【作者】 冯伟；

【导师】 曹卫星；

【作者基本信息】 南京农业大学 ， 生态农业科学技术， 2007， 博士

【摘要】 精确农业是现代农业生产中实现低耗、高效、优质和环境友好目标的根本途径，遥感技术可以快速获取农田作物生长状态的实时信息，为实施精确农业提供重要的技术支撑。高光谱遥感波谱具有连续、精细的特点，可显著增强对植株生物理化参数的探测手段和能力，提高作物生长监测的精度和准确性。本研究的目的是以小麦为对象，以系列田间试验为依托，综合运用高光谱遥感、生长分析、生理生化测试及数理统计分析等技术手段，分析不同氮素水平和品种条件下小麦冠层高光谱反射特征与氮素营养和长势指标之间的动态关系，确立氮素营养和长势指标的适宜光谱参数及相应监测模型，在探明小麦植株氮素状况与籽粒产量及蛋白质含量定量关系的基础上，建立了基于植株氮素营养的小麦籽粒产量和蛋白质含量光谱预测途径，从而为小麦生长特征的无损监测和精确管理提供理论基础和关键技术。本研究首先比较了不同氮素水平和不同品种条件下不同生育时期小麦冠层光谱反射率的变化模式。结果表明，随着施氮水平的提高，冠层反射光谱在近红外反射平台的反射率呈上升趋势，而可见光部分反射率则下降。小麦从拔节开始，冠层反射光谱在可见光波段先降低然后升高，以抽穗期前后反射率最低，而近红外区反射率则表现相反趋势；之后，随着生育进程推进，在可见光区反射增强，而近红外区反射较低，但在接近成熟时近红外区反射率急剧增强。这些光谱信息为进一步利用冠层反射光谱监测小麦生长状况和氮素营养状况提供了基础。利用冠层高光谱分析技术，提取了特征波段及多种光谱参数，研究了小麦氮素营养状态与冠层反射特征光谱的定量关系，建立了小麦植株氮素状况的敏感光谱参数及预测方程。群体叶片氮含量和积累量的敏感波段主要存在于近红外平台和可见光区，其中，红边区域表现最为显著。分别与叶片氮含量和氮积累量线性关系表现密切的光谱参数间存在差异，REP_LE、λ_O和mND705与叶片氮含量相关性好，方程拟合精度高；MSS-SARVI、FD742和PSSRb与叶片氮积累量关系最密切，方程拟合效果优于叶片氮含量。利用红边双峰特征构造新的红边参数，可以较好表达叶片氮素营养状态及变化，其中ND[RSDr(REP_IG)，LSDr(REP_IG)]对叶片氮含量方程拟合效果得到明显改善，而LSDr(REP_LE)与叶片氮积累量的关系非常密切，与FD742接近。经不同年际独立数据的检验表明，以REIP_LE和mND705两个光谱参数对叶片氮含量反应最敏感，而FD742和SDr／SDb两个光谱参数可以对叶片氮积累量进行有效监测，利用新构建的红边光谱参数建立叶片氮素营养监测模型均给出了理想的检验结果。比较而言，叶片氮积累量好于叶片氮含量。在分析不同氮素水平下小麦色素状况随生育期变化模式的基础上，讨论了色素状况与冠层高光谱反射率及光谱参数的关系，提出了小麦不同组分色素含量及密度的敏感光谱参数及预测方程。发现红边位置参数REP_IE和REP_IG与叶绿素含量关系较为密切，而与类胡萝卜素含量关系较弱。光谱参数VOG2、VOG3、RVI（810，560）、Dr／Db和SDr／SDb等与叶绿素密度关系较好，而与类胡萝卜素密度关系减弱。经独立试验资料检验，模型对色素含量估算效果以红边位置表现较好。VOG2、VOG3、Dr／Db和SDr／SDb对不同组分色素的估测检验结果均较好。总体上，利用关键光谱参数对叶片色素状况可以进行准确可靠的实时监测，其中，叶绿素a和叶绿素a+b的含量及密度监测效果最好。通过比较研究小麦叶片糖氮比随氮素水平的变化模式，确立了对小麦叶片糖氮比反应敏感的波段区域，明确了冠层反射光谱与叶片糖氮比的定量关系，得出小麦叶片糖氮比光谱监测的适宜时期为拔节期至灌浆中期。水分特征参数FWBI和Area980与叶片糖氮比关系密切，以指数方程拟合效果最好，而色素特征参数(R_750-800／R_695-740)-1和VOG2为变量指数方程拟合决定系数有所降低，但仍达极显著水平。独立年际试验数据的测试表明，利用FWBI、Area1190、(R_750-800／R_695-740)-1和VOG2四个光谱参数可以对生长盛期的小麦叶片糖氮比进行可靠的监测。在明确小麦叶干重和LAI随施氮水平和生育进程的动态变化模式的基础上，确立了对小麦叶干重和LAI反应敏感的波段区域，通过大量光谱参数的相关分析，建立了叶干重和LAI监测模型。对于不同试验条件下的叶干重和LAI，可以使用统一的光谱参数进行定量反演，其中基于RVI（810，560）、FD755、GMI、SARVI（MSS）和TC3等光谱参数的方程拟合效果较好。对表现较好的拟合方程进行检验，以参数RVI（810，560）、GMI、SARVI（MSS）、PSSRb、(R_750-800／R_695-740)-1、VOG2和mSR705为变量建立的叶干重和LAI监测模型均给出较好的检验结果，尤其是光谱参数RVI（810，560）、GMI和SARVI（MSS）可以对不同条件下小麦叶干重和LAI进行准确可靠的监测。对不同年份、品种和氮素水平下小麦籽粒蛋白质含量与不同生育期植株氮素状态的相关分析表明，利用前期叶片氮素状况可以预测成熟期籽粒蛋白质水平，其中通过开花期叶片氮含量和氮积累量可以进行可靠的估测。根据特征光谱参数-叶片氮素营养-籽粒蛋白质含量这一技术路径，通过将小麦叶片氮素监测模型融入链接，建立基于开花期高光谱参数的小麦籽粒蛋白质含量预测模型。模型检验显示，开花期光谱参数mND705、REP_LE、SDr／SDb和FD742可以对成熟期籽粒蛋白质含量进行准确预报，其中基于mND705参数的预测模型更为准确可靠。依据小麦产量形成的生物学特征，系统比较了植株氮素营养状况与籽粒产量之间的关系，提出了基于叶片氮素营养指标的籽粒产量预报模型。灌浆前期叶片氮积累量和叶面积氮指数均能够较好地反映成熟期籽粒产量状况，而叶片氮含量和氮积累量及叶面积氮指数在拔节-成熟期的累积值与成熟期籽粒产量的回归拟合效果更好。根据特征光谱参数-叶片氮素营养-籽粒产量这一技术路径，将小麦叶片氮素营养监测模型导入链接，建立基于开花期高光谱参数的小麦籽粒产量预测模型。检验结果表明，利用灌浆前期关键特征光谱指数可以有效地评价小麦成熟期籽粒产量状况，而拔节-成熟期特征光谱的累积值能够稳定预报不同条件下小麦成熟期籽粒产量的变化。综合分析了多种高光谱参数与小麦植株氮积累量的关系，确立了拟合度很好的光谱监测模型。基于植株氮积累量的积分累加值与对应籽粒氮积累状况的密切关系，构建了灌浆期籽粒氮积累量光谱估算方程，确定了小麦灌浆期地上部氮积累动态监测模型。独立年际试验数据的检验表明，基于高光谱指数SDr／SDb、VOG2、VOG3、RVI（810，560）、[(R_750-800)／(R_695-740)]-1和Dr／Db的监测模型可以实时评价小麦全生育期地上部氮素积累动态。更多还原

