摘要
目的:挖掘潜在的高光谱信息,建立精确的高光谱与水稻地上生物量(Above ground biomass,AGB)估测模型,方便农田管理者指导田间施肥。方法:以安徽科技学院小岗村现代生态农业研究所种植的水稻为研究对象,使用ASD FieldSpec■HandHeld^(TM) 2(HH2)光谱仪获得水稻孕穗后期抽穗前期345~1075 nm波段的光谱数据。经ViewSpecPro系统处理该生育时期的水稻冠层高光谱反射率数据,进行一阶和二阶变换,将各阶的光谱数据与地上部生物量进行相关性分析,筛选相关系数绝对值最大的5个波段,通过0阶、1阶和2阶微分光谱敏感波段的反射率与AGB数据拟合分析,建立水稻地上部生物量估测模型,并进行精度验证。结果:通过水稻AGB与3种变换的冠层光谱波段的相关性分析,表明0阶、1阶和2阶光谱反射率与AGB相关系数最大值出现的波段不同,分别为(658 nm)0.67、(768 nm)0.79、(763 nm)0.65;通过水稻AGB与3种变换的冠层敏感光谱波段的拟合分析,表明0阶、1阶和2阶光谱反射率与AGB拟合的最佳模型构建方式不同,对应的建模集R^(2)分别为0.69、0.80、0.41,RMSE分别为1571.69 kg/hm^(2)、1371.30 kg/hm^(2)、2039.22 kg/hm^(2),NRMSE分别16.8%、14.7%、21.8%;验证集中R^(2)分别为0.80、0.83、0.68,RMSE分别为993.50 kg/hm^(2)、890.68 kg/hm^(2)、1235.22 kg/hm^(2),NRMSE分别11.0%、9.8%、13.6%。结论:0阶和1阶中模型标准均方根误差都处于20%以内,说明模型的稳定性好,预测精度高,其中采用1阶幂函数可以更好地快速预测水稻地上生物量(验证精度高达到9.8%)。
Objective:To dig potential hyperspectral information to create accurate hyperspectral and rice above-ground biomass(AGB)estimation models for field managers to guide field fertilization.Methods:Rice planted in the modern ecological agriculture institute of Xiaogang Village of Anhui Science and Technology University was taken as a research object,and an ASD FieldSpec■ HandHeld^(TM) 2(HH2)spectrometer was used for obtaining spectral data of 345-1075 nm wave bands at the early stage of booting at the late stage of booting of the rice.Processing the rice canopy high spectral reflectance data in the growth period through a ViewSpecPro system,performing first-order and second-order transformation,performing correlation analysis on the spectral data of each order and the aboveground biomass,screening 5 wave bands with the maximum absolute value of correlation coefficient,establishing a rice aboveground biomass estimation model through fitting analysis of the reflectance of 0-order,1-order and 2-order differential spectrum sensitive wave bands and AGB data,and performing precision verification.Results:The correlation analysis between rice AGB and the three transformed canopy spectral bands showed that the bands with the maximum correlation coefficient between the 0th,1st and 2nd order spectral reflectance and AGB were 0.67(658 nm),0.79(768 nm)and 0.65(763 nm),respectively.The fitting analysis of rice AGB and three kinds of canopy sensitive spectral bands showed that the best model construction methods of 0-order,1-order and 2-order spectral reflectance were different from AGB.The corresponding modeling sets R^(2) were 0.69,0.80 and 0.41,RMSE were 1571.69 kg/hm^(2),1371.30 kg/hm^(2) and 2039.22 kg/hm^(2),NRMSE were 16.8%,14.7%and 21.8%,respectively.The validation set R^(2) were 0.80,0.83,0.68,RMSE were 993.50 kg/hm^(2),890.68 kg/hm^(2),1235.22 kg/hm^(2),NRMSE were 11.0%,9.8%,13.6%.Conclusion:The standard root-mean-square errors of the model in order 0 and order 1 were both within 20%,which indicates that the model had g
作者
陈小芳
李军
李新伟
周毅
CHEN Xiaofang;LI Jun;LI Xinwei;ZHOU Yi(College of Resources and Environment,Anhui Science and Technology University,Fengyang 233100,China)
出处
《安徽科技学院学报》
2021年第5期53-59,共7页
Journal of Anhui Science and Technology University
基金
新疆建设兵团绿洲生态重点实验室开放课题发展基金项目(201903)。
关键词
水稻
地上生物量
高光谱
模型
Rice
Above ground biomass
Hyperspectral
Model
