Comparative Study of Optimization Methods in a PM2.5 Transport Adjoint Model

Abstract

This paper focuses on the practical performances of the limited-memory BFGS (L-BFGS) method and the steepest descent method (GDM-S) by an adjoint data assimilation approach. The optimization procedure of the L-BFGS method in the ideal experiments clearly shows that the parameters should be scaled to similar magnitudes on the order of unity to improve the convergence efficiency. As compared with the GDM-S, the LBFGS method really uses much fewer steps to reach a satisfactory solution, but the performances are almost the same in the parameters inversions with the two optimization algorithms. In practical experiments, simulation results show good agreement with the observations of the period when the 21th APEC summit took place. Keywords-PM2.5 transport model; adjoint method; the L-BFGS method; the steepest descent method; parameters estimation