Exploration of the use of short‐wave infrared radiances in weather forecasts model Part <scp>II</scp> : Data assimilation and forecast impact assessment

Abstract The standalone non‐local thermodynamic equilibrium (NLTE) bias correction scheme, developed by Li et al. ( Journal of Geophysical Research , 125, 2020 and Quarterly Journal of the Royal Meteorological Society , 10.1002/qj.5020 , 2025), was tested within a data assimilation framework. The bias correction scheme utilizes a look‐up table and a linear regression method. Together with quality control procedures, this enables the assimilation of short‐wave infrared (SWIR) radiances. Initial evaluations revealed that significant residual biases remained in the polar region (latitude greater than 60° N/S) after applying the developed schemes to SWIR channels. This resulted in SWIR radiances that were not assimilated in the polar region. In the non‐polar region (latitude less than 60° N/S), there remained slight day/night discrepancies after applying the bias correction scheme. In this case, these biases arose from using different forecast models to determine the regression coefficients for the bias correction scheme and those employed by the data assimilation system. To address these biases, two additional quality control procedures were introduced. Six‐week experiments were conducted, assimilating 140 stratospheric SWIR channels over the non‐polar regions. Results were compared against a control, without SWIR assimilation. Three scenarios were examined: the assimilation of SWIR radiances, the assimilation of only solar‐day SWIR radiances and the assimilation of only solar‐night SWIR radiances. Observations were classified solar‐day radiances when the solar zenith angle was less than 90°, and solar‐night when it exceeded 120°. Overall, results were promising. The inclusion of SWIR channels did not degrade existing observations, and no significant temperature biases were found at levels where SWIR channels were sensitive over the course of the experiment. SWIR radiances benefited forecasts by improving 250‐ and 500‐hPa geopotential height anomaly correlations and reducing forecast temperature root‐mean‐squared error beyond 48 hours between 50 hPa and surface. Assimilating only solar‐day and only solar‐night SWIR radiances separately yielded better results than their combined assimilation.