Weather Forecasting: How It Works, From Data to Models to Forecasters

Who watches the sky before any model does

A weather forecast starts with data, not a hunch. Every day thousands of ground stations measure temperature, pressure, wind and rainfall; radar tracks rain and storms in real time, the same kind of data behind violent events like the ones covered in how thunderstorms form or how tornadoes form. But the largest share of the data comes from above.

According to NOAA/NESDIS, 85% of the data used in numerical weather models comes from polar-orbiting satellites, which measure temperature and moisture across the whole atmosphere, including over oceans where there are no ground stations. The CrIS sensor, for instance, returns detailed profiles of temperature and moisture that the model uses as a starting point.

A second kind of measurement travels from the ground upward, radiosondes. These are small instruments attached to a balloon that rises through the atmosphere, measuring pressure, temperature, humidity and wind throughout the climb. In the United States, the National Weather Service network launches roughly 75,000 a year, with two launches a day per station, at 00 and 12 UTC, the same times the large global models, including the ECMWF’s, are initialized.

Practical example: a weather balloon launched at 00 UTC climbs at roughly 300 meters a minute and can pass 35,000 meters before bursting as the pressure drops. Only about 20% of balloons are ever recovered; the rest are lost, but their job is already done the moment they finish transmitting data on the way up.

From data to model

With thousands of measurements collected within a few hours, a computer builds a snapshot of the atmosphere’s state at that exact moment, temperature, pressure, humidity and wind at every point of a grid, from the surface up through the higher layers. Starting from that snapshot, a numerical model applies the equations of physics that describe how air moves, how it heats up and how water vapor condenses, and calculates how that snapshot will change hour by hour.

The catch is that the starting snapshot is never perfect. ECMWF’s own training material on predictability explains the idea with a simple example: in a system governed by entirely deterministic equations, a tiny difference at the starting point, in the example given a value of 0.4 versus 0.4001, leads to completely different outcomes after just a few dozen steps. The atmosphere behaves the same way, a small measurement error today becomes an enormous one ten days later. That’s why the ECMWF states that, in practice, detailed forecasts hold up for about a week.

To push past that limit without pretending to a precision that doesn’t exist, forecasting centers turn to ensembles. Instead of running a single forecast, the ECMWF runs 51 in parallel, each with slightly different starting conditions and a resolution of about 9 kilometers. If the 51 runs stay close together, uncertainty is low; if they spread out widely, the forecast flags itself as less reliable. That’s how forecasts out to 15 days work: not hour-by-hour detail, but a probability, say a 60% chance of rain in a given area.

The part the model can’t do on its own

Numerical models don’t come out ready for the evening news. A forecaster compares the output of several models and the ensemble results, spots where they agree and where they don’t, and decides which scenario to communicate, with what margin of doubt. It’s the same reason a heat wave, like the ones covered in heat waves: how they form, gets announced days in advance but with varying confidence depending on how well the models agree with each other.

The European body that produces much of this forecasting, the ECMWF (European Centre for Medium-Range Weather Forecasts), is an international organization running around the clock, with about 500 staff across its sites in Reading, in the United Kingdom, Bologna and Bonn. Understanding how the atmosphere exchanges energy with the surface also helps make sense of longer-term phenomena, like those described in the greenhouse effect: what it is and how it works: the same underlying physics, applied over timescales far longer than a weekly forecast.