Page 1 COST 723 workshop, Sofia, Bulgaria 16 th – 19 th May 2006 The ASSET Intercomparison project (ASSIC) Author: W.A. Lahoz Thanks to ASSET partners Data Assimilation Research Centre, University of Reading RG6 6BB, UK

Page 2 COST 723 workshop, Sofia, Bulgaria 16 th – 19 th May 2006 You want to do chemical DA… You ask: What model should I use? What DA scheme should I use? What data should I assimilate? How complex must chemistry be? Note implicit assumption: chemical DA adds value To answer: Test DA in the context of constituent DA The ASSIC project: Comparison of ozone analyses Geer et al. (2006), ACPD (accepted)

Page 3 COST 723 workshop, Sofia, Bulgaria 16 th – 19 th May 2006 Paradigm: Two approaches: NWP models (GCMs) & chemical models (CTMs): 1) NWP + tracer (stratospheric methane) NWP + parametrization of chemistry (Cariolle + cold tracer) NWP + chemical module (coupled system) 2) CTM + tracer (no chemistry) CTM + parametrization (Cariolle + cold tracer) CTM + complex chemistry Note: approaches converging (e.g. coupled NWP/CTM)

Page 4 COST 723 workshop, Sofia, Bulgaria 16 th – 19 th May 2006 DA options: (1) Var: 3D-Var: DARC/Met Office 4D-Var: ECMWF; BIRA-IASB 3D-FGAT: MF/CERFACS (2) KF methods: KF + parametrization: KNMI (3) other: Direct inversion Options re observations: (1) Retrievals (L2): Constituents: NWP (DARC/Met Office) CTMs (Several groups) (2) Radiances (L1): Nadir: NWP systems Limb: ECMWF Not investigated here

Page 5 COST 723 workshop, Sofia, Bulgaria 16 th – 19 th May 2006 u Model: NWP GCM: DARC/ Met Office, ECMWF CTM: KNMI, BASCOE, Meteo-France/CERFACS Isentropic CTM: MIMOSA, Juckes (2006) u Data: MIPAS: all except KNMI SCIAMACHY: KNMI; columns / profiles u Ozone chemistry: None: MIMOSA, Juckes Cariolle scheme: (linearized ozone chemistry) Full chemistry: Reprobus (reasonable troposphere), BASCOE (diurnal cycle in mesosphere) u Assimilation techniques: 3D-Var, 4D-Var, sub-optimal KF, Direct Inversion ASSIC: analyses

Page 6 COST 723 workshop, Sofia, Bulgaria 16 th – 19 th May 2006

Page 7 COST 723 workshop, Sofia, Bulgaria 16 th – 19 th May 2006

Page 8 COST 723 workshop, Sofia, Bulgaria 16 th – 19 th May 2006

Page 9 COST 723 workshop, Sofia, Bulgaria 16 th – 19 th May 2006 Upper stratosphere

Page 10 COST 723 workshop, Sofia, Bulgaria 16 th – 19 th May 2006 Mid stratosphere – ECMWF PV at 850K

Page 11 COST 723 workshop, Sofia, Bulgaria 16 th – 19 th May 2006 Mid stratosphere: ozone at 10hPa, 17 th October 2003

Page 12 COST 723 workshop, Sofia, Bulgaria 16 th – 19 th May 2006 Ozone hole ECMWF assimilate MIPAS ozone

Page 13 COST 723 workshop, Sofia, Bulgaria 16 th – 19 th May 2006 Ozone hole: heterogeneous chemistry schemes u Correctly depleting ozone to near-zero: T < 195K term (Cariolle v1.2, v2.1) MOCAGE-PALM Reprobus – detailed scheme BASCOE v3q33 – PSC parametrization u 0.5 ppmm ozone remains (incorrect): KNMI, DARC – cold tracer Cariolle v1.0 adds too much ozone in vortex KNMI? u 1-2 ppmm ozone remains (incorrect): BASCOE v3d24 – detailed PSCBox scheme Affected by ECMWF temperature biases? MIMOSA, Juckes No chemistry; MIPAS data limitations

