利用高光谱技术估测大豆育种材料的叶面积指数  

Prediction of Leaf Area Index Using Hyperspectral Remote Sensing in Breeding Programs of Soybean

作　　者：齐波[1] 张宁[1] 赵团结[1] 邢光南[1] 赵晋铭[1] 盖钧镒[1] 

机构地区：[1]南京农业大学大豆研究所/国家大豆改良中心/农业部大豆生物学与遗传育种重点实验室(综合)/作物遗传与种质创新重点实验室,江苏南京210095

出　　处：《作物学报》2015年第7期1073-1085,共13页Acta Agronomica Sinica

基　　金：国家重点基础研究发展计划(973计划)项目(2011CB1093);国家高技术研究发展计划(863计划)项目(2011AA10A105);国家公益性行业(农业)科研专项经费项目(201203026-4);教育部高等学校学科创新引智计划(111工程)项目(B08025);教育部创新团队项目(PCSRT13073);江苏省优势学科建设工程专项;江苏省现代作物生产协同创新中心项目(JCIC-MCP);中央高校基本科研业务费项目(KYZ201202-8)资助

摘　　要：叶面积指数（LAI）是反映田间作物长势及产量潜力的重要参数,规模化育种要求及时、快速、无损地获取大量育种材料的田间生长信息。本研究利用52份大豆品种（系）的2年田间试验,在盛花期（R2）、盛荚期（R4）及鼓粒初期（R5）测定大豆冠层反射光谱,同步测定大豆LAI和地上部生物量（ABM）。结果表明,不同生育期LAI与冠层光谱在可见光波段（426~710 nm）均表现显著负相关（P〈0.05）,在近红外波段（748~1331 nm）均表现为显著正相关（P〈0.05）。根据文献已报道的植被指数与LAI的线性相关性分析,NDVI和RVI类型的植被指数能够较好地指示大豆LAI,进而在全光谱250~2500 nm范围内涵盖上述两种类型的植被指数,经对建立的大豆LAI线性与非线性模型综合评价,遴选出不同生育期敏感植被指数的最优估测模型。其中,R2期RVI（825,586）所建模型（y=0.03x1.83）、R4期RVI（763,606）所建模型（y=0.38e0.14x）及R5期RVI（744,580）所建模型（y=0.06x1.79）的预测表现最好,决定系数（R2）分别为0.677、0.639和0.664,相对标准误（RRMSE）均小于20%;模型验证的决定系数（R2＊）分别为0.643、0.612和0.634,均根方误差（RRMSE＊）约20%。进而发现针对LAI和ABM的RVI共性核心波段组合为R2期的825 nm和586 nm、R4期的763 nm和606 nm以及R5期的744 nm和580 nm。本研究结果可望为大豆规模化育种中获取大量不设重复试验材料的田间长势信息提供快速无损预测的技术支持。Leaf area index（LAI） is an important parameter in observing field growth status and yield potential of crop plants, which is important in evaluating field growth performance of breeding lines in modern large scale plant breeding programs. The measurement of LAI and aboveground biomass（ABM） was synchronized with the information collection of the canopy hyperspectral reflectance at R2, R4, and R5 growth stages in a field experiment with 52 soybean varieties under completely randomized blocks design with three replications in two years. The results indicated that LAI have significant positive correlation with canopy spectral reflectance in the visible region（426–710 nm） and significant negative correlation in the near infrared region（748–1331 nm）（P0.05）. According to the linear correlation analysis between the vegetation indices and LAI in the literature, NDVI and RVI are superior vegetation indices for soybean LAI prediction. The linear and nonlinear regression models of LAI on NDVI and RVI vegetation indices were constructed and evaluated for all two-band combinations in the full spectral range of 350–2500 nm under 1 nm windows. Three single-stage regression models, i.e. R2 RVI（825, 586） model（y = 0.03x1.83）, R4 RVI（763, 606） model（y = 0.38e0.14x） and R5 RVI（744, 580） model（y = 0.06x1.79） were selected and validated as the best ones with fitness of 0.677, 0.639, 0.664 and less than 20% relative standard error, respectively, with their validation determination coefficients of 0.643, 0.612, 0.634, and around 20% validation standard error, respectively. Furthermore, the common core two–band combinations for both LAI and ABM prediction at R2, R4, and R5 were selected as 825 nm and 586 nm, 763 nm and 606 nm, and 744 nm and 580 nm, respectively. The obtained indices along with their prediction models can provide a technical support for quick and nondestructive field survey of soybean growth status in large scale breeding programs.

关 键 词：大豆 高光谱 遥感 叶面积指数 地上部生物量 

分 类 号：S565.1[农业科学—作物学]