Chemical Data Assimilation at the Meteorological Service of Canada Richard Ménard, Alain Robichaud Paul-Antoine Michelangelli, Pierre Gauthier, Yan Yang, and Yves Rochon

Operations - Assimilation of surface ozone measurements Observation simulation experiment - vertical profile lidar/total column scanning Research - development of coupled meteorology-chemistry model and data assimilation

Models CHRONOS Limited area CTM gas phase chemistry operational since 2001 North America domain: 24 km emission inventory (forest fires emissions) AURAMS Limited area CTM gas phase, PM, aqueous chemistry operational (parallel run) since 2004 Online coupling with operational meteorological weather forecast model GEM

Assimilation and objective analysis using the model CHRONOS Objective analysis, each hour, 24/7, year round On the web since (experimental) June 2004 Multiyear analyses since the summer 2002 Plans for operational implementation for spring Near real-time ozone objective analysis

Enhancement of the observation network and real-time data transmission US EPA AirNow ground level ozone observations ~ 1500 hourly observations Additional rural and remote sites

Four additional ozone sondes in southern Canada for 2004 summer measurement campaign Data available on WOUDC and NATChem

Distribution of TEOM site across Canada. Distribution of AEROCAN

Error statistics obs – model (obs loc) = (true + obs error) - (true + model error) = obs error – model error  distance (km)

Objective analysisObservations Analysis increment Error statistics Emissions Met fields Chemical model Ozone objective analysis and assimilation using CHRONOS

Best overall fit with first order auto-regressive correlation model (FOAR) Fit of observation error variance, forecast error variance and correlation length scale Classification in terms of land use was found to be most significant Number of sites Observation error variance Correlation length scale Forecast error variance Total variance INDUSTRIAL AGRICULTURALRESIDENTIALCOMMERCIAL FOREST

Observation error variance – CHRONOS v EDT - August 2004

Forecast error variance – CHRONOS V EDT- August 2004

Monitoring of the error statisticschi-square

Verification Verifying against observations not used to produce the analysis 1/3 of observations used for verification (red) 2/3 of observation used to produce the analysis (blue)

Monitoring of the error statistics in operational mode using previous year statistics

Analysis error variance. Reduction due to observations Provide a method for observation network design

Applications Real-time best analysis for surface ozone (tool for environmental forecaster available on a hourly basis) Ozone climatology (concentrations, dose, cumulative index, SUM60,AOT40, flux, etc.) Give insight into possible model bugs & errors Optimal design of measurement network Forecasting Re-analysis (using CHRONOS in a 24H assimilation hindcast mode)

Maps of SUM60 (cumul. Sum > 60 ppb) (Summer 2002) MODEL OBSERVATIONS OBJECTIVE ANALYSIS

AVG. FLUX OF OZONE TO SURFACE VD*[ozone] – Aug NO O 3 ASSIMILATION WITH O3 ASSIMILATION ppb*m/s

Incremental analysis vs cloud Case study. May Z

Prediction (assimilation) ON OFFON

Impact of assimilating ozone on other species

Ongoing and future work Use of new biogenic emissions (AURAMS)

OSSE capabilities  Simulate an observation system (e.g. a new instrument) in a data assimilation environment to assess the impact of the observation system  Simulated truth, i.e. nature run, is created by a different model: SEF with CMAM chemistry  The “observations” are drawn from the nature run  3D Var + GEM_Tracer is used as the assimilation system  ORACLE space-based Differential Absorption Lidar (DIAL)  Ozone ; 1 km vertical resolution from 500 hPa to 1 hPa  TOVS total column ozone

Vertically resolved measurements

Forecast error variance ORACLE TOVS ORACLE + TOVS

Chemical-Dynamical Coupling in Data Assimilation Richard Ménard, Simon Chabrillat(*), Martin Charron, Dominique Fonteyn(*), Pierre Gauthier, Bin He, Jerzy Jarosz(**), Alexander Kallaur, Jacek Kaminski (**), Mike Neish, John McConnell, Alain Robichaud, Yves Rochon and Yan Yang Meteorological Service of Canada *Belgium Institute for Space Aeronomy **York University Environment Canada Meteorological Service of Canada Environnement Canada Service Météorologique du Canada

Outline  Objectives of the study  Implementation  Issues / Challenges development of GCCM development of coupled meteorology-chemistry data assimilation system computational data assimilation

