reanalyses

Reanalyses are retrospective, gridded estimates of the atmosphere (and often the ocean and land states) produced by repeatedly assimilating observations into a fixed numerical weather prediction model over a historical period. The goal is to create a continuous, homogeneous record of the past climate and weather states by combining the best available observations with a consistent physical model and data assimilation system. Unlike routine forecasts, reanalyses do not forecast future conditions; they reconstruct past states using a single, coherent framework.

Methodology centers on data assimilation, which blends observations with a forecast model to produce a best estimate of the atmospheric state. Historical observations from satellites, radiosondes, weather stations, and other sensors are quality controlled and ingested. The assimilation system—using methods such as variational approaches or ensemble techniques—remains fixed for a given reanalysis product, ensuring internal consistency but potentially propagating model biases over time as the real climate evolves.

Notable reanalyses include the NCEP/NCAR Reanalysis, ERA-40, ERA-Interim, ERA5, JRA-55, MERRA and MERRA-2, CERA-20C, and the 20th Century Reanalysis. Modern products typically provide global fields on multiple vertical levels with finer horizontal resolution and improved physics, while still prioritizing a continuous historical record from roughly the mid-20th century onward.

Uses of reanalyses encompass climate monitoring and trend analysis, initialization and boundary conditions for regional models, validation and diagnostic studies, and methodological research in data assimilation. Limitations include inhomogeneities arising from evolving observing networks and model systems, residual biases, and uneven data coverage, especially in earlier decades or remote regions. Data are commonly available in netCDF or GRIB formats through major data portals for scientific use.