【作者】 陈志强；

【导师】 白由路；

【作者基本信息】 中国农业科学院 ， 植物营养学， 2013， 博士

【摘要】 玉米是广泛种植的粮食作物，其产量的高低对世界粮食安全具有重要的影响。利用高光谱技术实时、快速监测玉米叶片氮素、叶绿素营养状况，能够合理指导其田间氮素管理，促使玉米的高产优质，体现现代农业的发展要求。本论文以两年玉米田间氮素梯度试验(N0～N6)为基础，以玉米叶片为基本研究对象，从不同角度(层次性、肥料用量、单株叶片光谱反射率平均值和SPAD值)探讨了叶片的光谱响应特性，明确了叶片光谱响应的敏感区域；通过筛选多种光谱参数，构建了基于单个和完整生育期的叶片氮素含量、SPAD值高光谱预测模型。全文的主要结论如下：(1)通过研究各个生育期(拔节期、大喇叭口期、开花吐丝期、灌浆期和蜡熟期)玉米叶片同一施肥水平不同层次、同一层次不同施肥水平的光谱反射率，明确了叶片光谱响应的敏感区域。低氮处理、植株下层次的玉米叶片，对光谱响应都较敏感。在350～760nm可见光波段(VB)和761～1300nm近红外波段(NIB)，叶片光谱响应的差别比较明显，VB区域内叶片光谱反射率随施肥量的增大而逐渐变小，NIB区域内叶片光谱反射率的变化规律较为复杂；在1301～2500nm短波近红外波段(SIB)，叶片光谱反射率间的变化较小。同一施肥水平下，玉米单株叶片光谱反射率平均值(LRM)在五个生育期的差别都很明显，主要位于VB和NIB区域；随着施肥量的增加，这种差别有逐渐减小的趋势。在VB区域，五个生育期的叶片光谱反射率都随着SPAD区间值的增大而逐渐变小；相同SPAD区间值内，叶片光谱反射率在各个时期的表现趋势也有所不同。(2)以单个生育期(拔节期、大喇叭口期、开花吐丝期、灌浆期和蜡熟期)数据构建玉米叶片氮素含量(LNC)、SPAD值高光谱预测模型时，LNC、SPAD值与叶片光谱反射率呈极显著负相关的波段都位于VB、SIB的部分区域；以差值参数(DSI)为基础构建的预测模型，能有效预测LNC、叶片SPAD值。(3)利用完整生育期(2011、2012、两年)数据构建玉米叶片氮素含量(LNC)、SPAD值高光谱预测模型时，叶片光谱反射率和一阶导数平均值曲线、LNC与光谱反射率相关系数曲线、SPAD值与光谱反射率及其一阶导数的相关系数曲线基本相同，LNC与一阶导数相关系数曲线的差别较大。以一阶导数为基础构建的预测模型，其预测效果不稳定；以DSI(R550附近，R680附近)、DSI(R680附近，R710附近)构建的预测模型，能有效预测LNC；以LCI、DSI(R550附近，R680附近)、DSI(R680附近，R710附近)构建的预测模型，能准确预测叶片SPAD值。(4)玉米穗位叶光谱响应的敏感区域位于可见光波段和近红外波段，叶片SPAD值与光谱反射率呈极显著负相关的波段位于VB、SIB的部分区域，以LCI、DSI(678，717)、DSI(549，678)为基础的预测模型，具备良好的评价效果和稳定性，能准确预测叶片SPAD值。(5)总体来讲，氮素营养下玉米叶片光谱反射率的差别明显区域位于可见光波段和近红外波段；叶片氮素含量(LNC)、SPAD值与光谱反射率呈极显著负相关的区域位于可见光波段，且两个相关系数最大值也位于该区域内；差值参数DSI(R550附近，R680附近)、DSI(R680附近，R710附近)都参与了最佳预测模型的构建。因此，利用高光谱遥感技术监测作物氮素、叶绿素状况时，应加强对该区域的重点研究，为研制低成本、高精度的便携式光谱诊断仪器提供数据支持。更多还原

