基于高光谱数据的冬小麦叶绿素含量估算模型

发布时间：2018-01-01 04:29

本文关键词：基于高光谱数据的冬小麦叶绿素含量估算模型　出处：《河北师范大学》2017年硕士论文　论文类型：学位论文

【摘要】：绿色植被在生态系统中扮演着不可或缺的角色,叶绿素作为表征植被生长状况和发育状况的重要因子,成为水文、气候、土壤、生态等循环过程中的重要输入参数,被广泛应用于气候变化、生态循环变化等研究中。同时,对农作物叶绿素含量的监测,还可以为农业管理者提供决策信息,实现农业变量水、变量肥管理,提高水肥利用率,对发展可持续的现代农业具有重要的意义。本文利用2014年4月28日至5月2日和2015年4月25日和26日,野外试验和室内实验收集的冬小麦LAI、平均叶倾角(MTA)、株高、叶绿素含量、含水量和光谱反射率等数据,利用PROSAIL模型模拟了小麦冠层反射率曲线,并与实测小麦冠层反射率曲线进行对比,分析了叶绿素含量、LAI水平、MTA水平和含水量对小麦反射率曲线的影响,分析了不同观测天顶角对植被冠层反射率的影响。分析了不同叶绿素含量对红边幅值、红边面积、NDVI、MCARI和CIred edge的影响,并根据模拟的反射率曲线提取了红边幅值、红边面积、NDVI、MCARI和CIred edge建立了叶绿素含量线性反演模型、反向神经网络(BPNN)反演模型、支持向量机(SVM)反演模型,并验证了其反演精度。基于不同的观测天顶角建立了NDVI、MCARI和CIred edge的线性反演模型、BPNN反演模型、SVM反演模型,并验证了其反演精度,得到以下结论:(1)利用PROSAIL模型模拟了某样点的冠层反射率曲线与野外试验利用SVC测得的该样点的反射率曲线进行了对比分析,模型模拟的反射率曲线与实测反射率曲线走势一致,在可见光范围内反射率值相差无几,在780nm以后模型模拟的反射率值稍高一些,说明了PROSAIL模型能够很好的模拟小麦冠层反射率。(2)对植被冠层反射率的敏感性分析发现:叶绿素含量主要在可见光范围内,对植被的冠层反射率有影响,随着叶绿素含量的增加反射率值减小;LAI对反射率曲线的影响主要在近红外波段,随着LAI值的增加,植被冠层光谱反射率增加;平均叶倾角(MTA)对植被冠层反射率的影响与LAI相反,在近红外波段随着平均叶倾角的增大,反射率值减小;植被含水量对其冠层反射率的影响波段为近红外波段,随着含水量的增加反射率值减小。(3)利用PROSAIL模型模拟了不同观测天顶角的植被冠层光谱反射率,选择了三个角度0°、36°、55°,在植被生化组分含量不变时,同一波段内的植被冠层光谱反射率随着观测天顶角的增大而上升。(4)分析了叶绿素含量对红边幅值、红边面积、NDVI、MCARI和CIred edge的影响发现:随着叶绿素含量的增加红边幅值、红边面积、NDVI和CIred edge的值也呈线性增加,而MCARI的值随着叶绿素含量的增加,逐渐减小。(5)利用红边幅值、红边面积、NDVI、MCARI、CIred edge建立了叶绿素估算模型,并进行了精度验证,结果表明:在线性估算模型中,MCARI和CIred edge的估算模型的精度最高,相关系数R2分别为0.95和0.939,均方根误差分别为2.789和2.806,相对误差分别为0.45和0.048。基于不同的观测天顶角建立了NDVI、MCARI、CIred edge的线性估算模型,在线性估算模型中,观测天顶角为55°时,CIred edge估算模型的反演精度最好,其模型相关系数R2为0.953,均方根误差和相对误差分别为7.088和0.094。(6)利用BP神经网络预测模型和支持向量机预测模型建立了叶绿素估算模型。在BPNN反演模型中,基于MCARI和CIred edge的建立的神经网络反演模型反演效果最好,MCARI模型的均方根误差为2.809,相对误差仅为0.046;CIred edge模型的均方根误差为2.600,相对误差为0.927。在SVM反演模型中,基于MCARI建立的模型反演效果最好,其均方根误差为2.863,相对误差为0.045。(7)当观测天顶角不同时,BPNN反演模型中,观测天顶角为0°时,CIred edge-BPNN模型的均方根误差和相对误差分别为7.265和0.107;SVM反演模型中,观测天顶角为55°,CIred edge-SVM模型的均方根误差和相对误差分别为7.185和0.095。
[Abstract]:Green vegetation plays an indispensable role in the ecological system, an important factor, chlorophyll as the characterization of vegetation growth situation and development status of the climate, soil, hydrology and become, important input parameters of ecological cycle process, is widely used in the study of climate change, ecological cycle changes. At the same time, the monitoring of chlorophyll content in crops also, for agricultural managers to provide decision-making information, realize the variable water agriculture, variable fertilizer management, improve water and fertilizer use efficiency, is of great significance to the sustainable development of modern agriculture. This paper from April 28, 2014 to May 2nd and April 25, 2015 and 26, the winter wheat LAI collected field test and laboratory experiment, the mean leaf angle (MTA) and the plant height, chlorophyll content, water content and spectral reflectance data to simulate the wheat canopy reflectance curve using the PROSAIL model, and with the measured wheat canopy The reflectivity curves are compared and analyzed the content of chlorophyll, LAI level, MTA level and the effect of moisture content on wheat reflectance curve, analyzes the influence of different zenith angle on canopy reflectance. Analysis of the amplitude of different chlorophyll contents of red edge, red edge area, NDVI, MCARI and CIred influence edge, and extracted. The red edge amplitude according to the simulation of the reflectivity, red edge area, NDVI, MCARI and CIred edge to establish the linear inversion model of chlorophyll content, BP neural network (BPNN) inversion model, support vector machine (SVM) inversion model, and verified the inversion precision. Different view zenith angle is established based on the linear NDVI. MCARI and CIred edge model