A comparison of interpolation processes: Applications to across-track scanning radiometers

This paper shows the first results of a more complete study on the comparison of interpolation methods applied to conically scanning and across-track scanning radiometers. The goal here is to present the performances of two interpolation algorithms for the interpolation of the raw measurements (samples) onto non-overlapping pixels, regularly localized along a scan of an across-track radiometer. The originality of the approach is the use of an end-to-end simulator including: (1) high resolution 2D realistic brightness temperature (TB) scenes computed from geophysical fields thanks to a radiative transfer model when previous studies have used synthetic ID profiles, (2) 2D convolution of the scene by Gaussian or measured antenna patterns at any pointing angle (defined by azimuth and elevation) when previous studies have used ID convolution of synthetic antenna patterns pointing at nadir, (3) notion of temporal integration when computing the raw radiometric measurements when previous studies have used instantaneous fields of view (IFOV) Both synthetic and high resolution realistic scenes, including or not radiometric noise, are used as referenced fields to assess the performances of these algorithms in term of radiometric accuracy and radiometric sensitivity. The simulation of the measurements is based on the convolution of the scene by the antenna patterns and takes into account the notion of Effective Field Of View (EFOV). The two interpolation processes are: (1) a purely geometric process based on the surface intersection between measurements (samples) and pixels -3dB beams projected on Earth. (2) the well-known Backus-Gilbert algorithm. The two methods show the same performance in term of radiometric accuracy when the Backus-Gilbert allows a better reduction of the radiometric noise.