An objective prior error quantification for regional atmospheric inverse applications

[摘要] Assigning proper prior uncertainties for inverse modelling of CO₂ is ofhigh importance, both to regularise the otherwise ill-constrained inverseproblem and to quantitatively characterise the magnitude and structure ofthe error between prior and "true" flux. We use surface fluxes derived fromthree biosphere models – VPRM, ORCHIDEE, and 5PM – and compare them againstdaily averaged fluxes from 53 eddy covariance sites across Europe for theyear 2007 and against repeated aircraft flux measurements encompassingspatial transects. In addition we create synthetic observations using modelledfluxes instead of the observed ones to explore the potential to infer prioruncertainties from model–model residuals. To ensure the realism of thesynthetic data analysis, a random measurement noise was added to the modelledtower fluxes which were used as reference. The temporal autocorrelation timefor tower model–data residuals was found to be around 30 days for both VPRMand ORCHIDEE but significantly different for the 5PM model with 70 days.This difference is caused by a few sites with large biases between the dataand the 5PM model. The spatial correlation of the model–data residuals forall models was found to be very short, up to few tens of kilometres but withuncertainties up to 100 % of this estimation. Propagating this errorstructure to annual continental scale yields an uncertainty of 0.06 Gt Cand strongly underestimates uncertainties typically used from atmosphericinversion systems, revealing another potential source of errors. Long spatiale-folding correlation lengths up to several hundreds of kilometres were determinedwhen synthetic data were used. Results from repeated aircraft transects insouth-western France are consistent with those obtained from the tower sitesin terms of spatial autocorrelation (35 km on average) while temporalautocorrelation is markedly lower (13 days). Our findings suggest that thedifferent prior models have a common temporal error structure. Separating theanalysis of the statistics for the model data residuals by seasons did notresult in any significant differences of the spatial e-folding correlationlengths.

[学科分类] 地球化学与岩石