Applications of Bivariate Fourier Series for Solving the Poisson Equation in Limited-Area Modeling of the Atmosphere: Higher Accuracy with a Boundary Buffer Strip Discarded and an Improved Order-Raising Procedure

Abstract

Abstract Bivariate Fourier series have many benefits in limited-area modeling (LAM), weather forecasting, and meteorological data analysis. However, atmospheric data are not spatially periodic on the LAM domain (“window”), which can be normalized to the unit square (x, y) ∈ [0, 1] ⊗ [0, 1] by rescaling the coordinates. Most Fourier LAM meteorology has employed rather low-order methods that have been quite successful in spite of Gibbs phenomenon at the boundaries of the artificial periodicity window. In this article, the authors explain why. Because data near the boundary between the high-resolution LAM window and the low-resolution global model are necessarily suspect, corrupted by the discontinuity in resolution, meteorologists routinely ignore LAM results in a buffer strip of nondimensional width D, and analyze only the Fourier sums in the smaller domain (x, y) ∈ [D, 1 − D] ⊗ [D, 1 − D]. It is shown that the error in a one-dimensional Fourier series with N terms or in a two-dimensional series with N2 terms, is smaller by a factor of N on a boundary-buffer-discarded domain than on the full unit square. A variety of procedures for raising the order of Fourier series convergence are described, and it is explained how the deletion of the boundary strip greatly simplifies and improves these enhancements. The prime exemplar is solving the Poisson equation with homogeneous boundary conditions by sine series, but the authors also discuss the Laplace equation with inhomogeneous boundary conditions.