【Abstract】 Maize is widely planted in the world, and its yield has important influence on food security.Nitrogen nutrition, chlorophyll status of maize leaves can be timely, rapidly monitored usinghyperspectral technology, which could be used to guide maize field nitrogen management, improvemaize yield and quality, embody development requirements of modern agriculture. In this thesis, a2-year maize field experiment of nitrogen gradient (N0~N6) was conducted. As the basic researchobjects, spectral response characteristics of maize leaves under many angles (levels, fertilizer dose, leafreflectance mean of per plant, SPAD value) were discussed, and spectral response sensitive areas ofmaize leaves were found. Various spectral parameters were selected to construct prediction models formaize leaf nitrogen contents, SPAD values based on single or the whole growth periods. The mainconclusions of this thesis were as follows:(1) Sensitive areas of maize leaves were found through studies on spectral response of maizeleaves under same fertilizer dose different levels, same levels different fertilizer dose at different growthstages (jointing stage, booting stage, anthesis-silking stage, filling stage and ripening stage). Maizeleaves of lower level or under low nitrogen fertilizer dose are more sensitive to spectral response. Thesensitive areas of maize leaves were in350~760nm visible band (VB) and761~1300nm near infraredband (NIB). Leaf spectral reflectance (LSR) in VB region decreased with the increase of fertilizer dose,and changes of LSR in NIB region were complex, while changes of LSR in1301~2500nm shortwaveinfrared band (SIB) were minor.Under the same fertilizer level, differences among leaf reflectance means (LRM) of per plant atfive growth stages were all obvious, sensitive areas were in VB, NIB regions, and the differencesdecreased with the increase of fertilizer dose. In VB region, LSR decreased with increase of SPADinterval values at five growth stages. Under the same SPAD interval value, changes of LSR at fivegrowth stages were different.(2) Prediction models for maize leaf nitrogen contents (LNC), SPAD values were constructedbased on data of a single growth stage (jointing stage, booting stage, anthesis-silking stage, filling stageand ripening stage). LNC, SPAD values were extremely significantly negatively correlated with LSR inpart of VB and SIB regions. LNC, SPAD values could be effectively predicted using prediction modelsbased on dissimilarity spectral index (DSI).(3) Prediction models for maize leaf nitrogen contents (LNC), SPAD values were constructedbased on data of the whole growth stages (2011,2012, two-year). The mean curves of LSR or its firstderivative, the correlation coefficient curves of LNC and LSR, SPAD values and LSR or its firstderivative were basically the same, while the correlation coefficient curves of LNC and first derivativewere different. Prediction models built with spectral parameters based on first derivative were unstable,LNC could be effectively predicted using models built with DSI (R550around, R680around) and DSI (R680around,R710around), maize leaf SPAD values could be accurately predicted using model built with LCI, DSI (R550around, R680around) and DSI (R680around, R710around).(4) Sensitive areas of maize ear leaves were in VB and NIB regions, and leaf SPAD values wereextremely significantly negatively correlated with LSR in part of VB and SIB regions. Evaluation andstability of prediction models built with LCI, DSI (678,717) and DSI (549,678) were good, thereforeleaf SPAD values could be accurately predicted.(5) Overall, spectral response sensitive areas of maize leaves were in VB and NIB regions underdifferent nitrogen fertilizer dose. LNC, SPAD values were extremely significantly negatively correlatedwith LSR in VB region, and the two maxima of correlation coefficients were in this region. The bestprediction models were all built with spectral parameters DSI (R550around, R680around) and DSI (R680around,R710around). Therefore, the VB region should be mainly studied when crop nitrogen, chlorophyll statuswere monitored using hyperspectral remote sensing technology, and which can provide experiment datafor developing low-cost, high-precision portable spectroscopy diagnosis instruments.更多还原