1. Slide 1 NWS Activities in Satellite Data Assimilation Andrew Collard, Daryl Kleist, John Derber, Li Bi, Lidia Cucurull, David Groff, Xu Li, Emily Liu, Quanhua Liu, Xiujuan Su, Cathy Thomas, Russ Treadon, P aul van Delst, Yanqiu Zhu 1 JCSDA Workshop May 21 2014

2. Slide 2 Plan of the Presentation Upgrades for QY14 implementation: Resolution Changes Radiative Transfer (CRTM) Enhanced Bias Correction AMSU-A Bug Fix SSMIS Atmospheric Motion Vectors MetOp-B GRAS and GPSRO QC MetOp-B IASI and Minor QC changes to satellite data Upgrades for Future Implementation Near Sea Surface Temperature Cloudy Radiance Slide 2 JCSDA Workshop May 21 2014

3. Slide 3 EnKF/DA-Resolution components in T1534 GFS Package (Daryl Kleist, Russ Treadon) JCSDA Workshop May 21 2014

4. Slide 4 Resolution for Data Assimilation Current operations Analysis done on linear grid that corresponds to the model truncation: T574 – 1152 x 576 However, the 80 member ensemble that prescribes 75% of the solution is at T254 – 512 x 256. Proposed configuration for T1534 SL (3072 x 1536) GFS Increase 80 member EnKF resolution to T574 (SL) Compute GSI-hybrid analysis increment at the ensemble resolution (T574 – 1152 x 576) Recall, 75% of the solution is driven by information at this (or coarser) resolution already Increment is transformed to wave space and added to full resolution background Note that the O-F is computed at truncated resolution (need to fix this in future) Static B has MPI bottleneck making it difficult to perform analyses on high resolution grids (being worked on) JCSDA Workshop May 21 2014

5. Slide 5 CRTM Upgrade (Dave Groff, Paul van Delst, Quanhua Liu) JCSDA Workshop May 21 2014

6. Slide 6 FASTEM-5 May 21 2014 JCSDA Workshop

7. Slide 7 Bias Correction Upgrades (Yanqiu Zhu) JCSDA Workshop May 21 2014

8. Slide 8 The control variables and are estimated by minimizing (Derber et al., 1991, Derber and Wu, 1998) is radiative transfer model Air-mass dependent in GSI Separate scan-angle dependent Original radiance bias correction scheme A two-step procedure where is pre-specified parameter is predictor term with predictor coeff. May 21 2014 JCSDA Workshop

9. Slide 9 Enhancements to the Radiance Bias Correction Scheme  Combine the scan angle and air-mass bias components inside the GSI variational framework, simplify the operational job suite  With the modified pre-conditioning, the pre-specified parameter for each predictor is removed  Add new emissivity sensitivity bias predictor to handle large land-sea differences  Automatically initialize the bias of any new radiance data  Automatically detect any new/missing/recovery of radiance data  Quickly capture any changes in the data and the data assimilation system, preferred scheme for re-analysis projects  Faster GSI minimization convergence 9

10. Slide 10 is computed outside GSI, a running average of OMF The predictors of air-mass are computed inside the GSI Enhanced scheme: a one-step procedure Original scheme: a two-step procedure Variational angle bias correction, is computed inside the GSI along with other predictor terms satang file no longer needed Special attention to the initialization of bias coefficients: using quality-controlled data from previous run; or mode of all data May 21 2014 JCSDA Workshop

11. Slide 11 Change of the Preconditioning Hessian w.r.t. predictor coeff. of the cost function Modified block-diagonal preconditioning is set to be the inverse of ( Dee 2004 ) Current preconditionerModified preconditioner For simplicity, only diagonal elements of are considered at each analysis cycle May 21 2014 JCSDA Workshop

12. Slide 12 Convergence comparison 00Z July 15, 2012 May 21 2014 JCSDA Workshop

13. Slide 13 Variance vs. Observation number A big drop of obs number prompts a jump of variance AIRS CH22 13  Adaptive background error variance for bias predictor coef.: set to be the estimate of analysis error variance of the coef. from previous analysis cycle. More & higher quality data smaller variance Enhancements to the Radiance Bias Correction Scheme (cont.)

14. Slide 14 Enhancements to the Radiance Bias Correction Scheme (cont.)  Provide a convenient and efficient way, via the newly added passive channel bias correction capability, to obtain the bias of any new satellite data that are not used but monitored for preparation for future use, such as the radiance data from NPP satellite 14 Passive channel bias correction: at the end of analysis minimizing the functional where

15. Slide 15 HGT 500hpa SH Anomaly Correlation May 21 2014 JCSDA Workshop

16. Slide 16 New Emissivity-based bias predictor JCSDA Workshop Old CRTM New CRTM (Others are various configurations using emissivity BC) New CRTM has larger land-sea differences which were being aliased into the CLW bias-correction for microwave instruments. This was solved by the addition of a new bias predictor applied over land to remove the systematic land-sea difference in radiance innovation. May 21 2014

