Adjoint sensitivity analysis for a three-dimensional photochemical model: application to Southern California

An adjoint method was used to investigate the sensitivity of peak ozone at selected sites in Southern California to nearly 900 model inputs including surface emissions, reaction rate coefficients, dry deposition velocities, boundary conditions, and initial conditions. Simulations showed large changes in ozone and ozone sensitivities at three sites investigated between summers 1987 and 1997 due to emission reductions. However, only small changes in ozone and ozone sensitivities were predicted between 1997 and 2010. Sensitivities of the differences in ozone between simulations with different emission scenarios were calculated and compared to sensitivities of ozone in each simulation. In some cases, the sensitivities of ozone differences were smaller than those of ozone itself, but in other cases, such as when the sensitivityto NOx emissions changed sign, sensitivities of differences were larger. The adjoint method was most useful for determining when and where model inputs affect, or have the potential to affect, an ozone response. For example, the method was used to plot the spatial distribution of important emission source regions to 1-hour versus 8-hour peak ozone. Changes in the distribution and sign of the adjoint function for emitted species revealed changes in the area of influence of pollutant emissions on peak ozone due to emission controls. The adjoint method provides useful information complementary to that obtained from forward sensitivity analysis methods.