摘要
为了准确快速地从118.75 GHz附近六通道亮温计算大气温度,作者开展了利用人工神经网络技术反演大气温度的数值模拟研究。与线性统计反演算法比较,海面上大气温度反演的总体均方根误差减小17%,陆面上大气温度反演的总体均方根误差减小15%。两种下垫面条件下的温度反演结果表明,近陆面的温度反演结果优于近海面的温度反演结果。另外,对温度廓线垂直结构反演性能的分析结果表明,对于具有较厚逆温层结构的温度廓线,神经网络反演对廓线的复现能力优于线性统计反演。
In order to retrieve atmospheric temperature profiles accurately and rapidly from 6 channels near 118.75 GHz, a numerical simulation study has been conducted using artificial neural networks. Comparison with a statistical retrieval method shows that more accurate results over ocean and over land can be obtained at all levels. The overall root mean square errors in retrieved profiles of a testing dataset are 17~ over ocean and 15 ~ over land less than that by using the statistical retrieval. In the cases of thick temperature inversion, the neural networks provide much better reproductions of the profiles than the statistical inversion.
出处
《气象科学》
CSCD
北大核心
2006年第3期252-259,共8页
Journal of the Meteorological Sciences
基金
国家重点基础研究发展规划项目(编号:2001CB309402)资助