Numerical Simulation Of Hyperspectral Radiative Transfer Based On The GPU Parallel Computing Research

With the rise of the hyperspectral atmospheric remote sensing technology, there are someplace which demands more standard on the real-time simulation and fast computation of the hyperspectral radiative transfer (HRT) such as the hyperspctral simulation technology, the remote sensing retrieval methods and the hyperspectral sensor simulation technique. In order to improve the efficiency of HRT computation, there are two ways to be investigated, one is to develop the fast radiative transfer computation algorithm, the other is to adopt the parallel computing technology.In this study, there are four interesting topics which have been done to satisfy the needs of the HRT computation as follows:1. We improve the scattering phase matrix decomposition approach of the SOSVRT(the Successive Order of Scatering Vector Radiative Transfer) model, and we also joint add the symmetric relations based on the reciprocity theorem.So, under the symmetric discetization angle, the methods can greatly improve the model computational efficiency.2. Based on the improved SOSVRT model, we study the factors of the radiance and polarization of high-resolution oxygen A-band spectrum, which are aerosol size diatibution, aerosol vertical profile, instrument slit function, and surface albedo, respectively. The results show that:1) The multiple scattering will broaden the contribution layer and thus reduce the vertical resolving power.2) The stronger the oxygen absorption, the higher the altitude of the maximum contribution layer is.3)The polarized signal mostly come from the upper level of the atmosphere and the polarized radiance is much less sensitive to the surface albedo than that of the reflected radiance, which can be used to invert aerosols and cirrus in upper troposphere.4) Polarization signal is more sensitive to aerosol size distribution than that of the radiance, so that it can give a better constraint results on aerosol microphysical properties for retrieval of aerosol vertical profiles.5) Instrument specification has a profound impact on the results of retrievals, and the lower resolution and the more stray light will reduce the dynamic range of oxygen absorption depths.3. TIRT(the hyperspectral Thermal Infrared Radiative Transfer model) has been developed on the CPU platform, and TIRT.CUDA(the hyperspectral Thermal Infrared Radiative Transfer model based on the CPU+GPU heterogeneous parallel platform) hase been developed. The TIRT.CUDA model reached 24 times speedup without considering the data reading time.4. We first developed a vector radiative transfer model considered scattering process named SOSVRT.CUDA based on the CPU+GPU parallel platform while other radiative transfer model based on the CPU+GPU parallel platform can only compute the thermal radiative transfer. However, only the single scattering process was taken into account in SOSVRT.CUDA and it reached 24 times speedup. The contribution ratio of the single scattering process can reach 99% in the shortwave infrared band under sunny days, as a result, the SOSVRT.CUDA model can be used to develop the remote sensing retrival algorithms in the shortwave infrared band and improve the computation efficiency.