梁红丽,张鹏,陈林,马刚,白文广,漆成莉. 2020. 基于梯度提升树的大气分层透过率快速计算方法[J]. 气象学报, (0):-, doi:10.11676/qxxb2020.055
基于梯度提升树的大气分层透过率快速计算方法
A machine learning method of efficiently computing level-to-space transmittances
投稿时间:2019-08-21  修订日期:2020-03-19
DOI:10.11676/qxxb2020.055
中文关键词:  大气透过率,红外辐射传输,机器学习,梯度提升树
英文关键词:Atmospheric transmittance, Infrared radiative transfer, Machine learning, GBT
基金项目:国家自然科学基金41675036
作者单位E-mail
梁红丽 中国气象科学研究院国家卫星气象中心 HelenLiang1@outlook.com 
张鹏 国家卫星气象中心 zhangp@cma.gov.cn 
陈林 国家卫星气象中心 chenlin@cma.gov.cn 
马刚 国家卫星气象中心 magang@cma.gov.cn 
白文广 国家卫星气象中心 bbaiwg@cma.gov.cn 
漆成莉 国家卫星气象中心 qicl@cma.gov.cn 
摘要点击次数: 271
全文下载次数: 234
中文摘要:
      大气透过率的计算是红外辐射传输计算的核心,RTTOV(Radiative Transfer for TOVS)通过建立大气廓线中温度、水汽、臭氧和其他气体浓度等参数和卫星通道透过率的统计关系,可实现卫星通道透过率和大气顶辐射率的快速准确计算。但在一些复杂吸收波段,如水汽波段,RTTOV的计算误差较大。为提高RTTOV在水汽敏感波段的计算精度,本研究利用机器学习中的梯度提升树(Gradient Boosting Tree,GBT)方法,选取从ECMWF(European Centre for Medium-Range Weather Forecasts)的IFS-137(The Integrated Forecast System, 137-level-profile)水汽廓线集中挑选的1406条廓线和由此计算的透过率真值作为样本,选取风云三号卫星上搭载的红外分光计(InfraRed Atmospheric Sounder, IRAS)通道12(7.33μm)进行了个例研究,分别建立了陆地和海洋晴空大气等压面至大气顶透过率和大气顶亮温的快速计算模型(GBT模型)。通过和透过率、亮温真值的比较,验证了GBT模型。比较的结果显示,GBT模型预测的透过率平均绝对误差(Mean Absolute Error, MAE)为:陆地0.0012,海洋0.0009,均方对数误差(Mean Squared Logarithmic Error, MSLE)为:陆地0.0215,海洋0.0095,均小于RTTOV直接计算的透过率的误差(陆地、海洋的MAE分别小0.0008和0.0010,MSLE分别小0.0135和0.0227);由GBT模型计算的亮温MAE分别为:陆地0.0949K,海洋0.0634K,均方根误差(Root Mean Square Error, RMSE)分别为:陆地0.1352K,海洋0.0831K,也都小于RTTOV直接模拟的晴空亮温误差(陆地、海洋的MAE分别小0.1685K和0.1466K,RMSE分别小0.1794K和0.1685K)本研究的结果表明,机器学习方法有提高水汽波段透过率和亮温计算精度的潜力,或可为辐射传输的快速计算提供可行的替代方法。
英文摘要:
      The calculation of atmospheric transmittance is the core of solving emitted infrared radiative transfer equations. As the layer transmittance of the atmosphere parameterized by functions of the mean layer temperature, water vapor, ozone and other gas concentration, RTTOV (Radiative Transfer for TOVS) is able to compute level-to-space transmittance and top-of-atmosphere radiance fast and accurately. However, the large computing error has been found in some strong absorption band, for example, the water vapor band. To solve this problem, a machine learning method called Gradient Boosting Tree (GBT) is utilized to compute transmittances in this paper. The 1406 typical profiles have been selected from ECMWF IFS-137 (European Centre for Medium-Range Weather Forecasts, The Integrated Forecast System, 137-level-profile) as training samples. The water vapor channel (7.33μm) of the IRAS (InfraRed Atmospheric Sounder)/FY-3 (Fengyun 3 series) is selected as case study. A fast model to compute level-to-space transmittance and the brightness temperature (BT) in the clear-sky by GBT method (GBT model hereafter) has been built. The calculated transmittances and BTs have been validated with the ground-truth. The comparison results show that the transmittances in the clear-sky calculated by GBT model are 0.0012(land) and 0.0009(ocean) in mean absolute error (MAE), 0.0215 and 0.0095 in mean squared logarithmic error (MSLE) respectively, which are smaller than those by RTTOV (0.0008(land) and 0.0010(ocean) in MAE, 0.0135(land) and 0.0227(ocean) in MSLE). In addition, the BTs in the clear-sky calculated by GBT model are 0.0949K(land) and 0.0634K(ocean) in MAE, 0.1352K(land) and 0.0831K(ocean) in root mean square error (RMSE) respectively, which are smaller than those by RTTOV(0.1685K(land) and 0.1466K(ocean) in MAE, 0.1794KK(land) and 0.1685K(ocean) in RMSE).The case study in this paper demonstrates that machine learning method has the potential capability to improve the accuracy to calculate the transmittance and BT in the water vapor bands. It provides an optional solution to the fast radiation transfer calculation.
查看全文   查看/发表评论  下载PDF阅读器
分享按钮