Assimilation of AIRS Data at the Met Office A.D. Collard and R.W. Saunders Met Office,Bracknell,UK 00/XXXX 1 Contents 00/XXXX Overview of AIRS Processing at the Met Office Cloud Detection Channel Selection Future Work 2 AIRS processing at the Met Office From NESDIS BUFR ingest To other European NWP centres 1DVar retrieval 3DVar assimilation of radiances 00/XXXX 3 Pre-processing Store incoming data on MetDB Monitoring stats radiances, retrievals O-B no. of obs and q/c flags Cray T3E supercomputer Current Status of AIRS Processing at the Met Office Simulated AIRS data is being received from NESDIS (M. Goldberg) and is being stored in our MetDB system. – 281 Channels, Reduced Spatial Sampling – BUFR format – Surface information added at Met Office before storage – Additional pre-processing steps may be performed, e.g., EOF based cloud detection (Lee, Smith and Taylor, 2001) 00/XXXX 4 Current Status of AIRS Processing at the Met Office (contd.) A 1DVar is done as further pre-processing before the assimilation stage. This includes: – Bias Correction – Cloud Detection – Channel Selection – Other QC – Production of Monitoring Stats 00/XXXX 5 Some 1DVar & Monitoring Details Uses RTTOV7 for RT (can also use Gastropod) – See talks by Matricardi et al. and Sherlock et al. 00/XXXX Newtonian or Marquardt-Levenberg Minimisation Variational Bias Correction (to be implemented) 6 Example O-B Plot 00/XXXX 7 Variational Cloud Detection (English, Eyre & Smith, 1999) Attempt to determine the probability of having cloud in the field of view given the observed radiances and the NWP background profile J Ln{P(cloud ¦ y obs , x b )} 12 (y )T {H(x b ) T BH(x b ) R}1 ( y ) Const. y y obs y (xb ) Clouds are flagged when J exceeds a certain threshold 00/XXXX 8 Cloud Detection Example 00/XXXX 9 Channel Selection (following Rodgers, 1996) Method: Choose those channels with the biggest impact on DFS. 1) Starting with A0=B test which channel will most improve the DFS 2) Update Ai using that channel 3) Repeat until a sufficient number of channels have been selected Rodgers speeds this process up by noting that, for diagonal (O+F), on adding a new channel, i, to the retrieval, the solution error covariance is changed from Ai-1 to Ai thus: A i A i 1 I hi A i 1hi T (hi is the Jacobian for channel i) 00/XXXX 10 1 A i 1hi T hi DFS for different channel selections 00/XXXX 11 “NESDIS 281” vs “Optimal Channels” 00/XXXX 12 Channel Selection Caveats 00/XXXX Channel Selections are based on different criteria Ozone is not considered here “Optimal” channel selection assumes a given B-matrix (and assumes it’s correct!) Channel selection is profile dependent 13 Conclusions and Future Work Simulated AIRS data is being ingested and preprocessed at the Met Office Software for cloud detection, quality control and the production of monitoring information is in place. Work continues on visualisation of monitoring data. 00/XXXX 14 Work on variational assimilation continues Conclusions and Future Work (contd.) 00/XXXX Channel selections issues should be explored further (after receipt of real data?) Studies on assimilation of cloudy radiances to be made. 15 00/XXXX 16