Inferring the unobserved chemical state of the atmosphere: idealized data assimilation experiments

Abstract:

Chemical data assimilation in numerical models of the atmosphere is a venture into uncharted territory, into a world populated by a vast zoo of chemical compounds with strongly non-linear interactions. Commonly assimilated observations exist for only a selected few of those key gas phase compounds (CO, O₃, NO₂), and assimilating those in models assuming linearity begs the question of: To what extent we can infer the remainder to create a new state of the atmosphere that is chemically sound and optimal? In our work we present the first systematic investigation of sensitivities that exist between chemical compounds under varying ambient conditions in order to inform scientists on the potential pitfalls when assimilating single/few chemical compounds into complex 3D chemistry transport models.

In order to do this, we developed a box-modeling tool (BOXMOX) based on the Kinetic PreProcessor (KPP, http://people.cs.vt.edu/~asandu/Software/Kpp/) in which we can conduct simulations with a suite of ‘mechanisms’, sets of differential equations describing atmospheric photochemistry. The box modeling approach allows us to sample a large variety of atmospheric conditions (urban, rural, biogenically dominated, biomass burning plumes) to capture the range of chemical conditions that typically exist in the atmosphere. Included in our suite are ‘lumped’ mechanisms typically used in regional and global chemistry transport models (MOZART, RACM, RADM2, SAPRC99, CB05, CBMZ) as well as the Master Chemical Mechanism (MCM, U. Leeds). We will use an Observing System Simulation Experiment approach with the MCM prediction as ‘nature’ or ‘true’ state, assimilating idealized synthetic observations (from MCM) into the different ‚lumped’ mechanisms under various environments. Two approaches to estimate the sensitivity of the chemical system will be compared: 1) adjoint: using Jacobians computed by KPP and 2) ensemble: by perturbing emissions, temperature, photolysis rates, entrainment, etc., in order to create gain matrices to infer the unobserved part of the photochemical system.