Tomorrow’s weather forecast is for an almost constant temperature just below freezing, accompanied by light snow. But this information on its own is of limited use, we also need to know if it is surprising or unusual - has it happened before? To make best use of weather forecasts and climate projections we need records of past weather to put them in context, and if we need good statistics for unusual events (storms, heatwaves, …) we need long records to provide enough observations.

One good source of such past weather data is the long reanalyses (for example 20CR and CERA-20C) which use weather-forecast models, constrained by observations of surface pressure to produce estimates of past weather back to at least 1900. In times and places where climate databases contain many weather observations, these reanalyses produce precise estimates of past weather,