Error correlation between CO 2 and CO as a constraint for CO 2 flux inversion using satellite data from different instrument configurations Helen Wang D.J. Jacob, M. Kopacz, D.B.A. Jones, P. Suntharalingam, J.A. Fisher, R. Nassar, S. Pawson, J.E. Nielsen, C. Clerbaux GEOS-Chem simulated column for January 2006 Similarities between CO 2 and CO CO 2 50 ppb 150 CO 380 PPM 388 Both share common combustion sources & transport : Correlation CO has stronger CO gradient : more sensitive to transport error CO is relatively easier to observe from space: additional useful information TIR (mid/upper toposphere) NIR (down to PBL) Active CO,CO 2 am CO 2, CO?CO 2 CO,CO 2 COPm ASCENDSOCO?GOSATSCIAIASITESAIRSMOPITT CO 2 CO joint inversion Inverse model minimizes cost function: Observation vector CO 2 and CO concentration State vector CO 2 and CO surface flux Forward model: Jacobian Observation error covariance a priori error covariance Coupling between CO 2 and CO occurs through off-diagonal elements in error covariance matrices S r : error correlation coefficient CO 2 :CO a priori error correlation is weak due to large uncertainties in CO emission factor. CO 2 :CO observational error correlation is potentially useful in joint inversion. observational error in inverse flux modeling Instrument error Representation error Model error Observational error instrument noise retrieval error smoothing error resolution &timing mismatch among retrieval, forward & inverse model transport & chemistry Covariance between CO 2 and CO occurs through model error : Model error is dominant for satellite CO [Heald et al., 2004] important for satellite CO 2 [Baker et al., 2008] r M =0.96 CO 2 (kg/m 2 ) CO (kg/m 2 ) Jan 2006 (65W, 54N) CO 2 CO model error correlation calculation: paired-model and paired forecast method For each grid box and each month: Correlate time series of differences between CO and CO2 runs 1. Paired model: Perform runs with the same surface fluxes driven by different meteorology for CO2 and CO 2. Paired forecast: 48-hour 24-hour chemical forecast that are valid at the same time Jan 2006 July 2006 Positive correlation in winter due to common combustion and combustion & respiration region overlap Negative correlation in summer due to CO 2 uptake by biosphere Correlation pattern is robust w.r.t. methods, meteorology & CO 2 flux changes of NTE magnitudes Model error correlation coefficients from paired model method 1PM CO : 1PM CO 2 without averaging kernels 14-region analytical inversion using pseudo data with & without CO 2 CO model error correlation CO combustion 50% CO 2 combustion 30% CO 2 biosphere 80% CO chemistry 30% Rest of the World 30% Sate vector X a priori uncertainty Assume 90% of observational error is due to model error Jan 2006 biosphere 14-region analytical inversion using pseudo data with & without CO 2 CO model error correlation CO combustion 50% CO 2 combustion 30% CO 2 biosphere 80% CO chemistry 25% Rest of the World 30% Sate vector X a priori uncertainty a posteriori error covariance matrix Smaller implies greater improvement Assume 90% of observational error is due to model error CO combustion CO 2 combustion CO 2 biosphere Short bar implies big improvement relative to single species inversion Joint inversion show substantial improvement during Dec Apr for mid high latitudes The Tropics show