Anu Dudhia

Abstract:

Line-by-line (LBL) methods of numerically solving the equations of radiative transfer can be inhibitingly slow. Operational trace gas retrieval schemes generally require much faster output than current LBL radiative transfer models can achieve. One option to speed up computation is to precalculate absorption cross sections for each absorbing gas on a fixed grid and interpolate. This work presents a general method for creating, compressing, and validating a set of individual look-up tables (LUTs) for the 11 most abundant trace gases to use the Reference Forward Model (RFM) to simulate radiances observed by the Infrared Atmospheric Sounding Interferometer (IASI) at a more operational pace. These LUTs allow the RFM to generate radiances more than 20 times faster than LBL mode and were rigorously validated for 80 different atmospheric scenarios chosen to represent variability indicative of Earth’s atmosphere. More than 99 % of all IASI simulated spectral channels had LUT interpolation errors of brightness temperature less than 0.02 K, several factors below the IASI noise level. Including a reduced spectral grid for radiative transfer sped up the computation by another factor of six at the expense of approximately doubling interpolation errors, still factors below IASI noise. Furthermore, a simple spectral compression scheme based upon linear interpolation is presented, which reduced the total LUT file size from 120 Gbytes to 5.6 Gbytes; a compression to just 4.4 % of the original. These LUTs are openly available for use by the scientific community, whether using the RFM or to be incorporated into any forward model.

Abstract:

The MIPAS spectrometer onboard the Envisat platform observed infrared emission from the Earth's limb between 2002 and 2012. It recorded high-resolution spectra during day and night, from pole to pole and between 6 and 70. km altitude in the nominal measurement mode or up to 170. km in special measurement modes, producing daily more than 1000 vertical profiles of various trace gases. The operational Level-2 data are processed by ESA/DLR but there exist three other, independent research Level-2 processors that are hosted by ISAC-CNR/University of Bologna, Oxford University, and KIT IMK/IAA. All four Level-2 processors rely on the same Level-1b data provided by ESA but their retrieval schemes differ. As part of ESA's Ozone Climate Change Initiative project, an intercomparison of the four MIPAS processors took place, in which vertical ozone profiles retrieved by these four processors from MIPAS nominal mode measurements were compared for 2007 and 2008. We present the results of this comparison exercise, which consisted of five parts: an information content study of the vertical averaging kernels, an intercomparison of zonal seasonal means and spreads, a determination of biases through comparison to ozonesonde and lidar measurements, a comparison to other satellite records (bias estimation and precision assessment with respect to ACE-FTS and Aura-MLS data), and a geophysical validation of the provided error bars using MIPAS-MIPAS collocations. The four processors demonstrate similar performance. All processors use the same Level-1b data from ESA, apply global fits, and use microwindows instead of the full spectrum. The main differences in the processing schemes include the choice of microwindows, the regularization approach, the treatment of negative retrieved values, and the cloud detection threshold. The different regularization schemes lead to a different trade-off between noise and resolution, but without a clear average advantage for any particular data set. The vertical resolution is typically 3-5. km and the single profile precision is about 2-3%. In the middle and upper stratosphere, at 25-45. km, all four MIPAS processors clearly show a high bias of 2 to 5% relative to all reference instruments. The similarity of the structure and magnitude of the bias among the MIPAS data sets indicates that the bias is most likely linked to the use of microwindows of the MIPAS AB band. The satellite intercomparisons show furthermore that for the KIT dataset, the onset of the high bias starts at a somewhat higher altitude (only above 35. km) than for the other three datasets. This is likely due to the more restrictive use of the AB band by the KIT processor, which comes at the cost of a coarser vertical resolution near the ozone volume mixing ratio (vmr) peak. In the troposphere, the Level-2 algorithms that suppress negative ozone values in the iterative retrieval process produce a larger positive bias than the algorithm that does not follow such a strategy. Our main conclusion is that the four MIPAS processors are more similar to each other than to any other reference instrument. This indicates that the observed biases are very likely instrument-related.