【Abstract】 Precision farming is a prime approach to realizing high yielding, good quality and low consumption in modem agricultural production. The operational precision farming has been hampered by a lack of timely distributed information of crop conditions, and remote sensing can rapidly determine the growth status of crop in the field, which offers important technical support for implementation of precision farming. In recent years remote sensing technology has been proved to easily monitor and forecast growth characters, nitrogen status, yield and quality formation of farm crop. The newly-emerged hyperspectral remote sensing is sensitive to specific crop parameters, with better estimation of various growth variables related to crop physiology and biochemistry.In this study, a series of field experiments with different wheat varieties and nitrogen levels were carried out in three years. Based on analysis of canopy spectral reflectance and assay of agronomic parameters and physico-chemical index, in wheat plant, the characteristics of canopy hyperspectral reflectance under different conditions and their correlation to nitrogen status, growth characters, and grain traits in wheat were quantified computed in this paper. The sensitive spectrum parameters and quantitive regression models of nitrogen status, pigment status, leaf area index, biomass were established. Based on the spectral monitoring of plant nitrogen status and the quantitative relationships between nitrogen status, nitrogen translocation, grain yield and protein content, the indirect approach predicting mature grain traits with reflectance spectra were developed, which provided technical basis for non-destructive monitoring and precise diagnosis of wheat growth.Comparison of variation patter in canopy reflectance under different nitrogen rates, growing stages and cultivars in wheat, showed that reflectance at near infrared flat （750-1300nm） increased with increasing nitrogen rates, whereas reflectance at visible band decreased. Reflectance at visible light initially decreased and then increased with growth progress after jointing, with the lowest value appeared around at heading. However, reflectance in near infrared had opposite trend, which initially increased and then decreased to the lowest from growth bloom stages to maturity. These results provide background spectral information for monitoring of growth characters, nitrogen status and grain formation with canopy reflectance spectra in wheat.Based on technique of hyperspectra analysis, many characteristic bands and derived spectral parameters were obtained. The quantitative relationships of leaf nitrogen status to canopy reflectance spectra, and the sensitive parameters and monitoring equations of leaf N status were put forward. The sensitivity bands occurred during visible light and near infrared region mostly, and a close correlation existed between red-edge district and learN status. The vegetable indices related most significantly to LNC differed from that of LNA. An integrated linear regression equation of LNC to REIPLE,λO and mND705 described the dynamic pattern of change in LNC in wheat, giving high determination of