17. Slide 17 AMSU-A Bug Fix (Emily Liu) JCSDA Workshop May 21 2014

18. Slide 18 AMSU-A Bug Fix The AMSU-A Cloud Liquid Water retrieval code returns CLW=0 when the sea surface temperature is below 0C. As CLW is a predictor used in bias correction of AMSU-A radiances, this means that uncorrected cloudy fields of view are being assimilated near the ice-edge where the water is below 0C but still not frozen (the sea freezes around -1.9C), are therefore systematically warm-biased. JCSDA Workshop May 21 2014

19. Slide 19 Impact on Analysis JCSDA Workshop May 21 2014

20. Slide 20 Impact on 48h Forecast JCSDA Workshop May 21 2014

21. Slide 21 Impact on SH 850hPa Wind JCSDA Workshop Control With Bug Fix Control with FASTEM-1 May 21 2014

22. Slide 22 SSMIS (Emily Liu) JCSDA Workshop May 21 2014

23. Slide 23 23 Jan Global Dsc Asc July Asc Dsc Global O-B (no bc) SSMIS F18 Bias Characteristics Jan Ascending NodeDescending Node July O-B (no bc) vs. Latitudes Ascending NodeDescending Node Jan July O-B (no bc) Ascending and descending biases are significant (summer season is worse) Ascending pass is warmer than the descending pass Latitudinal biases are significant; locations of max/min biases change with season T (K) Latitude T (K) May 21 2014 JCSDA Workshop

24. Slide 24 Variational Bias Correction Scheme Variational bias correction provides an automatic inter-calibration of the observing system in the context of the forecast model, producing bias corrections that improve the consistency of the information entering the analysis For instruments other than SSMIS the bias correction is calculated using five air-mass predictors and four scan angle predictors For SSMIS a variety of different additional predictors were tried based on experience at the Met Office, NRL and ECMWF The Met Office scheme was found to give the best results Air-massSSMIS specificScan angle (θ) const offset zenith angle cloud liquid water lapse rate lapse rate square * *node is +1 if ascending, -1 if descending Bias correction predictors used are: May 21 2014 JCSDA Workshop

25. Slide 25 Application of NWP Bias Correction for SSMIS F18 Ascending NodeDescending Node Latitude Unbias & Bias Corrected O-B O-B Before Bias Correction Global Dsc Asc O-B After Bias Correction Global Dsc Asc O-B Before Bias Correction O-B After Bias Correction Using Met Office SSMIS Bias Correction Predictors T (K) May 21 2014 JCSDA Workshop

26. Slide 26 Application of NWP Noise Reduction for SSMIS By design SSMIS oversamples the brightness temperature field at relatively high noise Must apply spatial averaging before assimilating the data to reduce the noise A spatial averaging scheme was implemented inside of analysis(GSI) for SSMIS May 21 2014 JCSDA Workshop F-18 F-17 F-16 F-18 NR F-17 NR F-16 NR

27. Slide 27 Impact is not significant in northern hemisphere Marginally significant positive impact in southern hemisphere Impact of Assimilating SSMIS into Current Operational System (1) 15 Jan to 30 Mar 2012 (00Z cycles only) Control Enhance Control Enhance Northern Hemisphere 500 hPa Geopotential Height Anomaly Correlation Southern Hemisphere 500 hPa Geopotential Height Anomaly Correlation May 21 2014 JCSDA Workshop

28. Slide 28 850hPa shows statistically significant positive impact in medium range Some small but significant positive impact in short range for 200hPa winds Tropical (20°N-20°S) 0 to 5 Day Wind Forecast RMS Errors Impact of Assimilating SSMIS into Current Operational System (2) 15 Jan to 30 Mar 2012 (00Z cycles only) 200hPa 850hPa Control Enhance Control Enhance May 21 2014 JCSDA Workshop

29. Slide 29 Atmospheric Motion Vectors (Xiujuan Su) JCSDA Workshop May 21 2014

30. Slide 30 Assimilation of GOES hourly and Meteosat water vapor satellite winds in GSI GOES hourly winds replace current 3 hour GOES satellite winds GOES winds are currently supplied at three-hourly intervals with observation times within one hour of the analysis time. We will change to receiving GOES hourly winds and will assimilate observations throughout the six-hour assimilation window. This results in the data counts increasing by a factor of six. The quality of data from both products are similar. Meteosat water vapor winds The data was not used in GSI, data count is about 7000 each cycle after thinning May 21 2014 JCSDA Workshop

31. Slide 31 Assimilation strategy GOES hourly winds Tighten current quality control, add normalized Estimated Error (a new QC variable from the data providers) as new quality and new observation error Meteosat winds Add new quality control, observation error and thinning the data May 21 2014 JCSDA Workshop

32. Slide 32 Impacts The impact on forecast skills by adding these data are neutral for most part with slight improvement on southern hemisphere. Adding these data also improve observation fit on satellite winds. May 21 2014 JCSDA Workshop