Page 14 COST 723 workshop, Sofia, Bulgaria 16 th – 19 th May 2006 Midlatitude UTLS: Payerne ozonesonde profiles Sonde at full resolution Sonde at analyses resolution

Page 15 COST 723 workshop, Sofia, Bulgaria 16 th – 19 th May 2006 u Through most of stratosphere (50 hPa-1 hPa) biases within +/- 10% and standard deviations less than 10% vs sonde & HALOE u Areas with worse agreement: Upper stratosphere and mesosphere Ozone hole UTLS Tropical tropopause Troposphere These areas have issues about fidelity of transport, chemistry and observations First results

Page 16 COST 723 workshop, Sofia, Bulgaria 16 th – 19 th May 2006 u Ozone hole A variety of heterogeneous chemistry schemes work well u Upper stratosphere / mesosphere Care needed to avoid biases in linear chemistry Newer linear chemistry schemes work well Diurnal variability? Are there remaining uncertainties in chemistry and instrument calibration? u Troposphere Relaxation to climatology? u Tropical tropopause Improvements needed in modelled transport and in observations u Extratropical UTLS Needs further investigation and case studies Problem areas Green -> amber -> red

Page 17 COST 723 workshop, Sofia, Bulgaria 16 th – 19 th May 2006 MIPAS profiles: If good data quality & coverage: Similarly good results obtained no matter DA method or system MIPAS ~5% higher than HALOE in mid/upper stratosphere & mesosphere (above 30 hPa), & O(10%) higher than ozonesonde & HALOE in lower stratosphere (100 hPa - 30 hPa) SCIAMACHY total columns: Almost as good as MIPAS analyses; analyses based on SCIAMACHY limb profiles are worse in some areas -> problems in SCIAMACHY retrievals Conclusions from ASSIC

Page 18 COST 723 workshop, Sofia, Bulgaria 16 th – 19 th May 2006 Choice of system: Systems based on different approaches, i.e. 3D-Var, 4D- Var, KF or direct inversion, CTM or GCM, show broadly similar agreement vs independent data Potential improvements: Improved ozone analyses can be achieved via better modelled chemistry & transport & better observations Likely that improvements will come through better modelling of background errors (B)

Page 19 COST 723 workshop, Sofia, Bulgaria 16 th – 19 th May 2006 Chemistry complexity: Schemes with no chemistry do well vs. independent observations; extremely fast; depend on quality & availability of MIPAS data These schemes are not as good for ozone hole -> limitations with MIPAS observations; CTMs & GCMs with chemistry do better Studies of choice of linear chemistry parametrization (Geer et al) Costs: GCM-based analyses -> substantially more computer power than CTMs; ozone DA relatively small additional cost when included in NWP system

Page 20 COST 723 workshop, Sofia, Bulgaria 16 th – 19 th May 2006 Weather forecasts -> add ozone (usually with simplified chemistry) to enhance pre-existing NWP system (GCM-based): DARC/MO, ECMWF Chemistry-> strong arguments to build ozone DA into CTM with sophisticated chemistry, taking met. input as given: BIRA-IASB Ozone -> middle way: use transport model, driven by pre-existing dynamical fields, in combination with simplified chemistry scheme (e.g. Cariolle): KNMI, GMAO Coupled system -> chemical module embedded in GCM – still early days: BIRA-IASB and MSC Canada CHOICE DEPENDS ON APPLICATION Conclusions about chemistry applications

Page 21 COST 723 workshop, Sofia, Bulgaria 16 th – 19 th May 2006 What comes from these studies: key ideas in chemical DA Ozone DA key driver in NWP & chemical models; not as much benefit as originally hoped in NWP Key reason: relatively poor information content of ozone data available from current operational satellites -> Need more data (IASI, GOME-2,…) Important area for NWP & chemical model DA: estimation of B Needs to take account of physical & chemical principles & statistical data One of biggest challenges in DA Bias an important issue in DA -> estimation of biases a major challenge Observations & models confronted to identify & attribute biases Implementation of bias correction schemes active field of research

