波段反射率和植被指数结合的作物生长季农田土壤水分估测  

Estimation of farmland soil moisture in crop growing season based on combination of band reflectance and vegetation index

作　　者：张传波 李卫国 王晶[2] 李伟 马廷淮[4] ZHANG Chuan-bo;LI Wei-guo;WANG Jing;LI Wei;MA Ting-huai(College of Agricultural Engineering,Jiangsu University,Zhenjiang 212013,China;Institute of Agricultural Information,Jiangsu Academy of Agricultural Sciences,Nanjing 210014,China;Fluid Machinery Engineering Technology Research Center,Jiangsu University,Zhenjiang 212013,China;Nanjing University of Information Science and Technology,Nanjing 210044,China)

机构地区：[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

出　　处：《江苏农业学报》2022年第1期111-118,共8页Jiangsu Journal of Agricultural Sciences

基　　金：国家重点研发计划项目(2021YFE0104400);江苏省农业科技自主创新资金项目[CX(20)2037]

摘　　要：为了建立快速、有效的农田土壤水分含量(SMCF)遥感估测方法,在江苏省连云港市东海县、泰州市兴化市和盐城市大丰区布设SMCF遥感估测试验。在获取作物冠层近红外波段反射率(Near-infrared bandreflectance,R_(nir))、红光波段反射率(Red bandreflectance,R_(red))以及SMCF的基础上,通过分析波段反射率和植被指数多个遥感光谱特征指标与SMCF之间的相关性,构建基于BP神经网络的SMCF遥感估测模型,并与多元线性回归模型估测精度进行比较。结果表明,R_(nir)、R_(red)、差值植被指数(DVI)和比值植被指数(RVI)与SMCF间呈正相关关系,归一化差值植被指数(NDVI)和SMCF间呈负相关关系,各指标与SMCF的相关性从高到低依次为R_(nir)>DVI>R_(red)>NDVI>RVI,其中R_(nir)与SMCF相关性最高,相关系数为0.765。利用BP神经网络建立的SMCF估测模型的决定系数(R^(2))为0.928,均方根误差(RMSE)为3.61%,平均相对误差(ARE)为9.07%。利用多元线性回归方法建立的SMCF估测模型的R^(2)为0.660,RMSE为7.65%,ARE为21.43%。二者相比可以看出,BP神经网络SMCF估测模型的估测效果明显优于多元线性回归模型,说明将神经网络算法与波段反射率和植被指数结合建模,可以有效提高SMCF的估测精度。To set up a fast and effective remote sensing estimation method for soil moisture content in farmland(SMCF),remote sensing estimation experiments of SMCF were designed in Donghai County of Lianyungang City,Xinghua City of Taizhou City and Dafeng District of Yancheng City,Jiangsu province.After acquisition of canopy near-infrared band reflectance(R_(nir))and red band reflectance(R_(red))of crops and SMCF,correlation between multiple indicators of remote sensing spectral features such as band reflectance,vegetation index and SMCF were analyzed,and a remote sensing estimation model of SMCF based on back propagation(BP)neural network was constructed.Besides,the estimation accuracy was compared with multiple linear regression model.The results showed that,R_(nir),R_(red),difference vegetation index(DVI)and ratio vegetation index(RVI)were in positive relationships with SMCF.The correlation coefficients between different indexes and SMCF were sequenced as follows:R_(nir)>DVI>R_(red)>NDVI>RVI.Among them,R_(nir) had the highest correlation with SMCF,with a correlation coefficient of 0.765.The determination coefficient(R^(2))of SMCF estimation model established by BP neural network was 0.928,the root mean square error(RMSE)was 3.61%,and the average relative error(ARE)was 9.07%.R^(2),RMSE and ARE of the SMCF estimation model established by multiple linear regression method were 0.660,7.65%and 21.43%,respectively.By comparing the two models,it can be seen that the estimation effect of BP neural network model on SMCF is obviously better than multiple linear regression model,which shows that the estimation accuracy of SMCF can be effectively improved by combining the neural network algorithm with band reflectivity and vegetation index.

关 键 词：农田土壤水分含量 作物生长 多光谱波段反射率 植被指数 神经网络 

分 类 号：S127[农业科学—农业基础科学]