摘要
藏北高寒牧区草地是中国高寒草地分布面积最大的地区。为了及时准确地获得该区域草地覆盖度的变化趋势,本研究利用多年气象数据、社会统计数据、GIMMS、MODIS 2种归一化植被指数(NDVI)数据作为参数,构建BP神经网络模型,估算2010—2014年藏北高寒草地年际变化趋势,并用主成分分析方法优化参数来改进模型。结果表明:(1)BP神经网络模型及其改进模型对藏北高寒草地覆盖度年际变化趋势与遥感值的相关系数分别为0.16、0.47,表明通过主成分分析优化参数后的BP神经网络模型具有较好的模拟效果;(2)2种BP神经网络估算的植被指数值与NDVI值平均误差率分别为2.36%、2.20%,均有较高的模拟精度;(3)从神经网络训练步数上看,BP神经网络结果训练收敛步长为5000,基于主成分分析的BP神经网络模型训练收敛步长为454,表明后者提高了计算效率,体现出良好的收敛性。因此,无论从年际变化趋势拟合程度、植被指数估算值精度还是从计算效率来看,改进的BP神经网络模型对于估算藏北高寒草地覆盖度变化更加行之有效。
Alpine grassland in northern Tibet is the largest alpine grassland area of China. The paper aims to timely and accurately obtain the change trend of grassland coverage in northern Tibet. We built the BP neural network model and estimated the trend of annual changes of alpine grassland in northern Tibet from 2010 to 2014, and used the principal component analysis method to optimize the parameters to improve the model by using the meteorological data, social statistics data, GIMMS NDVI and MODIS NDVI data as parameters. The results showed that： （1） the correlation coefficient of BP neural network model and its improved model to alpine grassland coverage change value and the remote sensing value in northern Tibet was 0.16 and 0.47, respectively, indicating that the BP neural network model had good simulation effect after optimizing the parameters through principal component analysis; （2） the average error rate of vegetation index and NDVI estimated by 2 BP neural networks was 2.36% and 2.20%, respectively, with high simulation accuracy; （3） from the training steps of neural networks, the training convergence step length was 5000 based on the BP neural network model, and the training convergence step length was 454 based on PCA-BP neural network model, it was shown that the latter improved the computational efficiency and had good convergence. Hence, the improved BP neural network model is more effective to estimate the alpine grassland coverage changes in northern Tibet whether from the fitting degree of annual variation trend, the precision value of vegetation index estimation, or the computational efficiency.
作者
罗布
拉巴
尤学一
Luo Bu;La Ba;You Xueyi(Tibet Institute of Plateau Atmospheric and Environmental Science, Lhasa 850000;Lhasa Branch of Chengdu Plateau Meteorological Institute, China Meteorological Administration, Lhasa 850000;College of Environmental Science and Engineering, Tianjin University, Tianjin 300072)
出处
《中国农学通报》
2018年第11期48-53,共6页
Chinese Agricultural Science Bulletin
基金
国家自然科学基金项目"气候变化背景下西藏阿里地区草地退化研究"(41165002)
中国气象局成都高原气象研究所开放基金项目"利用卫星遥感资料反演藏西北高寒牧区草地陆面温度"(LPM2014005)
