Efficient sensitivity computations in 3D air quality models

The prediction of ground level ozone for air quality monitoring and assessment is simulated through an integrated system of gridded models (meteorological, photochemical), where the atmosphere is represented with a three-dimensional grid that may include thousands of grid cells. The continuity equation solved by the Photochemical Air Quality Model (PAQM) reproduces the atmospheric processes (dynamical, physical, chemical and radiative), such as moving and mixing air parcels from one grid cell to another, calculating chemical reactions, injecting new emissions. The whole modeling procedure includes several sources of uncertainty, especially in the large data sets that describe the status of the domain (boundary conditions, emissions, chemical reaction rates and several others). The robustness of the photochemical simulation is addressed in this work through the deterministic approach of sensitivity analysis. The automatic differentiation tool ADIFOR is applied on the 3D PAQM CAMx and augments its Fortran 77 code by introducing new lines of code that additionally calculate, in only one run, the gradient of the solution vector with respect to its input parameters. The applicability of the approach is evaluated through a sensitivity study of the modeled concentrations to perturbations at the boundary conditions and the emissions, for three essentially dissimilar European Metropolises of the Auto-Oil II programme (Athens, Milan, and London).