Progress in the assimilation of GPS-RO at ECMWF

Progress in the assimilation ofGPS-RO at ECMWFSean HealyMike Rennie, Milan Dragosavac, Paul Poli, CarlaCardinali, Peter Bauer.

Impact of GPS-RO• The impact of GPS-RO NWP and climate reanalysis applicationshas been summarised in the keynote address by ECMWF’s DG.• Despite the relatively low observation number GPS-RO has asignificant impact.– Assimilation without bias correction– Superior vertical resolution• GPS-RO has reduced stratospheric temperature biases, anchoredradiance bias correction and is likely to be useful in the developmentof “weak constraint” 4D-Var.• Question: How do we improve the impact of GPS-RO? NWPusers and data providers must work together on this.

Improving the impact of GPS-RO• Improvements in how we assimilate the GPS-RO measurements willpartly depend on forward model development.– Summarise recent changes at ECMWF.– Outline our plans for the 2D operator.• Be aware of the limitations of the measurement technique.– The GPS-RO null-space.

Operational assimilation at ECMWF• We assimilate bending angles with a 1D operator. This ignores the 2Dnature of the measurement and integratesα( a)= −2a∞∫ad ln ndx2 2x−adx• The forward model now includes non-ideal gas effects when evaluatingthe model level heights (Aparicio et al, 2009) and when evaluating therefractivity on model levels based on Rueger and Thayer (Healy 2011).• I’d also recommend Aparicio and Laroche, PIP, JGR, 2011.• We also include a maximum refractivity gradient in the bending anglecalculation.

Maximum refractivity gradient in the observationoperator• Problem: assumption of linearity in incremental 4D-Var when closeto “ducting” (strong refractivity gradients) conditions.i+1iIn original version of the operatordid not calculate the bending below (i+1) ifn r≥ni i i+1 i+1However, problems arose if the separationwas only a few metres. Partial derivativescan get very large – eg,∂α∂qr

Previously in operationsOnly calculate bending angles below (i+1) if( ni+ 1 ir − n r ) > 101 i+iYes, it’s a bit ad-hoc. Seemed to work. But the MetOffice had 4D-Var convergence problems, so we put in amaximum gradient:⎛ dN ⎞max⎜− ⎟ = 0.157⎝ dx ⎠⎛ dN ⎞max⎜− ⎟ = 0.157 / 2⎝ dr ⎠

Testing maximum N gradient at ECMWF• Some reduction in the (o-b)/sigma_o bending angle biases in thetropical lower troposphere. EG, COSMIC-4 noise normalised stats.• What is the MAXIMUM GRADIENT GPS-RO can measure?Black = dN/dx max included

Tangent point drift• Work stimulated by paper by Foelsche et al (2010) and discussionswith Mike Rennie/John Eyre. Milan Dragosavac made the code changes.• The GPS-RO observation operator in the next operational cycle will includetangent point drift. (MF, NCEP already include this).• I did not think this was an important source of forward model error inthe stratosphere (where we were already doing a very good job!) butthat was incorrect.• Bending angle departure statistics are clearly improved and improved windscores.

(o-b) and (o-a) bending angle departure statistics(noise normalized)Black = tpd experimentRed = fixed location(Dotted line = analysisdepartures)COSMIC-4GRAS

Z500Reduction in 24 hrRMS errors asa result of TPDZ100

Stratospheric wind scores againstobservations: fractional reduction in RMS errorPositive impact

An agreed definition of “occultation point” (“meantangent point”) in BUFR files would be helpfulGRASTSX

2D bending angle operator plans• The ECMWF 2D bending angle operator is freely available as part ofthe GRAS SAF ROPP-4.0 (http://www.grassaf.org/software.php).• We’ll introduce tangent point drift in the 2D operator (but we’ll see ifwe can use batches of bending angles at a given lat/lon).• We’ll introduce a correction to account for the impact parametervariation along the ray path: da/ds=(∂n/∂θ)r. (I still do not understandwhy this is not a problem in the non-local phase/refractivityoperators ).• Investigate 2D/1D differences as a function of NWP resolution.• Investigate the water vapour information derived from 2D operator.

Comparing impact of GPSRO and IASI• Control: essentially the full observing system.• Baseline experiment: Control minus all level 1b sounder radiances,all GPSRO measurements and all aircraft temperaturemeasurements.• Investigated:– Baseline + GPSRO– Baseline + IASI– Baseline + IASI + GPSRO.

Comparison with IASI, relative to a poor baselinesystem.KEY QUESTION: How many GPSRO receivers do we need?Will COSMIC-2 surpass the impact of IASI?

(IASI + RO) vs RO fit to radiosonde TStandard deviation (K)Bias (K)(IASI + RO) is producing much better stratospheric biasesthan just GPSRO generally, but most obvious at theS.Pole. Why is GPSRO not doing a better job? TheGPSRO measurement has a “null space”.

Null space – how does the temperature difference atthe S.Pole propagate through the observation operator∆xH⋅∆x1K at ~25kmAssumed oberrorsThe null space arises because the measurements are sensitive to density asfunction of height (~P(z)/T(z)). A priori information is required to split this intoT(z) and P(z). We can define a temperature perturbation ∆T(z)~exp(z/H) which isin the GPSRO null space. Therefore, if the model background contains a bias ofthis form, the measurement can’t see or correct it.

Compare with Steiner et al(Ann.Geophs., 1999,17, 122-138)Temperature retrievalerror caused by a 5 %bias in the backgroundbending angle used inthe statistical optimization

New applications of GPS-RO• We have demonstrated GPS-RO can improve stratospherictemperature biases in NWP analyses and forecasts. (Despite thenull-space!)• As the GPS-RO time-series lengthens, the climate monitoring andreanalysis applications will become increasingly important.• In addition, we should start collaborations with the GCM developers,to promote the use of GPS-RO derived climatologies for testing theGCMs.– E.g., temperature climatology information between 200-5 hPa.

Summary• COSMIC GPS-RO measurements have had a significant impact onNWP. They are clearly an important component of the GlobalObserving System.• Summarized recent changes at ECMWF and outlined future planswith the 2D operator.• Noted the GPS-RO null space – temperature errors/biases thatGPS-RO cannot correct.• Many new applications can be anticipated in the coming years.

Day-5 RMS differences