coefficients and low standard errors. MSS-SARVI, FD742 and PSSRb were linearly related significantly to LNA, with higher R2 for LNA than that for LNC. The new developed red edge parameters based on red edge double peak could well describe the dynamic pattern of LNC and LNA changes in wheat, and the best indices was ND[RSDr(REP_IG), LSDr(REP_IG)] for LNC, and LSDr(REP_LE) for LNA with both high R² and low SE by regression analysis.The change characteristics of different pigment forms in leaves with development stages and quantitative relationships to canopy reflectance spectra and derived spectral parameters in wheat were investigated. The red edge position （REP） was highly correlated with leaf chlorophyll concentrations, with high R² in REP_LE, but R² between carotenoid and different spectral indices all decreased significantly. The correlations of chlorophyll density to VOG2, VOG3, RVI（810,560）, Dr/Db and SDr/SDb were higher. Testing the monitoring equations with independent datasets indicated that the red edge position were the best to predict leaf pigment concentrations, and VOG2, VOG3, Dr/Db and SDr/SDb were indicators of leaf pigment density. The overall results suggested that pigment concentrations and density in wheat leaf could be estimated by hyperspectral parameters selected, and the chlorophyll a and chlorophyll a+b status could be reliably estimated in wheat.The change patterns of leaf soluble sugar to nitrogen ratio with nitrogen levels, and the quantitative relationships to characteristic bands and sensitive parameters were analyzed. The proper time for monitoring leaf soluble sugar to nitrogen ratio should be from jointing to mid-filling, with best stage as anthesis. FWBI and Area980 of water-index were highly correlated with leaf soluble sugar to nitrogen ratio, and (R_750-800/R_695-740)-1 and VOG2 of pigment-index were also significantly related to leaf soluble sugar to nitrogen ratio, with the highest determination of coefficients from exponential equation. Testing of the monitoring models with independent dataset indicated that FWBI, Areal190, (R_750-800/R_695-740)-1 and VOG2 were the best indicators to estimate leaf soluble sugar to nitrogen ratio.Based on the change patterns of leaf area index and dry weight under different nitrogen rates with growth stages, correlations of LAI and leaf dry weight to canopy hyperspectral reflectance and spectral parameters were investigated, and sensitive spectral parameters and quantitative equations were developed to forecast LAI and leaf dry weight in wheat, with unified spectral parameters for LAI and leaf dry weight across a broad ranges of growth stages, nitrogen levels and growing seasons. Regression models with spectral variables as RVI（810,560）, FD755, GMI, SARVI（MSS） and TC3 produced better estimation of leaf dry weight and LAI. Testing of the monitoring models with independent dataset indicated that the above spectral indices gave accurate growth estimation, with more reliable estimation from RVI（810,560）, GMI and SARVI（MSS）.Relation