摘要
基于红外高光谱大气探测器AIRS实况观测资料,尝试用人工神经网络算法来实现晴空时大气的温度垂直廓线反演,重点将该算法与目前已经发展比较成熟的特征向量统计反演算法进行比较。结果表明,两种算法在计算时间效率和反演精度上相当。鉴于人工神经网络算法的统计物理本质,误差反向传播BP神经网络可以求解非线性问题,在优化训练样本和继续调试网络种类和网络参数的基础上,有望能进一步提高反演精度。
The artificial neural network algorithm is presented in this paper to retrieve the temperature profiles under clear skies by using AIRS（Atmospheric InfraRed Sounder） actual observations.The study is focused on the comparison of artificial neural network retrieval algorithm with eigenvector regression algorithm which has already been well developed.The results show that these two algorithms cost nearly the same computing time with comparative precision in the real AIRS data retrieval process.Due to the statistic-physical nature of the artificial neural network algorithm,it is expected to improve the temperature retrieval precision on the basis of selecting network type and modifying the network parameters sequentially.
出处
《大气科学学报》
CSCD
北大核心
2010年第3期341-346,共6页
Transactions of Atmospheric Sciences
基金
国家自然科学基金资助项目(40605009)
关键词
红外
高光谱
人工神经网络
反演
infrared
hyperspectral
artificial neural network
retrieval
