用神经网络进行多波段卫星信息的降水估测

ESTIMATION OF RAINFALL USING MULTISPECTRAL SATELLITE DATA BY BP NETWORK

本文用BP神经网络对日本GMS 5的VIS(可见光 )、IR1(红外 1)、WV(水汽波段 )、及红外 1与红外 2波段亮温差资料进行降水强度与降水落区的估测。经 1998年夏季 6～ 8月降水期间的 12 0 8个检验样本的检验 ,无雨、小到中雨和大到暴雨的三等级降水的估测准确率分别达到 89.0 4 %、70 .34%、6 0 .1%。

In this paper,the VIS ,IR,WV etc. of GMS-5 dataset are used by BP network for predicting the rainfall and rainfall area,the comparision result of 1208 test samples from Jun. to Aug. in 1998 has shown that the accuracy rate of no rain,light or moderate rain and heavy rain achieved 89.04 %,70.34 %,60.1 % respectively.