OptimizingCO2 observing networksin the presence of modelerror: results from TransCom 3

[摘要] We use a genetic algorithm to construct optimal observing networksof atmosphericconcentration for inverse determination of netsources.Optimal networks are those that produce a minimum inaverage posterior uncertainty plus a term representing the divergenceamong source estimates for different transport models.The addition of thislast termmodifies the choice of observing sites, leading tolarger networks than would be chosen under the traditionalestimated variance metric. Model-model differences behave likesub-grid heterogeneity and optimal networks try to average over someofthis.The optimization does not, however,necessarily reject apparently difficult sites to model.Although the resultsare so conditioned on the experimental set-up that the specificnetworks chosen are unlikely to be the best choices in the realworld, the counter-intuitive behaviour of the optimization suggeststhe model error contribution should be taken into account whendesigning observing networks.Finally wecompare the flux and total uncertaintyestimates from the optimal network with those fromthe 3 control case. The 3 control case performs well underthe chosen uncertaintymetric and the flux estimates are close to those from the optimalcase.Thus the 3 findings would have been similar ifminimizingthe total uncertainty guided their network choice.

[学科分类] 大气科学