融合叶绿素数据的棉花冠层光合速率高光谱估算建模  

Hyperspectral estimation modeling of photosynthetic rate in cotton canopy using chlorophyll data

作　　者：侯卜平 王家强[1] 李福庆 石靖 高菊 申栋妍 李克远 HOU Buping;WANG Jiaqiang;LI Fuqing;SHI Jing;GAO Ju;SHEN Dongyan;LI Keyuan(College of Agriculture,Tarim University,Key Laboratory of Genetic Improvement and Efficient Production of Characteristic Crops in Arid Areas of Southern Xinjiang,Research Center of Oasis Agricultural Resources and Environment of Southern Xinjiang,Alar,Xinjiang 843300,China)

机构地区：[1]塔里木大学农学院/南疆干旱区特色作物遗传改良与高效生产兵团重点实验室/南疆绿洲农业资源与环境研究中心,新疆阿拉尔843300

出　　处：《干旱地区农业研究》2025年第1期203-212,共10页Agricultural Research in the Arid Areas

基　　金：国家自然科学基金(32360547);新疆生产建设兵团科技创新人才计划项目(2022CB001-07);新疆生产建设兵团科技计划项目(2021DB019);阿拉尔市农业绿色高质量发展创新战略联盟(2021BB024)。

摘　　要：通过设置不同的灌水量梯度,获取棉花5个生育时期(蕾期、初花期、盛花期、花铃期、盛铃期)冠层的光谱反射率、叶绿素密度和叶片净光合速率(P_(n))数据,利用支持向量机(SVM)模型和随机森林(RF)模型,建立融合叶绿素密度数据和未融合叶绿素密度数据的冠层光合速率预测模型。结果表明:在水分胁迫下,叶绿素密度与净光合速率呈正相关关系;用CARS+SPA算法重复执行的方式进行特征波段筛选,降维效果显著,剔除冗余波段效率高,盛花期特征波段为332、347、416、466、672、695、711、733、752、848、954 nm和1069 nm。模型监测结果表明,融合叶绿素密度数据的模型拟合度优于未融合叶绿素的模型;比较不同模型的估算能力和模型精度,随机森林(RF)模型均优于支持向量机(SVM)模型;融合叶绿素密度的RF模型5个生育时期的建模集R^(2)分别为0.659、0.676、0.808、0.744和0.633,验证集R^(2)分别为0.635、0.675、0.786、0.725和0.627。与未融合叶绿素密度数据的模型相比,融合叶绿素密度数据模型建模集的R^(2)平均提高5.59%,RMSE平均降低2.92%,RPD平均提高7.26%;验证集的R^(2)平均提高4.12%,RMSE平均降低1.64%,RPD平均提高5.27%,表明融合叶绿素密度数据的棉花冠层光合速率光谱估测模型具有更高的拟合精度和稳定性。By applying different irrigation gradients,data on spectral reflectance,chlorophyll density,and canopy net photosynthetic rate(P_(n))were collected across five growth stages of cotton:budding stage,beginning flower stage,full blossom stage,blossing and boll-forming stage,and full bloom stage.A canopy photosynthetic rate prediction model was developed using support vector machine(SVM)and random forest(RF)algorithms,incorporating both chlorophyll-fused and non-fused data.The results demonstrated a positive correlation between chlorophyll density and the net photosynthetic rate under water stress.The CARS+SPA algorithm was employed to repeatedly perform feature band screening,achieving a remarkable dimension reduction effect and high efficiency in eliminating redundant bands.The feature bands were 332,347,416,466,672,695,711,733,752,848,954 nm and 1069 nm in full blossom stage.The monitoring results of the model showed that the model fitting degree of the fused chlorophyll data was better than that of the unfused chlorophyll data.Compared with the estimation ability and model accuracy of different models,the random forest(RF)model was superior to the support vector machine(SVM)model.The R^(2) of the calibration set of the RF model fusing chlorophyll density in the five growth periods were 0.659,0.676,0.808,0.744 and 0.633,respectively,and the R^(2) of the validation set were 0.635,0.675,0.786,0.725 and 0.627,respectively.Compared with the model without chlorophyll density data,the R^(2) of the calibration set increased by 5.59%on average,the RMSE decreased by 2.92%on average,and the RPD increased by 7.26%on average.The average R^(2) of the validation set was 4.12%higher than that of the unfused chlorophyll data,the RMSE was reduced by 1.64%on average,and the RPD was increased by 5.27%on average.The analysis demonstrated that the spectral estimation model of the cotton canopy photosynthetic rate,integrated with chlorophyll density data,exhibits superior fitting accuracy and stability.

关 键 词：棉花 光合速率 叶绿素密度 特征波段选择 光谱估测模型 

分 类 号：S562[农业科学—作物学] S127