An automated system of air monitoring is being developed in Ust-Kamenogorsk city. This system should be able to model contamination and asses air condition. It contains data assimilation module for chemistry composition measurements. Its data assimilation algorithm is based on variational approach and adjoint problem methods. In the present paper we consider results of sensitivity studies of the data assimilation algorithm for convection-diffusion model to various input data perturbations. Sensitivity study allows to make a conclusion on the relative importance of input variables for the data assimilation algorithm, to identify key variables and identify ones that may be excluded from consideration. The model of pollutants transport and turbulent diffusion can be expressed in the operator form:
tDtxtxrtxftxYAttx),(),,(),(),()(, Dxxx),()0,(0 (*see the annexed pdf)
Here A is a "spatial" matrix operator; f– a priori source functions; tx, - state-function (concentrations); )(0x-initial state, ,,0tDDt D – spatial domain; tt,0 - time interval; Y - vector of model parameters. Control variable for the assimilation is the emission rate r. Solution of the data assimilation problem is sought as the minimum of target functional combining a norm of control function and a discrepancy between the modeled and measured data [1,2]. Errors may be in the initial data (which is uncertain), in the assimilated measurements data (due to the instrument errors) and also coefficients of chemical transport model may contain errors due to the imperfect numerical modeling of meteorological parameters [3]. In this paper we performed numerical experiments to determine the sensitivity of the data assimilation result to the noise in the observations, as well as the sensitivity to perturbations in meteorological fields (model coefficients).
