How accurate are weather forecasts?

Forecast skill has improved dramatically and steadily, gaining roughly one day of accuracy per decade since the 1980s, so a modern five-day forecast is about as reliable as a one-day forecast was forty years ago. Current accuracy runs near 90 percent for five-day temperature and precipitation forecasts and around 80 percent at seven days. Beyond that, skill degrades quickly, and a ten-day forecast carries only modest advantage over simply predicting the historical average for that date. The gains came from satellite observation, vastly greater computing power, and ensemble forecasting, which runs a model many times with slightly varied starting conditions to map the range of plausible outcomes.

The hard ceiling is chaos, not effort. The atmosphere is a chaotic system in the technical sense that tiny differences in initial conditions amplify over time, the insight Edward Lorenz stumbled on in 1961 when a rounded input produced a wildly different forecast. That places a theoretical limit on deterministic prediction somewhere around two weeks. This is also why forecasts are increasingly expressed as probabilities, and why a 30 percent chance of rain is a meaningful statement rather than a hedge. Public perception of inaccuracy is partly a memory artifact: people remember the ruined picnic and forget the hundreds of correct forecasts they planned around without noticing.