Quantitative Analysis of DART Calibration Accuracy for Retrieving Spectral Signatures Over Urban Area

Discrete anisotropic radiative transfer (DART) calibration is an iterative inversion method that is applied to shortwave (SW) satellite images to get maps of spectral signatures (SS) of city materials at the satellite spatial resolution. Therefore, it is potentially a handy spectral unmixing tool. However, up to now, it has only been validated by comparing the time series of SW radiative budget
Q
^*
_SW
from a flux tower in Basel to DART simulated
Q
^*
_SW
using maps of SS derived from satellite images. This article thoroughly assesses the DART calibration accuracy with two synthetic case studies, called “ideal” and “nonideal,” for short wavelengths. In both cases, the satellite image is a DART simulated image of an urban scene with ground, buildings with various structures, water, and shrubs. In the ideal case, SS maps are the only unknowns in the inversion process. The mean absolute value of the relative errors over all bands for ground, roof, water, tree, and shrub maps were 0.013, 0.005, 0.027, 0.297, and 0.250. In the nonideal case, we considered an uncertainty on parameters assumed to be known in the ideal case: solar zenith angle (SZA); satellite image spatial resolution; pixel-shift; inaccuracy of landscape modeling; and modulation transfer function (MTF). It led to larger errors: for ground, roof, water, tree, and shrubs, the mean absolute value of the relative error was 0.233, 0.507, 3.088, 0.834, and 1.256, respectively. By descending order of importance, the parameters that most affect the accuracy of the retrieved SS of urban material were SZA, satellite image spatial resolution, pixel-shift, inaccuracy of three-dimensional urban scene modeling, and MTF.