for example, surface long-term weather forecast. is result broadens the interest for such missions, leading to the study of the Stratospheric Inferred Winds (SIW) [8] and SMILES-2 [9]-[11]. New features have been implemented in AMATERASU for conducting the simulation studies needed to design these missions. e large increase of the spectral informa-tion needed for wind measurements pushed us to acceler-ate the computations using Graphical Processing Units (GPUs). Also, the radiation polarization has been included to consider the Zeeman eect on the atomic and molecular oxygen lines targeted with SMILES-2, i.e., the change of the spectra due to the geomagnetic eld. ese features have been implemented with TensorFlow (TF), the opensource library developed by Google for machine learning (https://www.tensorow.org/). In this paper, we will rst provide a quick overview of SIW and SMILES-2, and of the simulation studies. e new AMATERASU features will be presented and we will dis-cuss the computation performances obtained with TF.Context2.1SIW and SMILES-2 missionsGlobal wind observations from the surface to the upper atmosphere are needed to improve the meteorological and climate prediction models which are being extended above 80 km [10]. e two past decades are considered as the golden age for the spatial observation of Earth, but wind data are barely available below 80 km [7]. To ll in the lack of measurements, NICT is studying a space-borne coherent Lidar to measure winds below 20 km, and passive THz heterodyne radiometers for the middle and upper-atmo-sphere (30–160 km). Figure 1 shows the wind component along the line-of-sight (LOS wind) obtained near 35 km with JEM/SMILES. is demonstrates the potential for such a technique to measure wind above about 30 km. is is roughly the lower limit for measuring wind because the wind induced Doppler shi is too small compared to the linewidth (the line is broadened by pressure).e missions SIW and SMILES-2 are designed for re-trieving horizontal winds, molecular composition and temperature using similar measurements. e observation method is shown in Fig. 2. e 2-d horizontal wind vector will be derived from perpendicular observations of the same air-mass using two antennas.SIW is a low-cost Swedish mission that will be launched in 2022/23. Chalmers Technical University (Sweden) leads the project and NICT is involved in the denition of the mission concept and the simulation studies [8]. e scien-tic payload is a small instrument (17 kg, 47 W) with the capability for measuring winds between 35 and 80 km (Fig. 3). It uses a double-side band (DSB) radiometer near 0.64 THz passively cooled (70 K below ambient tempera-ture). About 15 strong O3 lines will be measured for obtain-ing a wind signal to noise ratio larger than that of the 4-K cooled JEM/SMILES. SMILES-2 is a larger instrument (220 kg, ~300 W) equipped with 3 radiometers at 0.64, 0.76, and 2 THz, all cooled to 4.8~K [9][11]. e 0.64 THz band is the same as the SIW one. e mission could sense the altitude range between 20–160 km with a very high precision (130-180 K 2FiF1Left Panel: Daily average of quasi-zonal winds retrieved from JEM/SMILES near 35 km between 50N-55N [7]. The black dots show forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF, version cy23r4). Right panel: SMILES O3 line and LOS wind signature. The black-dashed lines show the 1- noise level on a single spectrum (4-K cooled radiometer with 300 K SSB system-noise temperature). 108 情報通信研究機構研究報告 Vol. 65 No. 1 （2019）4 衛星センサによる宇宙からの地球環境観測
元のページ ../index.html#112