||MCNC, Research Triangle Park, NC. North Carolina Supercomputing Center. ;National Oceanic and Atmospheric Administration, Research Triangle Park, NC. Atmospheric Sciences Modeling Div.;Environmental Protection Agency, Research Triangle Park, NC. Atmospheric Research and Exposure Assessment Lab.
In any simulation model, knowing the sensitivity of the system to the model parameters is of utmost importance. As part of an effort to build a multiscale air quality modeling system for a high performance computing and communication (HPCC) environment, the authors are exploring an automatic differentiation technique, Automatic Differentiation in FORtran (ADIFOR), for studying systematic model sensitivities. Instead of coding derivatives manually, one can apply ADIFOR without having to have an intimate knowledge of the algorithms implemented in a model. Also, most of the time, ADIFOR generates derivative code that follows the optimization schemes in the model. In this paper, we summarize the results from applying two sensitivity methods, ADIFOR and the brute-force method, to various advection schemes in air quality models.