A GPU-Based Solution for Ray Tracing 3-D Radiative Transfer Model for Optical and Thermal Images
Zunjian Bian,Jianbo Qi,J. -P. Gastellu-Etchegorry,J. -L. Roujean,Biao Cao,Hua Li,Yongming Du,Qing Xiao,Qinhuo Liu
DOI: https://doi.org/10.1109/lgrs.2022.3206312
IF: 5.343
2022-09-30
IEEE Geoscience and Remote Sensing Letters
Abstract:Three-dimensional (3-D) radiative transfer (RT) models are frequently recognized as a prerequisite when using high spatial resolution remote sensing (RS) data in heterogeneous surfaces. However, most studies of 3-D RT models have been restricted to limited applications due to the low computational efficiency. Therefore, this study proposed a graphic processing unit (GPU)-based solution for the ray tracing 3-D RT model. A state-of-the-art graphics and compute application programming interface (API), Vulkan, were introduced to implement the RT process. A bounding box method was adopted for the computation acceleration. By comparison with a central processing unit (CPU)-based solution, the performance efficiency of the proposed solution is significantly better: the simulation time of a GPU model is significantly reduced by more than 99% when facing a large-scale simulation mission. The simulation accuracy of the two solutions is similar, with root mean squared errors (RMSEs) lower than 0.005, 0.032, and 0.31 K for the red, near-infrared (NIR), and brightness temperature (BT) images, respectively. An evaluation based on airborne multiangle measurements also indicated that the accuracy of the proposed solution was satisfactory for simulating the red and NIR bidirectional reflectance factor (BRF) and BT directional anisotropies (DAs), with RMSEs lower than 0.003, 0.020, and 0.20 K, respectively, when treating the whole scene as a pixel. Considering the simulation accuracy and efficiency, a GPU-based model will be an important supplement to the CPU model.
imaging science & photographic technology,remote sensing,engineering, electrical & electronic,geochemistry & geophysics