支持向量机在AMSU-A临边调整中的应用  

USING SVM NETWORK TO ADJUST THE LIMB EFFECT OF AMSU-A

作　　者：谭永强[1,2] 费建芳[1] 张少洪[2] 徐宏[2] 

机构地区：[1]解放军理工大学气象学院,江苏南京211101 [2]空军装备研究院航空气象防化研究所,北京100085

出　　处：《热带气象学报》2010年第2期187-193,共7页Journal of Tropical Meteorology

摘　　要：采用支持向量机(SVM,Support Vector Machine)方法,对AMSU-A进行了临边调整试验。利用全球廓线数据集和快速辐射传输模式计算的理想亮温资料,以及AMSU-A全球实际亮温资料的分析表明,临边效应增大了窗区通道边缘视场的亮温,减小了5～14通道边缘视场亮温。临边效应对于各通道影响明显,且随着视角的增大而增大。通过理想试验分析表明,与多元线性回归方法相比,支持向量机方法对于窗区通道调整效果改进较多,对于通道5～14,同样优于多元线性回归方法。除窗区通道1、2、15边缘少数视场外,各视场调整均方根(RMS,Root Mean Square)误差在AMSU-A仪器噪声范围之内。对实际资料的试验表明,支持向量机方法调整效果同样优于多元线性回归方法。A nonlinear learning machine-Support Vector Machine （SVM） was used to adjust the instrument limb effect of the Advanced Microwave Sounding Unit-A （AMSU-A）. Respectively analyzing the ideal TBB data calculated from a fast transfer radiative model and global profile datasets and real data collected in April 2008 shows that the limb effect increased the window channel TBB, but decreased the TBB for Ch5, Ch6, Ch7 and Ch8, in off-nadir field of view （FOV）. The limb effect increased as the scan angle increased. The analyzed ideal data shows that the SVM network worked better than a multi-linear regression method, especially for window channels. Except for window Chl, Ch2 and Chl5, the limb-adjusted RMS error was less than the range of Noise Equivalent Delta Temperature （NEDT） for AMSU-A. The real data experiment shows that SVM network also worked better than the multi-linear regression method and the TBB adjusted by the SVM network limb may be used in weather and climate analysis.

关 键 词：大气探测 临边调整 支持向量机 AMSU-A 

分 类 号：P413[天文地球—大气科学及气象学]