Application of Adjoint Sensitivity Theory to an Atmospheric General Circulation Model

View More View Less
  • 1 Engineering Physics and Mathematics Division, Oak Ridge, National Laboratory, Oak Ridge, TN 37831
© Get Permissions
Full access

Abstract

Sensitivity studies with climate models usually involve rerunning the model. Since climate models take so much computer time, this approach can be used to calculate sensitivities to only a few parameters. This work demonstrates the use of a much more efficient method of estimating sensitivities—the adjoint method. About a hundred first-order parameter sensitivities can be estimated with the same amount of computer time as a single integration of the model.

The adjoint method is applied to the Oregon State University atmospheric general circulation model. Sensitivities estimated using the adjoint method agree within 20% to those calculated directly by rerunning. In addition, the contribution to each parameter sensitivity is broken down by the adjoint method into its temporal, physical and spatial components. The main conclusions of this work are (a) the adjoint method is feasible with the right approximations for models as large as general circulation models, and (b) the sensitivity information revealed by the adjoint method is useful in helping to understand the results of climate models.

Abstract

Sensitivity studies with climate models usually involve rerunning the model. Since climate models take so much computer time, this approach can be used to calculate sensitivities to only a few parameters. This work demonstrates the use of a much more efficient method of estimating sensitivities—the adjoint method. About a hundred first-order parameter sensitivities can be estimated with the same amount of computer time as a single integration of the model.

The adjoint method is applied to the Oregon State University atmospheric general circulation model. Sensitivities estimated using the adjoint method agree within 20% to those calculated directly by rerunning. In addition, the contribution to each parameter sensitivity is broken down by the adjoint method into its temporal, physical and spatial components. The main conclusions of this work are (a) the adjoint method is feasible with the right approximations for models as large as general circulation models, and (b) the sensitivity information revealed by the adjoint method is useful in helping to understand the results of climate models.

Save