摘要
针对碳储量回归预测模型存在共线性和精度较低的问题,利用森林资源二类调查数据和SPOT5影像数据对北京市延庆县的杨树林进行碳储量反演研究。先对选取的10个指标进行主成分分析,在此基础上采用径向基函数(RBF)神经网络方法构建碳储量反演模型,用预留测试样本验证,并与实测值进行比较。研究结果表明:SPOT5数据和二类数据可以很好地结合起来用于森林地上碳储量反演研究;PCA-RBF神经网络森林碳储量遥感反演模型拟合精度为99.90%,平均预测精度达到96.71%,预估效果较理想;模型训练完成后,可以应用于延庆县森林地上碳储量反演。
Aiming at the problem of multicollinearity and low precision predictions by the regression prediction model of carbon storage, this study used forest resource inventory data and SPOT5 image to retrieve the aboveground forest carbon storage of Populus forests in Yanqing County. Firstly, 10 factors were analyzed by principal components analysis. Then this paper introduced a method based on PCA and radial basis function （RBF） neural network for predicting forest carbon storage. The research results show that forest resource inventory data combined SPOT5 image is very useful for retrieving study of carbon storage of Populus forests~ the fitting precision of the PCA-RBF neural network model was 99.90% ,and the average prediction reached 96.71%. The model has a good retrieval accuracy, which can be well used for retrieval of regional aboveground forest carbon storage.
出处
《中国农业大学学报》
CAS
CSCD
北大核心
2012年第4期148-153,共6页
Journal of China Agricultural University
基金
国家"十一五"科技支撑计划(2006BAD23B05)
国家级林业推广项目(201145)
关键词
森林碳储量
SPOT5
主成分分析
遥感反演
RBF神经网络
forest carbon storage
SPOT5
principal component analysis
remote sensing retrieval
RBF neuralnetwork
