【作者】 杨杰；

【导师】 曹卫星；

【作者基本信息】 南京农业大学 ， 生态学， 2009， 博士

【摘要】 作物生长光谱监测的主要任务是要确立能够反映作物生长状态的敏感波段和特征指数,并建立光谱指数与农学参数之间的定量关系。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,为定量估测植株单一生化组分状况提供了有效途径。本研究以水稻为对象,以系列田间试验为依托,综合运用高光谱信息分析、生理生化测试及数理统计建模等技术手段,分析不同氮素水平和品种条件下水稻叶片高光谱反射特征与叶片和群体氮素及色素之间的定量关系,进而确立基于叶片高光谱监测水稻单叶和群体氮素及色素之间的适宜光谱指数及相应监测模型,从而为水稻氮素营养的无损监测和精确诊断提供理论基础和关键技术。首先分析了不同施氮量和不同氮含量水平下水稻叶片光谱反射特征的变化模式。结果表明,随着土壤施氮量和叶片氮含量水平的提高,叶片反射光谱在可见光区的反射率下降,负相关显著；而近红外波段反射率则小幅上升,但相关性不高。叶片氮含量与一阶导数的相关关系表明在可见光至红边区众多波段均达到极显著相关水平,相关系数呈明显的峰谷特征。可见,不同试验条件下水稻叶片的氮素状况和反射光谱特征呈现了明显的动态变化模式,为进一步定量解析叶片反射光谱与氮素营养状况的关系提供了丰富的信息基础。基于叶片高光谱特征,系统分析了水稻叶片氮素组分与叶片反射光谱、一阶导数光谱、两波段组合的比值(SR)、归一化(ND)及差值(SD)等光谱指数的相关关系。叶片全氮和蛋白氮含量的敏感波段主要位于可见光绿光区520-590 nm及红边区域695-715 nm,其中红边区域表现最为显著。通过比较不同波段组合的不同形式光谱参数发现,以700-702 nm附近波段与近红外短波段的比值组合估算水稻上部叶片的全氮和蛋白氮含量的效果最好,其次为黄光区583-587 nm左右波段与近红外短波段的比值组合。其中,窄波段比值指数SR(R780, R702)和SR(R770, R700)分别为估算水稻叶片全氮和蛋白氮含量的最佳光谱变量,而基于有效组合区域内光谱指数敏感性的宽波段比值指数SR[AR(763-860), AR(697-707)]和SR[AR(746-815), AR(697-705)]分别用于全氮和蛋白氮含量估算,表现出与窄波段组合相似的敏感性和预测力,表明在此敏感组合区域内带宽选择对反演结果的准确度影响不大。最佳差值和归一化差值指数仅出现在740-755 nm小范围内,组合波段邻近,且与最佳一阶导数波段相近,但总体表现均不及比值组合优秀。系统分析了已有叶绿素敏感光谱指数及新型两波段组合的比值和归一化光谱指数与叶片叶绿素含量的关系,提出了水稻叶片不同组分叶绿素含量的敏感光谱指数及预测方程。发现红边波段构成的比值或归一化光谱指数均可较好地指示水稻上部4叶的叶绿素含量。估算叶绿素a(Chla)和叶绿素总量(Ch1a+b)的敏感区域一致,最佳比值指数均为SR(R730, R710),最佳归一化指数分别为ND(R780, R710)和ND(R780, R712);估算叶绿素b (Ch1b)的最佳比值和归一化光谱指数分别为SR(R780, R725)和ND(R780, R725)。此外,引入445 nm波段反射率对上述光谱指数进行修正,可以降低模型的预测误差,提高模型的稳定性。当宽波段比值指数SR[AR(720-740),AR(705-715)]用于叶片Ch1a和Cb1a+b含量估算,SR[AR(750-850), AR(715-735)]用于叶片Chlb含量估算,以及宽波段归一化指数ND[AR(750-850), AR(705-715)]、ND[AR(750-850), AR(706-718)]和ND[AR(750-850), AR(715-735)]分别用于Ch1a、Cb1a+b和Ch1b含量的估算时,均表现出敏感度和稳定性的统一,且波段选择较灵活,从而有助于指导便携式叶绿素监测仪的研制开发。进一步分析了不同叶位叶片光谱或单叶光谱组合与水稻群体叶片氮素组分的关系,明确了基于叶片尺度光谱监测群体叶片全氮和蛋白氮含量的适宜取样叶位、敏感波段及光谱指数,并建立了相应的监测模型。基于叶片光谱监测群体叶片全氮含量和蛋白氮含量时,顶2叶和顶3叶为适宜取样叶位,而顶2叶和顶3叶的光谱平均值(L23)为最理想的叶片光谱形式；敏感波段主要位于可见光绿光及红边区；可见光580 nm附近波段及红边702 nm附近波段与近红外短波段的比值组合和群体叶片全氮及蛋白氮含量关系最为密切。估算全氮含量以绿光窄波段指数SR(R780, R580)和宽波段指数SR[AR(750-850), AR(568-588)表现最好；而估算蛋白氮含量则以红边窄波段指数SR(R780, R701)和宽波段指数SR[AR(750-850), AR(697-706)]表现最好。利用叶片特征光谱指数可以对群体叶片全氮和蛋白氮含量进行准确可靠的监测。基于不同生态点、年份、品种和施氮水平下4个大田试验资料,通过比较不同叶位(组合)叶片光谱指数与群体叶片Ch1a和Ch1a+b含量的关系,确立了适宜于监测群体叶片Ch1a和Ch1a+b含量的关键叶位、敏感光谱指数及监测方程。不同叶位叶片光谱对群体叶片叶绿素含量的估算效果存在明显差异,基于顶2叶和顶3叶平均光谱构建的敏感光谱指数和监测方程表现较好。比值和归一化指数均可较好的估算群体叶片叶绿素含量,但最佳组合的中心波段有所差别,归一化组合以560±10 nm vs. NIR和710±6 nm vs. NIR表现较好,比值组合以554±10 nm vs. NIR和718±6 nm vs. NIR表现较好。由此,提出绿光归一化指数ND(R776, R560)、和ND[R(750-850), R(550-850)]、绿光比值指数SR(R554, R776)和SR[R(544-564), R(750-850)]、红边归一化指数ND(R780, R710)和ND[R(750-850), R(704-716)]、以及红边比值指数SR(R718, R780)和SR[R(712-724), R(750-850)]可用于水稻群体叶片Ch1a和Ch1a+b含量的估算,模型校正及检验结果均显示了各参数的可靠性和适用性,尤以红边组合指数的敏感性表现最好。通过分析水稻单叶类胡萝卜素(Car)含量和类胡萝卜素/叶绿素比值(Car/Ch1)与不同波段组合的多种类型光谱指数的相关性,构建了适于水稻单叶Car含量和Car/Ch1比值监测的敏感光谱指数及监测方程。结果显示,723 nm附近波段与近红外波段的比值组合及713 nn附近波段与近红外波段的归一化组合可以较好地监测水稻单叶Car含量,基于窄波段组合SR(R723, R770)和ND(R770, R713)及宽波段组合SR[AR(715-729), AR(750-820)]和ND[AR(740-840), AR(707-719)]的监测方程线性拟合效果较好,独立资料的检验亦表明各方程稳定性高,可以对不同条件下水稻单叶Car含量进行可靠的监测。由于水稻叶片Car/Ch1比值的变化与叶片衰老或胁迫程度的关系密切,因而提出SR(R698,R712)、ND(R716, R695)、SR(R615, R713)和ND(R737, R622)用于灌浆中后期衰老叶片Car/Ch1比值的监测,线性拟合关系良好。进一步分析水稻不同叶位叶片光谱指数与群体叶片Car含量间的关系,同样发现以顶2叶和顶3叶的平均光谱表现较好,不同波段组合的光谱指数中,以蓝光比值指数SR(R466, R496)和蓝光归一化指数ND(R466, R496)建立的监测方程表现最好,可用于不同生育期水稻群体叶片类胡萝卜素含量的定量监测。更多还原

