Fast Sensitivity Analysis of Three-Dimensional Photochemical Models

Photochemical air quality models increasingly are being used to understand the atmospheric dynamics of air pollutants and as the basis of emission control regulations. The response of the these model predictions to system parameters or emission controls provides valuable information for the strategy design to improve air quality. Such information can be pursued via sensitivity analysis, the systematic calculation of sensitivity coefficients, to quantitatively measure these dependencies. However, sensitivity analysis has not seen as wide of use as desired, in part because of the implementation complexity as well as computational limitations. For these reasons, sensitivity analysis has been applied primarily to subsystems of air quality models (e. g. Koda et al., 1974; Rabitz et al., 1983; Milford et al., 1992), or to limited aspects in air quality models (Cho et al., 1987). The “brute-force” method has been the most frequently used to determine model sensitivities, but it rapidly becomes less viable and prohibitively inefficient for a model when a large number of sensitivity coefficients needs to be computed.