Oceanic Data Analysis Using a General Circulation Model. Part II: A North Atlantic Model

Authors:
  • ETEli Tziperman
    Eli Tziperman
  • WTWilliam Carlisle Thacker
    William Carlisle Thacker
  • RLRobert Bryan Long
    Robert Bryan Long
  • SHShow-Ming Hwang
    Show-Ming Hwang
  • SRStephen R. Rintoul
    Stephen R. Rintoul

Abstract
A general circulation model and North Atlantic climatological data of temperature salinity, wind stress, evaporation minus precipitation, and air–sea heat fluxes are used to examine the possibility of solving inverse problems using a full-scale numerical GCM and real oceanographic data, combined through an optimization approach. In this study several solutions for the model inputs and the structure of the cost function as a function of the model inputs are examined to demonstrate two of the main difficulties confronting such large-case nonlinear inverse problems (about 30 000 unknowns and a similar number of constraints for the problem examined here). The first is the possible existence of local minima of the cost function, which prevents convergence of the optimization to the global minimum representing the desired optimal solution for the model inputs. The second difficulty, which seems the dominant one for many of the problem examined in this part as well as in Part I, is the ill conditioning ... Abstract A general circulation model and North Atlantic climatological data of temperature salinity, wind stress, evaporation minus precipitation, and air–sea heat fluxes are used to examine the possibility of solving inverse problems using a full-scale numerical GCM and real oceanographic data, combined through an optimization approach. In this study several solutions for the model inputs and the structure of the cost function as a function of the model inputs are examined to demonstrate two of the main difficulties confronting such large-case nonlinear inverse problems (about 30 000 unknowns and a similar number of constraints for the problem examined here). The first is the possible existence of local minima of the cost function, which prevents convergence of the optimization to the global minimum representing the desired optimal solution for the model inputs. The second difficulty, which seems the dominant one for many of the problem examined in this part as well as in Part I, is the ill conditioning ...