Page 22 COST 723 workshop, Sofia, Bulgaria 16 th – 19 th May 2006 Model deficiencies in transport & chemistry limit value of DA V. difficult for assimilated data to represent processes on timescales much longer than typical assimilation cycle – O(weeks to months) Expanding area for future DA activities is likely to be air quality Experience gained in stratospheric chemical DA -> starting point for tackling technical problems associated of tropospheric chemical DA Currently a wealth of constituent data from research satellites Danger that this will diminish just as DA techniques to exploit those data reaching maturity -> What happens after Envisat & Eos Aura?

Page 23 COST 723 workshop, Sofia, Bulgaria 16 th – 19 th May 2006 DA invaluable for use of stratospheric constituent measurements: Fills gaps between observations Allows use of heterogeneous measurements Numerical model -> information to be propagated forward in time: combination of measurements available at different times & locations Properly applied, DA can add value to observations & models, compared to information that each can supply on their own DA underpins evaluation of impact of current observation types using OSEs, and future global observing system using OSSEs DA ADDS VALUE BUT, for proper use, limitations must be borne in mind Finally…

Page 24 COST 723 workshop, Sofia, Bulgaria 16 th – 19 th May 2006 BE VERY CAREFUL HOW YOU USE DA…

Page 25 COST 723 workshop, Sofia, Bulgaria 16 th – 19 th May 2006 EXTRA SLIDES

Page 26 COST 723 workshop, Sofia, Bulgaria 16 th – 19 th May 2006 i. DA fills in observational gaps and can do so more accurately than linear interpolation of observations ii. By filling in observational gaps in an intelligent and statistically justifiable manner, DA provides initial conditions of constituent fields, consistent with observations & model, for chemical forecasts iii. DA (using inverse modelling ideas) allows quantification of sources & sinks of constituents (see, e.g., Meirink et al. 2005). A model cannot obtain initial conditions nor quantify sources & sinks objectively iv. DA allows better use of observations by assimilation of radiances – this should result in superior constituent retrieval capabilities by providing better a priori information for the retrieval process Added value

Page 27 COST 723 workshop, Sofia, Bulgaria 16 th – 19 th May 2006 v. DA confronts observations & models objectively by introducing error statistics (B & R) and allow the possibility of identifying problems with model or observations. By contrast, direct comparison of a model with observations does not necessarily have this same rigorous foundation vi. Errors & biases of both model & observations can be quantified in the process of tuning a DA scheme for internal consistency, e.g., Chapnik et al. (2006); Dee & da Silva (1999) vii. DA, using OSEs and OSSEs, can be used to evaluate incremental value of current observations and of (often expensive) future missions – models cannot be readily used for this purpose

Page 28 COST 723 workshop, Sofia, Bulgaria 16 th – 19 th May 2006 viii. Because CTMs can be very sensitive to the forcing fields (winds and temperature), their representation of constituent fields could be in error if forcing fields are in error (R. Ménard, pers. comm., 2005) ix. Dispersive nature of winds (Schoeberl et al. 2003) means CTMs will tend to provide inadequate budget distributions if, e.g., mean age of air is too young (Meijer et al. 2004) In both cases, DA keeps the model state on track, i.e., as consistent as possible with information from both constituent observations & forcing fields See Lahoz, Fonteyn & Swinbank, submitted to QJ

Page 29 COST 723 workshop, Sofia, Bulgaria 16 th – 19 th May 2006 The SH polar vortex split of Sep 2002: ozone at one level MIPAS ozone DARC analyses Blue: low ozone; Red: high ozone; 10 hPa Courtesy Alan Geer HOW DA CAN ADD VALUE