What a Forecast Actually Is
A probability dressed as certainty, derived from a snapshot that no one quite gets right.
A probability dressed as certainty, derived from a snapshot that no one quite gets right.
The Measurement Problem That Starts Everything
A weather forecast is, at its core, a statement about a physical system's future state derived from its present one. That sentence sounds simple enough. It contains a trap.
To say where the atmosphere will be tomorrow, you need to know where it is today — not approximately, not in a few representative spots, but everywhere at once. Temperature, pressure, humidity, wind speed and direction: across the full volume of air from the surface to the stratosphere, over land and ocean alike, at the same moment. The World Meteorological Organization ↗ exists largely to coordinate that impossible task across nearly two hundred countries, because no single nation can measure its slice of the atmosphere in isolation and expect a useful result.
Once you have that snapshot — assembled from radiosondes, buoys, aircraft that report automatically, satellites and surface stations — you hand it to a physical model. The model is a numerical description of the atmosphere's governing equations: conservation of mass, momentum and energy, thermodynamics. Run those equations forward in time and the model produces a new state, then another, then another. The forecast is the chain of states the model predicts.
A probability dressed as certainty, derived from a snapshot that no one quite gets right.
What this means in practice is that a forecast is not an opinion. It is the logical consequence of a set of equations applied to a set of initial conditions. Robert FitzRoy understood the underlying ambition when he built the first storm warning service in the 1860s; the mathematics to realize it came later, through Vilhelm Bjerknes's 1904 programme and, eventually, Jule Charney's first successful numerical weather prediction ↗ on ENIAC in 1950. The leap from intuition to calculation took roughly a century.
What each step contributes
- The snapshotmeasuring the present state everywhere, simultaneously
- The modelrunning the atmosphere's governing equations forward in time
- The chain of stateseach forecast step depends on the last; error compounds
None of this makes the forecast certain. The initial snapshot is always incomplete. Errors in the starting state grow — Edward Lorenz showed in the 1960s that even tiny discrepancies amplify through the dynamics of the atmosphere itself. The forecast is therefore not a prediction so much as a best trajectory through a space of possible futures, and the further ahead it reaches, the wider that space becomes. When you check tomorrow's temperature on your phone, you are reading the end of a very long chain of measurement, mathematics, and managed uncertainty.
Elsewhere in Arithmetic
A forecast you calculate rather than guess. Everything in this sectionis about how a forecast is calculated.
- Six Weeks of Arithmetic for Six Hours of WeatherLongRichardson calculated a single six-hour forecast by hand in 1922.
- Bjerknes states the problemLongBefore anyone could calculate a forecast somebody had to say precisely what calculating one would mean: a set of equations, a set of initial measurements, and no meteorology in between.
- Why it came out wrongLongThe arithmetic was sound; the initial data was not smoothed, so the calculation amplified noise that was never in the atmosphere.
- Sixty-four thousand people with slide rulesMediumHis estimate of what it would take to keep pace with the weather, arranged in a hall like an orchestra, is the clearest picture of what a model does.