Methane observations from TROPOMI suggest CAMS products underestimate methane across the UK

Satellite observations of greenhouse gases are used to constrain some global forecast models and are included in reanalysis models. The Copernicus Atmosphere Monitoring Service (CAMS) produces twice daily atmospheric composition forecasts using a data assimilation approach, with satellite observations forming part of the initial conditions in this forecast model. Global reanalysis of atmospheric composition, through data assimilation, is also produced by CAMS, where currently the fourth generation of the ECMWF global reanalysis (EAC4) is used.

The TROPOspheric Monitoring Instrument (TROPOMI) makes daily, high-resolution observations of atmospheric methane (CH₄) in the short-wave infrared band from space. The high-resolution of TROPOMI observations allows for urban and regional-scale CH₄ emissions evaluation but cloud coverage and data quality can limit the number of days with useful data. TROPOMI CH₄ observations are not included in the data assimilation of the CAMS global atmospheric composition forecast model nor in CAMS EAC4 and can therefore be used to independently evaluate atmospheric CH₄forecasts and EAC4 model estimates.

There are eight days in 2019-2020 with TROPOMI CH₄ observations that have sufficient coverage and pixel-density across the UK for comparison with CAMS daily CH₄ forecasts and EAC4 reanalysis values. We find average negative biases of ~55 ppb in CAMS forecasts and ~50 ppb in CAMS reanalysis compared to TROPOMI observations for these eight days across the UK. Differences could be due to i) the anthropogenic emissions used in the models; ii) biases in the stratosphere part of the CAMS models; iii) the TROPOMI retrieval algorithm, where biases could arise from the surface albedo and aerosol optical thickness values for certain pixels. To better understand and attribute the biases in CAMS we plan to explore parts of the CAMS model that relate to the stratospheric bias. Correct for the biases in CAMS yields average differences of around ±30 ppb across the UK suggesting additional discrepancies resulting from random error.