33. Slide 33 GPS Radio Occultation (Lidia Cucurull) JCSDA Workshop May 21 2014

34. Slide 34 GPS RO changes RO observations from MetOp-A and MetOp-B are rejected below 8 km (refractivity and bending angle) Improved the quality control for RO observations (refractivity and bending angle) in the lower troposphere under the presence of large vertical gradients of atmospheric refractivity (primarily moisture) Bending angle: (a) if model detects 75% of the critical value around the height of the observation (we look at several model layers surrounding the observation), the observation is rejected if at/below this model layer. If several layers exist, we chose the top layer. (b) if bending angle > 0.03 rad and model detects at least 50% of critical gradient around the observation height, we select the observation within the profile with the largest bending angle. Any observation within the same profile and below the selected observation is rejected. May 21 2014 JCSDA Workshop

35. Slide 35 500-mb Geopotential Height anomaly correlation May 21 2014 JCSDA Workshop

36. Slide 36 Other Satellite Data Usage Changes (Andrew Collard, Li Bi) JCSDA Workshop May 21 2014

37. Slide 37 Other Satellite Data Usage Changes 1) Turn on new instruments: IASI MetOp-B (same as MetOp-A) 2) Turn off known bad channels: NOAA-19 AMSU-A Ch 7 NOAA-19 MHS Ch 3 Aqua AIRS Ch 321 3) Increase obs errors for sounding channels on ATMS: Chs 6-10 increased to 0.4K (from 0.3K) Chs 11-12 increased to 0.45K (from 0.35 and 0.4K) 4) Turn on cloud detection channels for monitored instruments: NOAA-17 HIRS, NOAA-19 HIRS, GOES-13 and GOES-14 sounders (we might also want to remove some of these instruments completely) 5) Turn on OSCAT (unfortunately OSCAT is no longer working) JCSDA Workshop May 21 2014

38. Slide 38 Projects for Future Implementations JCSDA Workshop May 21 2014

39. Slide 39 Near Sea-Surface Temperature (NSST) (Xu Li) JCSDA Workshop May 21 2014

40. Slide 40 Why is SST, an oceanic variable, analyzed in an atmospheric data assimilation system? More consistent initial conditions A single cost function and air-sea interaction More effective use of satellite data Assimilating satellite radiance directly to analyze the oceanic variable Taking advantage of the atmospheric data assimilation Advanced and updated frequently A direction to coupled data assimilation A single cost function for two media The covariance between the atmosphere and ocean Not included yet EnKF is able to handle it

41. Slide 41 Improved Oceanic Component within the NCEP GFS (1/2) An SST analysis scheme has been developed within the NCEP GFS (3DVAR GSI) SST extended to NSST (Near-Surface Sea Temperature), a T-Profile due to diurnal warming and sub-layer cooling The foundation temperature(Tf) is selected as the oceanic analysis variable NSST model developed (plus CRTM) to relate Tf and observed satellite radiance and oceanic temperature The combination of Tf and NSST profile provides the more appropriate boundary conditions of radiative transfer model and atmospheric forecasting model All the data are assimilated directly, including wavelength dependent satellite radiance and depth dependent in situ buys and ships sea temperature. Tf is analyzed 6-hourly together with atmospheric variables by minimizing a single cost function with GSI Partially atmosphere-ocean coupled in prediction mode

42. Slide 42 Improved Oceanic Component within the NCEP GFS (1/2) Results based on the cycling runs for one summer and one winter season SST analysis: Improved, in terms of (O-B) statistics against buoys, such as more Gaussin, lower bias and RMS, more used data SST prediction: Improved against buoys but degraded against own analysis, since more variability introduced in the new scheme and the suppressed variability in the control run Atmosphere prediction against own analysis: improved in tropics, neutral in higher latitude areas Conclusions An atmosphere-ocean partially coupled data assimilation and prediction system has been developed within the NCEP GFS and the results are encouraging Fully Coupled data assimilation (future) Strongly coupling (the combination of NSST and the NCEP CFS) Atmosphere-ocean convariance with Hybrid EnKF

43. Slide 43 The Impact of the NSST on atmospheric prediction: RMS difference (EXP – CTL), Tropics, Jan. 2011. Based on 124 (31 days, 4 times a day) 7-day forecasting. Positive impact in tropics at all levels except for near surface in T and RH (since more oceanic variability introduced in the NSST and the suppressed oceanic variability in the control run)

44. Slide 44 Cloudy Radiance Assimilation (Li Bi, Haixia Liu, Cathy Thomas, Xiaoyan Zhang, Yanqiu Zhu) JCSDA Workshop May 21 2014

45. Slide 45 Cloudy Radiance Projects Use new overcast radiance output from CRTM to convert GSI IR cloud detection scheme to radiance space. Microwave cloudy radiance assimilation Following ECMWF approach Focussing on non-precipitating clouds initially Infrared cloudy radiance assimilation CTP/cloud pressure retrieval approach Direct cloud assimilation approach Cloud cleared infrared radiances Use either new or existing cloud cleared radiance algorithm inside the GSI. Work on examining and extending the ensembles to make maximum use of cloud information May 21 2014 JCSDA Workshop

46. Slide 46 Any Questions JCSDA Workshop May 21 2014