【Abstract】 The primary task of crop growth spectral monitoring is to determine sensitive wavebands and characteristic parameters for reflecting crop growth status, and establish quantitative relationship between agronomic variables and spectral indices. In the past few years, the newly emerged hyperspectral remote sensing, with the characteristics of high resolution, consecutive wavebands and rich data, can significantly enhance the ability of detecting specific crop variables related to physiology and biochemistry, and have opened up new possibilities for quantifying single biochemical index in plant. In this study, five field experiments were conducted with different nitrogen (N) rates and rice cultivars across three growing seasons at different eco-sites. Based on analysis of leaf spectral information and assay of physico-chemical index in rice plant, the characteristics of leaf hyperspectral reflectance under different conditions and their correlation to leaf and canopy nitrogen status and correlative biochemical components in rice were quantified in this paper, then some new spectral indices and quantitative regression models were developed for estimating single leaf and canopy N and chlorophyll (Ch1) concentration. The prospective results would provide technical basis for non-destructive monitoring and precise diagnosis of wheat growth.Comparison of variation pattern in leaf reflectance under different N supply rates and leaf N concentration (LNC) showed that with increasing soil N supply rates and LNC, reflectance at visible band were decreased, presenting significant negative correlation; but reflectance at near infrared flat region (NIR,750-1300nm) were increased slightly, with weakly positive correlation. The correlation of leaf first derivative spectra to LNC indicated that highly significant correlation were reached at many bands located in visible and red edge regions, correlation spectrum presented obvious peak-valley characteristics. These changed patterns of leaf spectra and N status at different experimental condition provided a basis for analyzing and constructing quantitative relationships of leaf N nutrition to hyperspectral characters of leaf reflectance in rice. Based on technique of leaf hyperspectra analysis, many characteristic bands and derived spectral indices were obtained. The quantitative relationships of LNC and leaf protein N concentration (LPNC) to leaf reflectance spectra, first derivatives, and all combinations of two wavebands between 350 and 2500 nm as simple ratio spectral indices (SRs), normalized difference spectral indices (NDs) and simple difference spectral indices (SDs) were developed. The results indicated that the sensitivity bands mostly occurred in 520-590 nm within green light region and 695-715 nm within red edge region, and a close correlation existed between red-edge region and LNC and LPNC. Comparison of prediction ability of different algorithmic spectral indices indicated that the SRs were the most effective approach for predicting LNC and LPNC in the top leaves of rice plant, especially with the ratio of reflectances in the NIR to reflectances centered at 700-702 nm, and next to reflectances centered at 583-587 nm within yellow region. Two narrow bands spectral indices as