支持向量机方法反演温湿廓线  

Temperature and humidity profile retrieval by support vector machine

作　　者：谭永强[1] 费建芳[1] 

机构地区：[1]解放军理工大学气象学院,江苏南京211101

出　　处：《解放军理工大学学报（自然科学版）》2010年第6期676-680,共5页Journal of PLA University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金重点资助项目(40830958);国家973计划资助项目(2009CB421502)

摘　　要：为提高温湿廓线反演效率,提出了一种基于支持向量机(SVM)反演大气温湿廓线的方法。利用欧洲中期天气预报中心(ECMW F)的RTTOV-8-7前向辐射传输模式和60L-SD廓线集生成了AM SU模拟亮温资料,对模拟亮温资料进行温湿廓线反演试验。试验发现:相对于多元统计回归反演方法,地面至10 hPa层次温度反演平均均方根误差减少了11.5%,温度反演误差分布与权重函数峰值能量高度层分布的密集区域基本一致;水汽反演优于统计反演方法较多,在地面,SVM水汽反演误差减少了39.9%。反演实验说明,SVM可以较好地描写温湿廓线反演中非线性映射关系。分别比较模拟亮温中加入1、2、3倍噪声的情况,发现该方法反演温湿廓线均具有较好的抗噪声作用。To improve the efficiency of retrieval,a new method based on support vector machine（SVM） was presented to retrieve temperature and humidity profile.Using RTTOV-8-7 forward model of European Centre for Medium-Range Weather Forecasts（ECMWF） and 60L-SD profile databases,the brightness temperatures received in AMSU instrument were simulated.The retrieval experiment shows that from the surface to the level of 10 hPa,the mean temperature RMS error of SVM retrieval method decreased by 11.5% than the multi-statistics retrieval method;temperature retrieve error is small while the layer with high density of weighting function peak;Humidity retrieval result is much better than the statistics retrieval method,SVM humidity retrieval error decreased by 39.9% in surface.The retrieval experiment shows that SVM technique can describe nonlinear mapping relationship very well in temperature and humidity profile retrieval.When respectively added noise to 1,2 and 3 times of the standard noise on each channel,the retrieval results also showed that the SVM method has good result.

关 键 词：温度廓线 湿度廓线 反演 支持向量机 

分 类 号：P458[天文地球—大气科学及气象学] E915[军事]