An evaluation of radiative transfer modelling errors in AMSU-A data

TitleAn evaluation of radiative transfer modelling errors in AMSU-A data
Publication TypeMiscellaneous
Year of Publication2016
AuthorsLupu, C, Geer, A, Bormann, N, English, S
Secondary TitleTechnical Memorandum

Systematic biases are observed between AMSU-A observations and simulations from numerical weather prediction models. They are not just a simple offset but instead exhibit a complex global pattern that appears to be related to the temperature profile. These errors could arise from inaccuracies in the radiative transfer model calculations used to simulate AMSU-A radiances from the model fields, from poorly-characterised aspects of the instrument calibration and spectral response function, or from errors in the forecast model state. Two hypotheses to explain and model the biases in AMSU-A are examined in this work. First, the optical depth of the atmosphere may not be sufficiently well-modelled due to unknown systematic errors in the the radiative transfer model. In order to correct this, an empirically-derived multiplicative scaling can be applied to the simulated optical depths (the so-called gamma-correction). Second, the reported locations of AMSU-A passbands may be incorrect, with empirically-derived corrections of up to 68 MHz having been suggested. Corrections to the radiative transfer modelling based on these hypotheses have been tested in the ECMWF data assimilation system and evaluated in terms of first-guess departure fits to independent observations and forecast scores, or in terms of residual biases to the AMSU-A data itself.

When AMSU-A channels 6-8 are simulated using corrected channel central frequencies (passband shifts), there are generally clear improvements in the fit between model and observations in these channels. However some degraded fits between model and observations are noticed for upper-tropospheric and lower-stratospheric AMSU-A channels, as well as for independent observations (e.g., ATMS, IASI). This also translates into medium-range forecast score degradations. The gamma-corrections generally do the best in terms of fits to AMSU-A channels 6-8 and unlike the passband shifts they do not degrade the fit between model and observations in upper-tropospheric and lower-stratospheric AMSU-A channels and the short-range forecasts in the lower stratosphere for independent observations. However, neither of the available hypotheses can fully explain the origin of the systematic biases between models and AMSU-A observations, and both show some degradations in medium-range forecasts.