Advances in retrieving XCH 4 and XCO from Sentinel-5 Precursor: improvements in the scientific TROPOMI/WFMD algorithm

The TROPOspheric Monitoring Instrument (TROPOMI) on board the Sentinel-5 Precursor satellite enables the accurate determination of atmospheric methane ( CH 4 ) and carbon monoxide ( CO ) abundances at high spatial resolution and global daily sampling. Due to its wide swath and sampling, the global d...

Full description

Bibliographic Details
Published in:Atmospheric Measurement Techniques
Main Authors: O. Schneising, M. Buchwitz, J. Hachmeister, S. Vanselow, M. Reuter, M. Buschmann, H. Bovensmann, J. P. Burrows
Format: Article in Journal/Newspaper
Language:English
Published: Copernicus Publications 2023
Subjects:
Online Access:https://doi.org/10.5194/amt-16-669-2023
https://doaj.org/article/0191b90dc34b4c68914bcc77fca57a4b
Description
Summary:The TROPOspheric Monitoring Instrument (TROPOMI) on board the Sentinel-5 Precursor satellite enables the accurate determination of atmospheric methane ( CH 4 ) and carbon monoxide ( CO ) abundances at high spatial resolution and global daily sampling. Due to its wide swath and sampling, the global distribution of both gases can be determined in unprecedented detail. The scientific retrieval algorithm Weighting Function Modified Differential Optical Absorption Spectroscopy (WFMD) has proven valuable in simultaneously retrieving the atmospheric column-averaged dry-air mole fractions XCH 4 and XCO from TROPOMI's radiance measurements in the shortwave infrared (SWIR) spectral range. Here we present recent improvements of the algorithm which have been incorporated into the current version (v1.8) of the TROPOMI/WFMD product. This includes processing adjustments such as increasing the polynomial degree to 3 in the fitting procedure to better account for possible spectral albedo variations within the fitting window and updating the digital elevation model to minimise topography-related biases. In the post-processing, the machine-learning-based quality filter has been refined using additional data when training the random forest classifier to further reduce scenes with residual cloudiness that are incorrectly classified as good. In particular, the cloud filtering over the Arctic ocean is considerably improved. Furthermore, the machine learning calibration, addressing systematic errors due to simplifications in the forward model or instrumental issues, has been optimised. By including an additional feature associated with the fitted polynomial when training the corresponding random forest regressor, spectral albedo variations are better accounted for. To remove vertical stripes in the XCH 4 and XCO data, an efficient orbit-wise destriping filter based on combined wavelet–Fourier filtering has been implemented, while optimally preserving the original spatial trace gas features. The temporal coverage of the data records has ...