SR(R.78o, R702) and SR(R770, R700) were developed to estimate LNC and LPNC, respectively; and further, based on sensitivity analysis of SRs located in effective combined region, tow broad band spectral indices as SR[AR(763-860), AR(697-707)] and SR[AR(746-815), AR(697-705)] were also determined to estimate LNC and LPNC, respectively, giving similar sensitivity and prediction ability with narrow SRs, indicated that selective bandwidth in effective region had little effect on prediction accuracy. The optimal SDs and NDs for estimating LNC and LPNC only located in a small regions of 740-755 nm, the wavebands for constructing spectral indices were adjacent, and close to optimal first derivatives, but they were all inferior to SRs.The relationships of leaf Ch1 concentration to new SRs and NDs with two wavelengths combinations, and existing Ch1 sensitive spectral indices were systematically analyzed, some sensitive spectral indices and monitoring equations were put forward for leaf Ch1 concentration estimation. Analysis showed that the best indicators for estimating leaf Ch1 concentration in rice were SRs and NDs calculated in the red edge region. The sensitive region for estimating of chlorophyll a (Ch1a) and total chlorophyll (Ch1a+b) concentration were consistent, the best SRs were uniform as SR(R730, R710), and the best NDs were ND(R780, R710) and ND(R780, R712), respectively; the best SRs and NDs for estimating chlorophyll b (Ch1b) were SR(R780, R725) and ND(R780, R725), respectively. In addition, modifying the above spectral indices with the reflectance at 445 nm could reduce the predicting error of models, and increased the extrapolation potential for the model. Some broad bands spectral indices were further developed for Ch1a, Ch1b and Ch1a+6 concentration estimation, respectively, as SR[AR(720-740), AR(705-715)] for leaf Chla and Ch1a+b and SR[AR(750-850), AR(715-735)] for leaf Chlb concentration estimation, and ND[AR(750-850), AR(705-715)], ND[AR(750-850), AR(706-718)] and ND[AR(750-850), AR(715-735)] for leaf Chla, Chla+b and Chlb concentration estimation, respectively, each of them had similar sensitivity as narrow bands spectral indices. This would help to development of portable Ch1 monitoring instrument.Further analysis were conducted on the relationships of leaf reflectance derived from different positions or single leaf spectral combination to canopy LNC and LPNC, and proper leaf position, key hyperspectral indices and quantitative monitoring model for accurate prediction of canopy LNC and LPNC were determined. The performance of different leaf hyperspectral indices for estimating canopy leaf N status were different with changed leaf position, top 2nd and 3rd leaf were ideal sampling position for monitoring canopy LNC and LPNC, average reflectance of top 2nd and 3rd leaf (L23) could help to improve the sensitivity and stability of key hyperspectral parameters, as ideal combination of key leaf position. The SRs combined of 702±nm vs. NIR were the most effective approach for predicting canopy LNC and LPNC in rice, and next were 580±nm vs. NIR. Green ratio indices as SR(R780, R580) and SR[AR(750-850), AR(568-588)] were developed for estimating canopy LNC; and red edge ratio indices as SR(R780, R701) and SR[AR(750-850), AR(697-706)] were