AIRS and Model Reanalysis Data Sets

Many  of the quantities produced by AIRS Level 2 processing, such as temperature and water vapor vertical structure, are also available as weather and climate model output, and as model reanalyses. Many reanalysis systems assimilate AIRS radiances. Consequently, much of the information from AIRS is incorporated in reanalysis data. Also, reanalysis data are reported synoptically, on regular space-time grids. This simplifies their use, especially compared to AIRS Level 1 and Level 2 products.

The primary advantage of AIRS retrievals over reanalyses is that AIRS observations preserve instantaneous relationships between quantities like temperature, water vapor, and cloud. In reanalysis systems, these relationships are affected by both the physical models embedded in the systems and by assimilated observations. As a result, reanalysis features are often roughly correct, but with details misplaced in space and time. Also, the uncertainties introduced by the model and assimilation process are often difficult to quantify.

The primary disadvantage of the AIRS system is incomplete sampling because infrared sounding is limited to regions of thinner clouds. In addition, the AIRS retrieval algorithm has its own challenges in quantifying uncertainties in inferred geophysical quantities.

Reanalyses may directly or indirectly affect AIRS retrievals through their presence in the algorithm first guesses.  The AIRS-team algorithm uses a neural network trained on ECMWF for versions 6 and 7 (Milstein & Blackwell, 2016).  CLIMCAPS uses the MERRA2 reanalysis directly as its first guess for temperature, water vapor and ozone (Gelaro et al., 2017).

The tradeoffs between direct AIRS observations and model reanalyses are complex and beyond the scope of this document. Interested readers are encouraged to examine the peer-reviewed publications on the AIRS web page at https://airs.jpl.nasa.gov/ for additional information.