Shi Yi-Ning, Yang Jun, Weng Fuzhong
Opt Express. 2021 Feb 1;29(3):4700-4720. doi: 10.1364/OE.417153.
Satellite data assimilation requires a computationally fast and accurate radiative transfer model. Currently, three fast models are commonly used in the Numerical Weather Prediction models (NWP) for satellite data assimilation, including Radiative Transfer for TIROS Operational Vertical Sounder (RTTOV), Community Radiative Transfer Model (CRTM), and Advanced Radiative transfer Modeling System (ARMS). ARMS was initiated in 2018 and is now becoming the third pillar supporting many users in NWP and remote sensing fields. Its radiative transfer solvers (e.g. Doubling Adding method) is inherited from CRTM. In this study, we propose a Discrete Ordinate Adding Method (DOAM) to solve the radiative transfer equation including both solar and thermal source terms. In order to accelerate the DOAM computation, the single scattering approximation is used in the layer with an optical depth less than 10 or a single scattering albedo less than 10. From principles of invariance, the adding method is then applied to link the radiances between the layers. The accuracy of DOAM is evaluated through four benchmark cases. It is shown that the difference between DOAM and DIScrete Ordinate Radiative Transfer (DISORT) decreases with an increase of stream number. The relative bias of the 4-stream DOAM ranges from -5.03 % to 5.92 % in the triple layers of a visible wavelength case, while the maximum bias of the 8-stream DOAM is only about 1 %. The biases can be significantly reduced by the single scattering correction. Comparing to the visible case, the accuracy of the 4-stream DOAM is much higher in the thermal case with a maximum bias -1.69 %. Similar results are also shown in two multiple-layer cases. In the MacBook Pro (15-inch, 2018) laptop, the 2-stream DOAM only takes 1.68 seconds for calculating azimuthally independent radiance of 3000 profiles in the hyper-spectral oxygen A-band (wavelength ranges from 0.757 µm to 0.775 µm), while the 4-stream DOAM takes 4.06 seconds and the 16-stream DOAM takes 45.93 seconds. The time of the 2-, 4- and 16- stream DOAM are 0.86 seconds, 1.09 seconds and 4.34 seconds for calculating azimuthally averaged radiance. DISORT with 16 streams takes 1521.56 seconds and 127.64 seconds under the same condition. As a new solver, DOAM has been integrated into ARMS and is used to simulate the brightness temperatures at MicroWave Humidity Sounder (MWHS) as well as MicroWave Radiation Imager (MWRI) frequencies. The simulations by DOAM are compared to those by Doubling Adding method and accuracy of both solvers shows a general agreement. All the results show that the DOAM is accurate and computational efficient for applications in NWP data assimilation and satellite remote sensing.
卫星数据同化需要一个计算速度快且准确的辐射传输模型。目前，数值天气预报模型（NWP）中常用于卫星数据同化的有三种快速模型，包括泰罗斯业务垂直探测器辐射传输模型（RTTOV）、社区辐射传输模型（CRTM）和先进辐射传输建模系统（ARMS）。ARMS于2018年启动，如今正成为支持NWP和遥感领域众多用户的第三大支柱。其辐射传输求解器（如加倍相加法）继承自CRTM。在本研究中，我们提出一种离散坐标相加法（DOAM）来求解包含太阳和热源项的辐射传输方程。为了加速DOAM计算，在光学厚度小于10或单次散射反照率小于10的层中使用单次散射近似。然后根据不变性原理，应用相加法来连接各层之间的辐射亮度。通过四个基准案例评估了DOAM的精度。结果表明，DOAM与离散坐标辐射传输模型（DISORT）之间的差异随着流数的增加而减小。在可见光波长案例的三层中，4流DOAM的相对偏差范围为-5.03%至5.92%，而8流DOAM的最大偏差仅约为1%。通过单次散射校正可显著降低偏差。与可见光案例相比，在热红外案例中4流DOAM的精度更高，最大偏差为-1.69%。在两个多层案例中也显示了类似结果。在苹果MacBook Pro（15英寸，2018）笔记本电脑上，2流DOAM计算高光谱氧A波段（波长范围为0.757 µm至0.775 µm）中3000个廓线的方位角无关辐射亮度仅需1.68秒，而4流DOAM需4.06秒，16流DOAM需45.93秒。计算方位角平均辐射亮度时，2流、4流和16流DOAM的时间分别为0.86秒、1.09秒和4.34秒。在相同条件下，16流的DISORT分别需要1521.56秒和127.64秒。作为一种新的求解器，DOAM已被集成到ARMS中，并用于模拟微波湿度探测器（MWHS）以及微波辐射成像仪（MWRI）频率下的亮温。将DOAM的模拟结果与加倍相加法的模拟结果进行比较，两种求解器的精度总体一致。所有结果表明，DOAM在NWP数据同化和卫星遥感应用中准确且计算效率高。
