摘要
在全国挑选了东北、华北、华中、西南、华南、西北、新疆和西藏８个试验区，采用人工神经网络和逐步回归方法，应用温度、降水对植被指数进行预报（气候因子超前土地覆盖特征量２４个月）。试报结果表明：在对植被指数的预报上，人工神经网络优于逐步回归。同时尝试性的将神经网络预报方法与逐步回归方法结合起来作预报，即应用逐步回归挑选出来的预报因子作为神经网络的外部输入精选因子，进行神经网络模拟预报。研究表明，对神经网络的预报因子进行精选，可事先排除一些干扰信息，对提高神经网络的预报准确率有所帮助。
Observed data of temperature and precipitation are used to predict Normalization Difference Vegetation index (NDVI) in eight selected areas which is Dongbei, Huabei, Huazhong, Xinan, Huanan, Xibei, Xinjiang and Tibet in China by methods of artificial neural networks and stepwise regression. The climatic factors are 24 months ahead of NDVI. The testing forecast shows that the method of artificial neural networks is much better than that of stepwise regression in prediction of NDVI. Meanwhile above two methods are combined to predict NDVI, that is the factors selected by stepwise regression are used to be the input factors of artificial neural networks in predictions. The result shows that combining the two methods is helpful in improving forecast accuracy.
出处
《地理学报》
EI
CSCD
北大核心
2000年第B11期64-70,共7页
Acta Geographica Sinica
基金
中国科技部"九五"重中之重项目!(96-908-03-03)&&
关键词
土地覆被
气候预测模型
植被指数
人工神经网络
逐步回归
normalization difference vegetation index (NDVI)
artificial neural networks
stepwise regression
