Calibration of Urban Surface Simulated Radiance Using Sentinel-2 Satellite Image and the DART Model

Satellite images are occasionally limited in some applications because of their specified imaging time and observation angles, and it is one solution to simulate them with radiative transfer models, which can reduce these limitations. However, there are many numerical differences between the simulated images and the satellite images. In order to calibrate simulated radiance, the discrete anisotropic radiative transfer (DART) model is used to simulate the red, green, and blue band images corresponding to the Sentinel-2 images, and an iterative algorithm, which is based on the assumption that the pixel radiance is the cumulation of all the elements radiance within the scene, is proposed to minimize discrepancies between the simulated images and satellite images. The results show that: 1) the algorithm is effective to enhance the similarity of the simulated images and Sentinel-2 images. There are differences in the red, green, and blue bands of 0.70 W
$\cdot $
mt
$^{-2}\cdot {\mathrm { m}}^{-1}\cdot {\mathrm { sr}}^{-1}$
,
$1.73~{\mathrm { W}}\cdot {\mathrm { m}}^{-2}\cdot {\mathrm { m}}^{-1}\cdot {\mathrm { sr}}^{-1}$
, and
$- 1.57~{\mathrm { W}}\cdot {\mathrm { m}}^{-2}\cdot {\mathrm { m}}^{-1}\cdot {\mathrm { sr}}^{-1}$
between the calibrated images and Sentinel-2 images and the relative differences of 0.77%, 2.30%, and 2.33%, respectively. The radiance differences between the simulated images (before and after calibration) and satellite images decreased from 5.34–
$77.78~{\mathrm { W}}\cdot {\mathrm { m}}^{-2}\cdot {\mathrm { m}}^{-1}\cdot {\mathrm { sr}}^{-1}$
to 0.01–
$4.01~{\mathrm { W}}\cdot {\mathrm { m}}^{-2}\cdot {\mathrm { m}}^{-1}\cdot {\mathrm { sr}}^{-1}$
, and the correlation coefficient increased from 0.39–0.41 to 0.60–0.64 and 2) it is varied for the different ground elements’ responses to the iteration algorithm; the simpler the ground element is, the better the results become. The methodology is useful to improve the precise of urban radiance simulation and is potential in minimizing errors in parameter estimation and expanding temporal analysis.