inversion, BPNN inversion model, SVM inversion model, and verified the accuracy of inversion, we get the following conclusions: (1) simulated canopy reflectance curve of a sample point and field test using PROSAIL model The reflectance curves using SVC measured the samples were compared and analyzed, simulated reflectivity curves and measured reflectance curve consistent with the trend, in the range of visible light reflectance values not much difference between after 780nm reflectance model, the simulation value is slightly higher, the PROSAIL model can simulate the wheat canopy reflectance as well. (2) the sensitivity analysis of vegetation canopy reflectance showed that chlorophyll content mainly in the visible range, impact on vegetation canopy reflectance, chlorophyll content increased with the decrease of reflectivity; the effect of LAI on the reflectance curve is mainly in the near infrared band, with the increase of LAI value, increasing vegetation canopy reflectance; the mean leaf angle (MTA) effect on vegetation canopy reflectance and LAI on the contrary, in the near infrared band increases with the mean leaf angle, reflectivity decreases; vegetation water content on the crown Effect of band reflectance layer for near infrared band, along with the increase of water content in reflectance values decreases. (3) simulated vegetation canopy spectral reflectance of different zenith angle by using the PROSAIL model, choose the three angles of 0 degrees, 36 degrees, 55 degrees, in the vegetation biochemical component unchanged, vegetation canopy spectral reflectance the same band increases with the increasing of viewing zenith angle. (4) analysis of the amplitude of the chlorophyll content of the red edge, red edge area, NDVI, MCARI and CIred found that the influence of edge: with the increase of the content of the chlorophyll red edge amplitude, red edge area, NDVI and CIred edge values increased linearly while the MCARI value decreases gradually with the increase of chlorophyll content. (5) the use of red edge amplitude, red edge area, NDVI, MCARI, CIred, edge established the chlorophyll estimation model, and the accuracy was verified, the results show that: in the linear estimation model, MCARI and CIred edge The estimation model of the highest accuracy, the correlation coefficient R2 were 0.95 and 0.939, the root mean square error were 2.789 and 2.806 respectively, the relative error is 0.45 and 0.048. NDVI, based on different zenith angle MCARI, CIred linear edge estimation model, the linear estimation model, the zenith angle is 55 degrees CIred edge, estimated the accuracy of model inversion is the best model, the correlation coefficient R2 is 0.953 and the root mean square error and relative error were 7.088 and 0.094. (6) model of chlorophyll estimation model and SVM prediction model using BP neural network. The BPNN inversion model, MCARI and CIred edge of nerve network inversion model inversion effect based on the best, the root mean square error of MCARI model is 2.809, the relative error is only 0.046; the root mean square error of CIred edge model is 2.600, the relative error is 0.927. in SVM inversion model, Model inversion based on MCARI best, the root mean square error is 2.863, the relative error is 0.045. (7) when the zenith angle is not at the same time, BPNN inversion model, the zenith angle is 0 degrees, the root mean square error of CIred edge-BPNN model and the relative errors are 7.265 and 0.107; SVM inversion model. The zenith angle is 55 degrees, the root mean square error CIred edge-SVM model and the relative errors were 7.185 and 0.095.

【学位授予单位】：河北师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S512.11

【相似文献】