欢迎访问《麦类作物学报》编辑部网站

作者	单位
张传波，李卫国，张宏，李伟，马廷淮，张琤琤，陈华	（1．江苏大学农业工程学院，江苏镇江 212013；2.江苏省农业科学院农业信息研究所，江苏南京 210014；3.江苏大学流体机械工程技术研究中心，江苏镇江 212013；4.南京信息工程大学，江苏南京 210044）

摘要点击次数: 362

全文下载次数: 560

中文摘要:

为探讨基于神经网络对小麦地上部生物量（aboveground biomass，AGB）进行遥感估测的可行性，在江苏省泰州泰兴市、盐城大丰区和宿迁沭阳县布设冬小麦大田试验，在对冬小麦近红外波段反射率（near-infrared band reflectance，REF_nir）、红光波段反射率（red band reflectance，REF_red）、归一化差值植被指数（normalized difference vegetation index，NDVI）、差值植被指数（difference vegetation index，DVI）、比值植被指数（ratio vegetation index，RVI）、土壤调节植被指数（soil adjusted vegetation index，SAVI）和优化土壤调节植被指数（optimized soil adjusted vegetation index，OSAVI）等7个遥感光谱指标与冬小麦生长指标（LAI和AGB）进行相关性分析基础上，构建基于BP神经网络的冬小麦AGB估测模型，并与多元线性回归估测模型进行精度比较。结果表明，冬小麦拔节期REF_red、NDVI、RVI、SAVI、OSAVI和LAI与AGB之间存在较好相关性，其中LAI与AGB的相关性最高（相关系数为0.782），SAVI与AGB的相关性最低(相关系数为0.647)。利用BP神经网络建立的冬小麦AGB估测模型AGB_BP的决定系数（r）为0.918，均方根误差（root mean square error，RMSE）为582.9 kg·hm^-2，平均相对误差（average relative error，ARE）为18.4%。利用多元线性回归分析建立的冬小麦AGB估测模型AGB_RA的r为0.784，RMSE为871.1 kg·hm^-2， ARE为32.6%。利用冬小麦抽穗期AGB实测数据再对模型AGB_BP和AGB_RA进行验证，其RMSE分别为1 140.4和1 676.7 kg·hm^-2， ARE分别为20.5%和33.1%。由此可以看出，冬小麦估测模型AGB_BP精度优于模型AGB_RA，说明利用多个遥感光谱指标结合LAI建模可以有效提高冬小麦AGB的估测精度。

英文摘要:

In order to explore the feasibility of remote sensing estimation of wheat above ground biomass(AGB) based on neural network,field experiments of winter wheat were carried out in Taixing City of Taizhou City,Dafeng district of Yancheng and Shuyang County of Suqian City,Jiangsu Province.By analyzing the correlation between near-infrared band reflectance(REF_nir),red band reflectance(REF_red),normalized difference vegetation index(NDVI),difference vegetation index(DVI),ratio vegetation index(RVI),soil adjusted vegetation index(SAVI),optimized soil adjusted vegetation index(OSAVI) and winter wheat growth indices(leaf area index and biomass),a winter wheat biomass estimation model based on BP neural network was constructed,and the estimation accuracy was compared with multiple linear regression model.The results showed that REF_red,NDVI,RVI,SAVI,OSAVI,leaf area index(LAI) and AGB had a good correlation at jointing stage of winter wheat.Among them,LAI had the highest correlation with AGB，with a correlation coefficient of 0.782; SAVI had the lowest correlation with AGB,with a correlation coefficient of 0.647.The winter wheat biomass estimation model AGB_BP established by BP neural network had a coefficient of determination(r) of 0.918,with root mean square error(RMSE) of 582.9 kg·hm^-2,and average relative error(ARE) of 18.4%.The r of the winter wheat biomass estimation model AGB_RA established by multiple linear regression analysis was 0.784，with the root mean square error of 871.1 kg·hm^-2,and the average relative error of 32.6%.Using winter wheat above-ground biomass estimation models AGB_BP and AGB_RA to estimate AGB at heading stage of winter wheat,the root mean square errors were 1 140.4 and 1 676.7 kg·hm^-2,respectively,and the average relative errors were 20.5% and 33.1%,respectively.The comparison between the two models showed that the accuracy of the winter wheat AGB_BP estimation model was better than that of the AGB_RA model,indicating that using multiple remote sensing spectral indicators combined with LAI modeling could effectively improve the estimation accuracy of winter wheat biomass.

查看全文查看/发表评论下载PDF阅读器

关闭