On “Gridless” Interpolation and Subgrid Data Density

Abstract

Nearest-neighbor gridding, binning, and bin-averaging procedures are performed routinely to map the irregularly sampled data onto a grid for data analysis and assimilation. Because these procedures are actually an interpolation procedure based on a piecewise constant function as the interpolation kernel, they tend to discard the subgrid locations of the data. Use of a locally continuous function for the interpolation kernel can preserve the subgrid location information in the data, at the cost of numerical sensitivity to the spatial variation in data density. This paper suggests a simple numerical procedure, based on a single correlation coefficient parameter, to eliminate such numerical sensitivity.