Sean Healy Presented by Erik Andersson Use of COSMIC data in ECMWF’s global data assimilation system for numerical weather prediction Sean Healy Presented by Erik Andersson COSMIC in Global NWP

Outline Performance of GPSRO in a recent adjoint-based impact study: forecast error sensitivity to observations (FSO) Investigating the surface pressure information derived from GPSRO measurements GRAS/COSMIC consistency Summary COSMIC in Global NWP

Forecast Error Sensitivity to Observations (FSO) Data assimilation scientists have developed adjoint-based tools to estimate by how much various observation types contribute to the reduction of 24-hour forecast error. www.ecmwf.int/newsevents/meetings/workshops/2009/Diagnostics_DA_System_Performance Carla Cardinali has recently completed this type of calculation for the ECMWF 4D-Var data assimilation system GPSRO has performed well COSMIC in Global NWP

Forecast sensitivity to observations (FSO) J is a measure of the forecast error (“dry energy norm”, ps, T, u,v) Forecast error sensitivity to the analysis Analysis solution Rabier F, et al. 1996. Analysis sensitivity to observation and background The tool provides FSO for each assimilated observation, which can be accumulated by observation type, subtype, variable or level The forecast sensitivity (J is a scalar of the forecast error) equation can be expressed as a product of the forecast sensitivity with respect to the initial conditions and the analysis sensitivity with respect to the observations. The analysis solution can be expressed as the contribution of the background information plus the innovation vector. If we take the analysis rel xa=xb+Kdy and compute the analysis sensitivity with respect to the observation we obtain the transpose of the K-matrix gain. After some substitutions and applying numerical techniques to solve dJ/dy (Krylov solution). Once dJ/dy is computed we can now take the delta J and rearrange it by substituting the xa=… solution. The forecast error can be gathered over different subsets (type, subtype, variable and level) δy COSMIC in Global NWP

Observations’ contributions to decreased forecast error Operational FC system, Sept-Dec 2008 COSMIC in Global NWP

Observations’ contributions to decreased forecast error Operational FC system, Sept-Dec 2008 GPS-Radio Occultation COSMIC in Global NWP

Summary statistics by observation type Mean sensitivity of An to Obs Global observation influence on analysis: GI=7% Global background influence I-GI=93% Information content (DFS) COSMIC in Global NWP

Surface pressure information derived from GPSRO measurements The integration of the hydrostatic equation is part of the GPSRO observation operator because the bending angle and refractivity values are given as a function of a height co-ordinate. 1D-Var studies (Healy and Eyre, 2000) suggest that it should be possible to derive useful surface pressure information from the GPSRO measurements. We have recently performed experiments where all surface pressure information is blacklisted to see if COSMIC and GRAS can constrain the surface pressure field. Period June-July, 2009. Verified against ECMWF operations. COSMIC in Global NWP

Just to show the number of conventional Ps obs. COSMIC in Global NWP

Southern Hemisphere results (24 hour forecast mean error) GPSRO bias quite stable The GPSRO seems to constrain the Ps bias. “Control” is the full observing system. Similar temporal evolution in NH and tropics COSMIC in Global NWP

SH – sigma of 24 Hour error COSMIC in Global NWP

500Z height score (SH) The Ps measurements don’t have much impact from ~day-4 when GPSRO assimilated. HOWEVER, I’m currently looking at another period to see if I can reproduce this result. COSMIC in Global NWP

GRAS-COSMIC mean differences We expect GPSRO measurements from different instruments to have similar bias characteristics, but operational monitoring has shown that the GRAS and COSMIC bending angle biases differ by about 0.2% in the lower-mid stratosphere. In operations, the COSMIC departures were in better agreement with ECMWF forecasts and we initially assumed that the problem was with the GRAS processing. However, Christian Marquardt (EUMETSAT) demonstrated at the January 2009 AMS meeting that the problem was caused by the smoothing of the COSMIC phase delays at UCAR. UCAR proposed modifications to their processing and made 3 months (Nov, Dec, 08 and Jan 09) data available to the NWP centres. We used this data to investigate the GRAS COSMIC consistency. Revised data processing at UCAR has been operational since October 11, 2009. COSMIC in Global NWP

Global bending angle (o-b)/b departure statistics from ECMWF operations for Aug. 20 to Sept. 20, 2009 GRAS COSMIC-6 COSMIC-4 This is a typical result derived from operations before UCAR made the change. The COSMIC instruments agree with each Other but not with GRAS. COSMIC in Global NWP

Experiments with Modified COSMIC data Statistics for Dec 08 (NH) GRAS Much better consistency after UCAR processing change. But what causes the biases? Good agreement between GPSRO instruments, but what causes the –ve bias? COSMIC in Global NWP

Dec 08 Statistics when aircraft temperatures are blacklisted Aircraft T values bias the analyis warm, peaking at 200 hPa by ~0.5 K. This shifts all the stratospheric model levels upwards, Increasing the forward modelled bending angles. Part of the bias is caused by aircraft temp measurements which are known to be biased warm – stratospheric model levels too high, so the simulated bending angles are biased high. COSMIC in Global NWP

Summary FSO diagnostics show that GPSRO is an important observing system. We are currently investigating the surface pressure information content of GPSRO. Consistency between GRAS and COSMIC measurements much better since the processing change at UCAR. Part of the negative bending angle bias is caused by biased Aircraft T measurements. We plan to bias correct the aircraft Temperature measurements. COSMIC in Global NWP