Investigating the Sensitivity of NCEP's GFS to Potential Detector Heterogeneity of Hyperspectral Infrared Sounders

Infrared sounding instruments use large arrays of detectors to make simultaneous observations in multiple field‐of‐view (FOV). For Numerical Weather Prediction (NWP) models to assimilate multiple detector observations, FOV responsivities should be well matched so that forecasts are not degraded due to dis‐similar FOV properties. In this pseudo‐observing system simulation experiment (OSSE) study, hyperspectral infrared observations are assimilated to assess the impact of FOV heterogeneity on analyses and forecasts. The Cross‐track Infrared Sounder (CrIS) was selected as the representative instrument because it is an operational instrument with the lowest noise. The use of CrIS to this study is only limited to its detector configuration, orbital parameters, magnitudes of the instrument noise, and detector‐to‐detector bias. Perturbations in radiance to simulate detector heterogeneity are modeled to be wavelength and detector dependent. Results show that detector heterogeneity can be detrimental to analyses and forecasts. The data thinning routine relies on the brightness temperature of a surface sensitive channel to determine which profiles are used. If the FOV heterogeneity results in an increase in the brightness temperature, the FOV will be preferred, and this extraneous information will be used. The variational bias correction scheme does not account for biases associated with different FOVs. Unremoved biases are assimilated as “information” resulting in a biased analysis. Biases propagate and grow with time as the global data assimilation system cycles. The bias correction coefficients of all satellite observations are solved simultaneously with the atmospheric state analysis. In this way, the detector biases can influence other satellite observations. Key Points IR hyperspectral detector heterogeneity influences selection of clear profiles as NCEP's current thinning algorithm prefers warmer pixels Bias correction schemes poorly characterize biases introduced by detector heterogeneity Unremoved biases are assimilated as “information” resulting in a biased analyses and degraded forecasts