values for wind, temperature, pressure and moisture\. Watch the dots settle into the grid and the boxes change from the first guess to the analysis\.

There are always gaps, so the first guess fills in wherever there are no measurements\. That is why a good analysis depends on the forecast before it\.

### 5. Step the physics forward

The model then applies the physical laws for moving air, heat and water to each box\. It takes a step of a few minutes, then another, until it reaches days ahead\. Drag the slider to run the forecast forward and watch the storm cross the grid towards the coast\.

ECMWF’s global model uses boxes about 9 km wide with 137 layers up through the atmosphere\. Supercomputers repeat this calculation several times a day\.

### 6. Run it 51 times

Small errors in the starting picture grow as the forecast runs\. Forecasters therefore run the model again and again, each time from a slightly different start\. The tracks stay close together at first and spread out later\. Where they agree, the forecast is reliable\. Where they spread, it is uncertain\.

An ensemble has one control run \(the bright track\) and 50 runs with small changes to the starting picture\. Even tiny measurement errors can change where a storm goes days later\.

### 7. From runs to a forecast

Count the runs that bring rain to the town: here 20 of 51 do, so the forecast says about a 40% chance of rain\. Forecasters check the computer output against their experience and local knowledge, then turn it into the forecast you see on your phone\.

A 40% chance of rain does not mean rain for 40% of the day\. It means that on days like this, rain falls at your location about 4 times in 10\.
