Assimilation of radiance observations from geostationary satellites: second year report

This report summarises the work carried out during the second year of this EUMETSAT Research Fellowship. All-sky radiance (ASR) data from the SEVIRI instruments on Meteosat-8 and Meteosat-11 remain key inputs to the ECMWF operational data assimilation system and as such great care is taken to ensure observation quality and continuity. In this reporting period we have had to manage a number of potentially disruptive real-time events. Firstly a large, sudden bias change in the 7.3µm water vapour channel on Meteosat-8 occurred in January 2019, which remains unexplained, but was sufficiently short-lived that the impact on operations was minimal. The second event was a scheduled decontamination of the SEVIRI instrument onboard Meteosat-11. Such exercises (from previous experience) are known to change the bias characteristics of the observed radiances, particularly in the 13.4µm channel, which is used for cloud initialisation, and this was again the case. The rapid adjustment of the ECMWF adaptive bias correction system (VarBC) ensured a smooth transition and there was no adverse impact on the assimilation system. The final issue was the dissemination of unphysical ASR radiance data during a sun-satelliteground-station co-linearity event. This highlighted some weaknesses in our real-time data screening processes and has now resulted in the implementation of additional upstream checks to protect against similar occurrences in the future. Efforts have continued to increase the exploitation and impact of SEVIRI ASR data, as well as clear sky radiance (CSR) data from other instruments in the ECMWF assimilation and forecasting system. One important area is to extend the assimilation of geostationary radiances to include window channels. These can convey important information about boundary layer humidity (in the tropics) and assist the vertical localisation of humidity from the sounding channels above. Results so far suggest that there is additional skill to be gained from window channels, but that the screening of cloud/aerosol as well as handling time varying biases is challenging. Another important development area is that of further improvements to the specification of observation errors for ASR data. Following on from the success of the implementation of inter-channel correlations for radiances from geostationary satellites (Burrows [2018]), focus has turned to the possibility of varying observation errors throughout the 4D-Var assimilation time window. Currently, all ASR/CSR observations are assigned the same observation error regardless of when they are observed. While the observation quality does not change, observation errors should include errors of representation and account for the growth of model error over the assimilation window, suggesting that data at the end of the window should be down-weighted (Howes et al. [2017]). However, a previous EUMETSAT sponsored study by McNally [2019] has demonstrated that observations at the end of the window are the most critical and influential, and suggests that down-weighting these would likely result in degraded forecast skill. This issue is clearly most acute for geostationary radiances as they report with high temporal frequency over the 4D window. A number of different time-varying observation error models have now been tested in dedicated assimilation experiments. While the results of these do demonstrate that the assimilation system is very sensitive to such modifications, none have emerged (so far) as a clear improvement and investigations in this area continue. In addition to maintaining and optimising the use of current geostationary radiance data from EUMETSAT, a significant amount of effort has been directed towards preparing for the assimilation of radiance data from the the future MTG-IRS mission. Some of the enhancements that will be delivered with MTGIRS can actually be investigated using existing data from a number of different sources. Firstly, The Research Report No. 51 1 Geostationary radiance assimilation Chinese geostationary satellite FY-4A carries an experimental hyperspectral infrared instrument (called GIIRS) and data from this instrument became available in 2019. An initial assessment of the radiances has been performed and uncovered a number of anomalies including horizontal banding, spectral shift, inter-dwell variability, intra-dwell variability and spurious geometric features. However, despite these issues, early indications suggest that, with highly selective screening, enough good quality GIIRS data can be be recovered to allow sensible prototype assimilation trials to begin. Secondly, MTG-IRS will deliver radiance data with very high time and spatial resolution and we can begin to gain some experience of how best to exploit these particular characteristics of the future IRS by working with existing radiance observations from the GOES-16 Advanced Baseline Imager (albeit with low spectral resolution non-hyperspectral data). Unfortunately, to be in a position to perform data assimilation experiments, a number of significant data issues had to be investigated including spurious image striping and cloud/aerosol contamination in the window channels. However, these have now been largely resolved by NOAA/NESDIS/STAR, and GOES-16 half-hourly data have been activated as part of the ECMWF operational assimilation system. Experiments are now underway to explore if radiance data with a repeat frequency in the range 10-20 minutes (IRS will have 15 minute repeat) can be successfully digested by the 4D-Var assimilation system.