developed for estimating canopy LPNC. The effect of model simulation and test indicated that these spectral indices constructed with leaf hyperspectral data could be effectively used for accurate estimation of canopy LNC and LPNC in rice under different growing conditions.Quantificational relationship between canopy leaf Chi concentration and leaf spectral indices derived from top four leaves at different growth position or their certain combination were also analyzed, and some key hyperspectral indices were developed for accurate prediction of canopy leaf Ch1 concentration, thus quantitative monitoring model for canopy leaf Ch1 concentration were determined. The results indicated that the performance of estimating canopy leaf Ch1 concentration by leaf hyperspectral indices were different with changed leaf position, average reflectance of top 2nd and 3rd leaf (L23) was more effective than others, as ideal selection of leaf spectrum. Both SRs and NDs were effective approach for predicting canopy leaf Chla and Chla+b concentration in rice, but centre waveband of optimal combination were different. The NDs located in 560±10 nm vs. NIR and 710±6 nm vs. NIR, and the SRs located in 554±10 nm vs. NIR and 718±6 nm vs. NIR, which were more remarkable than other wavebands combination. Thus green NDs as ND(R776, R560) and ND [R(750-850), R(550-570)], green SRs as SR(R554, R776) and SR[R(544-564), R(750-850)], red edge NDs as ND(R780, R710) and ND[R(750-850), R(704-716)], and red edge SRs as SR(R718, R780) and SR[R(712-724), R(750-850)] were developed to estimate canopy leaf Ch1a and Chla+b concentration, especially red edge wavebands combination, with higher sensitivity were more advised.The change characteristics of single leaf carotenoid (Car) concentration and carotenoid/chlorophyll ratio (Car/Ch1) in rice with development stages and quantitative relationships to leaf reflectance spectra and derived spectral indices were investigated. The results indicated that the SRs using reflectance around 723 nm combined with NIR or the NDs using reflectance around 713 nm combined with NIR could be used to estimate leaf Car concentration, among which the SR(R723, R770) and ND(R770, R713) have the best performance. Broad bands combinations as SR[AR(715-729), AR(750-820)] and ND[AR(740-840), AR(707-719)] in red edge region also have a good correlation with leaf Car concentration. Tests with independent dataset showed that leaf Car concentration in rice could be predicted effectively by above hyperspectral indices. Because changes of leaf Car/Ch1 ratio were direct related to leaf senescence or stress, SR(R698, R712), ND(R716, R695), SR(R615, R713) and ND(R737, R622) were developed for estimating leaf Car/Ch1 ratio in the mature stage, especially in the process of leaf senescence, but much work also remains to be done to test and perfect it. Further, Relationships of leaf hyperspectral indices to canopy leaf Car concentration were also analyzed, results show that average spectral reflectance of top 2nd and 3rd leaf was more suitable for monitoring canopy leaf Car concentration, blue spectral indices as SR(R466, R496) and ND(R466, R496), with a good accuracy and precision, were presented for canopy leaf Car concentration estimation.更多还原