D7 Reanalysis and Derivative Products

Topic

Meteorological reanalyses are historical reconstructions of the atmosphere's state, produced by combining—via a data assimilation model—absolutely all available observations from any given moment in the past (radiosondes, satellites, buoys, aircraft, surface stations) with a numerical prediction model; this process employs an assimilation technique that adjusts the model's state to match these observations across an entire time window rather than at a single instant. One of the most widely used meteorological reanalyses spans from 1940 to the present, featuring a horizontal resolution of approximately 31 kilometers and 137 vertical levels, and generates a new dataset every hour; other meteorological centers produce their own global reanalyses with specifications that are similar but not identical.

In practice, meteorological reanalyses are the only source of atmospheric data that is simultaneously homogeneous, global, and three-dimensional; consequently, they serve as the foundation for studying past climate, weather extremes, and the quality of climate models themselves. However, these reanalyses do carry biases stemming from changes in the observing system over time: for instance, satellites began to be incorporated into the available observations toward the end of the 1970s, and the design of radiosondes themselves evolved over the years, meaning the mix of observations feeding the reanalysis is not exactly the same at the beginning of the period as it is at the end.

Based on this same principle, there are global precipitation products that combine satellite estimates with ground-based rain gauges to achieve complete coverage of precipitation across the planet. Some of these global precipitation products cover a period as long as that of meteorological reanalyses—from the late 1970s to the present; Others covered more limited periods—spanning little more than a decade—before being replaced by newer products that began operating from the year 2000 onwards.

Internally, the variational assimilation technique used to generate both meteorological reanalyses and their derived products works by minimizing a cost function that weighs two factors simultaneously: how well the final result fits the available observations, and how much that result deviates from the background state held by the numerical prediction model prior to incorporating those observations. The outcome of this minimization process is known as the "analysis"—the best possible estimate of the atmosphere's initial state, serving as the starting point for any subsequent forecast.

Since neither meteorological reanalyses nor their derived products are free from bias, quantifying the magnitude of this error has become just as important as producing the reanalysis itself. Different meteorological reanalyses—generated by various centers using distinct combinations of observations—are compared against one another; the resulting discrepancies serve as a measure of the uncertainty inherent in each dataset. Thus, meteorological reanalyses remain the most comprehensive historical reconstructions of the atmosphere available, provided the user understands the biases and uncertainties associated with them.

ID:606