基于人工神经网络的雅鲁藏布江流域NDVI预测模型

雅鲁藏布江流域既是生态资源的宝库,又是全球气候变化的敏感区。研究基于2000-2015年的MODIS数据和30个地面站点气象数据资料,在分析雅鲁藏布江流域的NDVI归一化植被指数时空变化特征的基础上,分别采用偏相关分析和主成分分析法,辨识了影响各子流域NDVI变化的主导气候因素,并此基础上构建了基于人工神经网络的雅鲁藏布江流域NDVI预测模型。结果表明:①雅鲁藏布江流域NDVI整体上呈现出从流域的上游到流域的下游逐渐增加的趋势;②主成分分析(PCA)和偏相关分析(PAR)的结果表明,降雨和气温的影响主要集中在前3个月且气温的影响大于降水;③分别构建了ANN-PCA、ANN-PAR和ANN模型,其率定期NASH效率系数平均值达到0.75,0.71,0.63,验证期平均达到0.73,0.69,0.62。结果表明,因子筛选能够显著提高模型精度,所建的模型精度较高,能够较好的模拟和预测雅鲁藏布江流域的NDVI时空变化趋势。

Prediction of NDVI in the Yarlung Zangbo River Using Artificial Neural Networks

The Yarlung Zangbo River Basin is not only a treasure of ecological resources,but also a sensitive area for global climate change.Based on the MODIS data from 2000 to 2015 and meteorological data of 30 ground stations,the spatial-temporal variation characteristics of the NDVI(the normalized difference vegetation index)in the Yarlung Zangbo River Basin are analyzed.The partial climatic analysis and principal component analysis are adopted to identify the dominant climatic factors that affect the NDVI change in each subzone.On this basis,the NDVI prediction models based on artificial neural networks are proposed and applied to Yarlung Zangbo River Basin.The results show:①The NDVI in the Yarlung Zangbo River Basin is gradually increasing from the upstream to the downstream.②The results of principal component analysis(PCA)and partial correlation analysis(PAR)show that rainfall and temperature in the first three months are the main factors affecting vegetation.③The ANN-PCA,ANN-PAR and ANN models are proposed and applied in the Yarlung Zangpo River Basin.The average Nash coefficients are 0.75,0.71 and 0.63 in calibration period respectively,and 0.73,0.69 and 0.62 in the verification period respectively.The results show that climatic factor identification can improve the model accuracy significantly.The proposed models achieve satisfied accuracy and can be applied to predict the spatial and temporal trend of NDVI in the Yarlung Zangbo River Basin.