Application of ADIFOR for air pollution model sensitivity studies

Typical computational methods of sensitivity analysis are discussed. Automatic differentiation addresses the need for computing derivatives of large codes accurately, regardless of the complexity of the model. Automatic differentiation in FORTRAN (ADIFOR) is a source transformation technique that accepts FORTRAN coded program for the computation of a function and generates portable FORTRAN code for the computation of the derivatives of that function. ADIFOR is introduced and applied to a comprehensive atmospheric chemistry/transport/radiative-transfer model to study the sensitivity of photochemical ozone production with respect to aerosol. The modeling results indicate that aerosol interaction with ozone may be as important as NOx and non-methane hydrocarbon (NMHC) emissions in determining ozone production. The presence of scattering and/or absorbing aerosols in the atmosphere can cause significant differences in calculated ozone levels. Normalized sensitivity coefficients show that ozone and other photochemical oxidants are most sensitive to the aerosol single scattering albedo, which determines the scattering efficiency of the aerosol. ADIFOR is demonstrated to be an effective tool for sensitivity analysis in air pollution modeling