Where the data comes from
Satellites, balloons, aircraft, ships, buoys and stations, and how data assimilation turns them into a starting point.
The observing systems
| Source | What it measures | Strengths and limits |
|---|---|---|
| Satellites (geostationary and polar orbiting) | Temperature and humidity through the atmosphere, clouds, winds from moving clouds, sea surface temperature | By far the largest number of observations and the only source over remote oceans; measurements are indirect and must be converted |
| Satellite radar instruments | Wind over the sea surface (scatterometers), wave height and sea level (altimeters) | Direct information about the ocean where no one measures; narrow swaths, gaps between passes |
| GNSS signals through the atmosphere | Temperature and humidity profiles from the bending of navigation-satellite signals | Accurate and unaffected by clouds; less certain in the lowest, moist layers |
| Weather balloons (radiosondes) | Wind, temperature, humidity and pressure up through the atmosphere | Accurate vertical profiles, but launched from a limited number of sites, mostly twice a day |
| Aircraft | Wind and temperature along flight routes and during climb and descent | Dense along air routes and around airports; few over the open ocean away from routes |
| Weather stations on land | Pressure, wind, temperature, humidity, rain at the surface | Many and frequent, but only where people live; station sites differ in quality |
| Ships | Pressure, wind, temperature, sea temperature, sometimes waves | Reports along shipping lanes; the reports from yachts and ships at sea fill real gaps |
| Moored and drifting buoys | Pressure, wind, waves, sea temperature | Accurate and continuous at their location; limited number |
| Radar | Rain and, in some cases, wind within range of the coast | Detailed over land and near coasts; used in some fine-scale models every hour |
| Ocean floats | Temperature and salinity from the surface down to about 2 km | The main feed for ocean models; little direct use for the atmosphere |
From observations to a starting point
Observations arrive at different times, in different places, with different errors, and none is complete. Data assimilation combines them with the background: the previous short forecast from the same model, which already knows how the atmosphere was moving. Each observation is weighted by how accurate it is expected to be, and so is the background. The result, the analysis, is a complete, physically consistent state for every box and layer of the grid.
- Large centres repeat this cycle every few hours: collect, assimilate, forecast, then use that forecast as the background for the next cycle.
- Some fine-scale models assimilate radar and surface data much more often, as often as every hour, to keep a close hold on the latest showers.
The role of historical data
A common belief is that a model finds similar weather in the past and copies it. A physics-based model does not do this: it solves the equations from the current state. Past data is still essential, in other ways:
- Reanalysis. Centres re-run the assimilation over past decades with one fixed, modern system, producing a consistent record of the atmosphere (ERA5 is a well-known example, reaching back to 1940). It is used to study climate and to test and improve models.
- AI models are trained on this record: see AI weather models.
- Correction and checking. Past forecasts are compared with what actually happened to find systematic errors, which can then be corrected: see Checking a forecast against reality.