Ocean Sci. Discuss., 9, 3593-3642, 2012
www.ocean-sci-discuss.net/9/3593/2012/
doi:10.5194/osd-9-3593-2012
© Author(s) 2012. This work is distributed
under the Creative Commons Attribution 3.0 License.
Review Status
This discussion paper has been under review for the journal Ocean Science (OS). Please refer to the corresponding final paper in OS.
A comparison between gradient descent and stochastic approaches for parameter optimization of a coupled ocean–sea ice model
H. Sumata1, F. Kauker1,2, R. Gerdes1,3, C. Köberle1, and M. Karcher1,2
1Alfred Wegener Institute for Polar and Marine Research, Bremerhaven, Germany
2Ocean Atmosphere Systems, Hamburg, Germany
3Jacobs University, Bremen, Germany

Abstract. Two types of optimization methods were applied to a parameter optimization problem in a coupled ocean–sea ice model, and applicability and efficiency of the respective methods were examined. One is a finite difference method based on a traditional gradient descent approach, while the other adopts genetic algorithms as an example of stochastic approaches. Several series of parameter optimization experiments were performed by minimizing a cost function composed of model–data misfit of ice concentration, ice drift velocity and ice thickness. The finite difference method fails to estimate optimal parameters due to an ill-shaped nature of the cost function, whereas the genetic algorithms can effectively estimate near optimal parameters with a practical number of iterations. The results of the study indicate that a sophisticated stochastic approach is of practical use to a parameter optimization of a coupled ocean–sea ice model.

Citation: Sumata, H., Kauker, F., Gerdes, R., Köberle, C., and Karcher, M.: A comparison between gradient descent and stochastic approaches for parameter optimization of a coupled ocean–sea ice model, Ocean Sci. Discuss., 9, 3593-3642, doi:10.5194/osd-9-3593-2012, 2012.
 
Search OSD
Discussion Paper
    XML
    Citation
    Final Revised Paper
    Share