of leaf nitrogen status to grain protein index under different wheat cultivar and growth stages was revealed in this study. Results showed that grain protein character at maturity could be forecasted by plant nitrogen status of pre-maturity, with anthesis as ideal proper stage. Based on the technical route of key spectral parameters-leaf N indices-grain protein contents, total predicting models on grain protein content were constructed by linking the two set of models with leaf N nutrition as intersection, namely monitoring model on leaf nitrogen status with hyperspectral remote sensing and predicting model on grain protein content based on leaf nitrogen status. Testing of the predicting models with independent datasets indicated that the spectral indices of REP_LE, mND705, SDr/SDb, and FD742 at anthesis gave accurate estimation of grain protein contents in wheat, with more reliable estimation from mND705.Based on the biological mechanism of yield formation, relationship of leaf nitrogen status to grain yield was compared among different growth stages under different wheat cultivars across two years. The results showed that there were significant relationships of grain yield to leaf N accumulation and leaf area N index at initial filling, and the sum of leaf N content, leaf N accumulation and leaf area N index from booting to maturity were reliable indices for predicting yield. Based on the technical-route of characteristic spectral parameters-leaf N nutrition-grain yield, predicting models on grain yield were constructed with canopy hyper-spectral parameters at initial filling and cumulative value of key spectral parameters from booting to maturity in wheat by linking the two sets of models with leaf N nutrition as intersection, namely monitoring model on leaf nitrogen status with hyperspectral remote sensing and predicting model on grain yield based on leaf nitrogen status. Testing of the predicting models with independent two-year dataset indicated that the above linked models gave accurate yield estimation.The regression analyses between vegetation indices and plant N accumulation indicated that several key spectral parameters could accurately estimate the changes in plant N status across different growth stages, nitrogen levels and growing seasons, with same spectral parameters for each wheat cultivar. The cumulative value of plant N accumulation from anthesis to specific day were highly correlated with grain N accumulation at corresponding day, and monitoring models on grain N accumulation were constructed with plant N accumulation during filling. Total monitoring models on above-ground N accumulation during filling period were established using canopy hyper-spectral parameters by adding grain N accumulation to plant N accumulation. Tests with other independent dataset showed that several key spectral indices such as SDr/SDb, VOG2, VOG3, RVI（810,560）, [(R_750-800)/(R_695-740)]-1 and Dr/Db could be used to predict above-ground nitrogen accumulation at both pre-anthesis and grain filling.更多还原