Reanalysis data

Reanalysis data is a gridded historical record of the atmosphere and ocean, produced by running a single, fixed version of a modern weather model over decades of past observations, using the same data-assimilation process a live forecast uses, but applied retrospectively. Well-known examples include ERA5 from the European Centre for Medium-Range Weather Forecasts and the NOAA/NCEP reanalyses.

Because the model and assimilation method are held constant across the whole period, reanalysis produces a spatially and temporally consistent estimate of variables like temperature, wind, pressure and sea surface temperature everywhere on a grid, even where and when no observation was actually taken. This differs from a live numerical weather prediction run, which uses the best available model version at forecast time and changes as models are upgraded, making raw forecast archives inconsistent for long-term comparison.

Reanalysis is the standard input for training climate models, benchmarking climate model downscaling, calculating long-term climate normals, and driving impact models like crop or hydrology simulations that need a complete historical weather record with no gaps. The main pitfall is forgetting that reanalysis is a model output constrained by observations, not a direct measurement: it can be more accurate than sparse observations in data-poor regions, but it can also carry systematic model biases that a single ensemble forecasting re-run does not reveal without independent validation.