Estimation of Initial Field in the Bohai Sea with the Adjoint Method: A Comparative Study on Optimization Algorithms

Abstract:

The adjoint assimilation technique is used to invert the prescribed initial field in the Bohai Sea. Based on this technique, the practical performances of the limited-memory BFGS (L-BFGS) method, the Regularization method, and the Gradient Descent (GD) method are investigated computationally through a series experiments. Experimental results demonstrate that the prescribed initial field can be successfully estimated by these three methods. Inversion result with the Regularization method is better than that with the L-BFGS method, although errors of observations are higher. Though higher simulation errors than L-BFGS and Regularization method, the difference between the prescribed distribution and inversion result is the lowest, indicating that inversion result with the traditional GD method is the best.

• Periodical: Applied Mechanics and Materials (Volumes 571-572)
• Pages: 196-200
• DOI: https://doi.org/10.4028/www.scientific.net/AMM.571-572.196
• Online since: June 2014
• Authors: Chun Hui Wang, You Li Shen*, Xian Qing Lv
• Keywords: Adjoint Method, Gradient Descent, Initial Field, L-BFGS, Regularization

References

[1] S.Q. Peng, L. Xie, Effect of determining initial conditions by four-dimensional variational data assimilation on storm surge forecasting, Ocean Model., 14(1-2) (2006) 1–18.

DOI: 10.1016/j.ocemod.2006.03.005

[2] C.T. Allen, G.S. Young, S.E. Haupt, Improving pollutant source characterization by better estimating wind direction with a genetic algorithm, Atmos. Environ. 41(11) (2007) 2283–2289.

DOI: 10.1016/j.atmosenv.2006.11.007

[3] X.Q. Lv, J.C. Zhang, Numerical study on spatially varying bottom friction coefficient of a 2D tidal model with adjoint method, Cont. Shelf Res. 26(2006) 1905-(1923).

DOI: 10.1016/j.csr.2006.06.007

[5] H. Elbern, A. Strunk, H. Schmidt , O. Talagrand, Emission rate and chemical state estimation by 4-dimensionalvariational inversion, At mos. Chem. Phys. 7 (2007) 3749-3769.

DOI: 10.5194/acp-7-3749-2007

[6] J. Nocedal, S.J. Wright, Numerical optimization, 2nd ed. Springer: New York, (2006).

[7] D.C. Liu, J. Nocedal, On the limited memory BFGS method for large scale optimization. Math. Program. 45 (1989) 503–528.

DOI: 10.1007/bf01589116

[8] J. Nocedal, L-BFGS subroutine, software for large-scale unconstrained optimization. Available at: http: /www. ece. northwestern. edu/~nocedal/lbfgs. html.

[9] A.K. Alekseev, I.M. Navon, J. L. Steward, Comparison of advanced large-scale minimization algorithms for the solution of inverse ill-posed problems. Optim. Meth. Software. 24(1) (2009) 63-87.

DOI: 10.1080/10556780802370746

[10] A.N. Tikhonov, Solution of incorrectly formulated problems and the regularization method, Sov. Math. Dokl. 4 (1963) 1035-1038.

[11] V.A. Morozov, Methods for solving incorrectly posed problems. NewYork: springer, (1984).

[12] Y.H. Dai, Convergence properties of the BFGS algorithm. SIAM J. Optim. 13(3) (2002) 693–701.

DOI: 10.1137/s1052623401383455

[13] W.F. Mascarenhas, The BFGS method with exact line searches fails for non-convex objective functions. Math. Program. 99 (2004) 49–61.

DOI: 10.1007/s10107-003-0421-7