Development of General Circulation and Chemistry Model (GCCM) Global Environmental Multiscale (GEM) model operational NWP model at Meteorological Service of Canada semi-Lagrangian, adjoint + TLM global uniform/variable resolution stratospheric version hybrid vertical coordinate 80 levels, top 0.1 hPa 240 × 120 (1.5 degree) radiation, k-correlated method (Li and Barker 2003) uses as input H 2 O, CO 2, O 3, N 2 O, CH 4, CFC-11, CFC-12, CFC-113, CFC-114 sulfate, sea salt, and dust aerosols. non-orographic gravity wave drag (Hines) Dynamics and physics

Kinetic PreProcessor (KPP) symbolic computation to generate production and loss terms jacobian, hessian, LU decomposition matrices Online J calculation (MESSy code, Landgraf and Crutzen 1998) All species advected and gas phase chemistry solved with Rosenbrock or Fully implicit chemical solver (45 min time step) Implementation of TLM and adjoint. Choice of species and chemical reaction (gas phase) CMAM / BIRA-IASB Choice of bulk or sized-resolved PSC’s and aerosols (heterogeneous chemistry) Canadian Middle Atmosphere Model (CMAM) Danish Meteorological Institute Chemistry

Data assimilation system Stratospheric assimilation inherits the characteristics of the operational assimilation 3D Var and 4D Var –AMSU-A (channel added) and AMSU-B microwave channels –GEOS infrared radiances –Data quality control with BG check and QC-Var –Conventional meteorological data CMCNCEPUK MetOfficeECMWF

4D Var offers a more natural framework for the assimilation of time series of data, such satellite data Decomposition of assimilation algorithms in basic operations, e.g. PALM Modular approach to the development of 4D-Var –3D-Var: observation operators, background-error representation, etc. –GEM: direct (nonlinear), tangent linear and adjoint models Coupling of those modules is insured by an external coupler Assimilation is now running on the IBM-p690 –Current cycle: 5 nodes (40 PEs) Meteorological 4D Var (operational since 03/05)

Chemical data assimilation MSC: Real-time assimilation of surface ozone since York University-MSC : Coupled meteorology-chemistry data assimilation MOPITT CO Siberian forest fires August BASCOE : Belgium Assimilation System for Chemical Observation from Envisat (operational 4D Var CTM)

Development of the coupled dynamical-chemical data assimilation system 3D Var-CHEM Addition to an abritrary number of chemical tracer in the operational 3D Var Can accommodate cross-error covariance either operator form or explicit form

Not all chemical species are observed Analysis splitting ? only observed variables in control vector The problem of minimizing with respect to x and u is mathematically equivalent to minimizing followed by the update (Ménard et al. 2004) 4D Var extension Uses same solver as in 3D Var tangent linear integration

 Distributed computing / distributed memory GCCM OpenMP, MPI VAR-CHEM OpenMP, MPI (temp. solution analysis splitting )  Transport Can save computation in semi-Lagrangian advection transport upstream point (D or M) is the same for all advected species xxx interpolation weights C i (x) are the same for all advected species e.g. cubic Lagrange interpolation Computational Issues D M A

Data assimilation issues Because the ozone production rate increases with decreasing temperatures, in regions dominated by photochemistry (above 35 km) a negative correlation between temperature and ozone would occur Haigh and Pyle (1982), Froideveau et al. 1989, Smith 1995, Ward 2002 Cross-error covariance models e.g. Temperature-Ozone

For data at a given level, perturbations can fit an expression of the form with a correlation that can be up to 0.92 above 42 km, and increase linearly from zero to 0.92 between 37 km to 42 km.

Where we are after five months Development of the GCCM with York chemistry completed, and heterogeneous chemistry well underway. Kinetic preprocessor completed Validation of stratospheric meteorology has been made in both climate and assimilation mode 3D Var-CHEM is completed and operational Constructing the error statistics using differences of forecast (Rochon’s method) Development of 4D Var underway Short term plans (next three months) Validation of York (gas phase and heterogeneous) chemistry Completion of the chemical interface, and implementation of BIRA chemistry Validation of the error statistics using innovations and NMC method Validation of the coupled chemistry-dynamics assimilation over selected period of time Implementation of coupled chemical-dynamical 4D Var Start of monitoring of MIPAS observations – development of bias correction