Extension and evaluation of sensitivity analysis capabilities in a photochemical model

  • Authors:

  • S. L. Napelenok;D. S. Cohan;M. T. Odman;S. Tonse

  • Affiliations:

  • Atmospheric Sciences Modeling Division, Air Resources Laboratory, National Oceanic and Atmospheric Administration, In partnership with the United States Environmental Protection Agency, 109 T.W. A ...;Department of Civil and Environmental Engineering, Rice University, 6100 Main Street, MS 317, Houston, TX 77005, USA;Department of Civil and Environmental Engineering, Georgia Institute of Technology, 311 Ferst Drive, Atlanta, GA 30332, USA;Lawrence Berkeley National Laboratory, 1 Cyclotron Road, Berkeley, CA 94720, USA

  • Venue:
  • Environmental Modelling & Software
  • Year:
  • 2008

Quantified Score

Hi-index 0.00

Visualization

Abstract

The decoupled direct method in three dimensions (DDM-3D) provides an efficient and accurate approach for probing the sensitivity of atmospheric pollutant concentrations to various changes in photochemical model inputs. The implementation of DDM-3D for the widely used Community Multiscale Air Quality (CMAQ) model was updated to account for recent changes in the base model and to include additional chemical mechanisms and advection schemes. The capabilities of CMAQ-DDM-3D were extended to enable execution using multiple processors in parallel and the computation of sensitivities to chemical reaction rate constants. The resulting direct sensitivity modeling system was tested for statistical agreement with the traditional difference method for calculating sensitivities, considering a summer episode in a domain covering the continental United States. Sensitivities to domain-wide and sector specific emissions, initial/boundary conditions, and chemical reaction rates were compared and found to be in good correlation for both primary and secondary air pollutants. The scalability of CMAQ-DDM-3D to the number of processors used in parallel was also examined. Sensitivity calculations were found to scale in a similar way to the base model, where the benefit to model runtime of adding more processors diminished for simulations that used more than eight processors.