基于光谱反射率的寒地水稻叶片氮含量预测

Prediction of Leaf Nitrogen Content of Rice in Cold Region Based on Spectral Reflectance

为实现利用水稻叶片光谱指数实时预测叶片群体的氮素含量,采集了不同年份中氮素、品种差异下寒地水稻主要生育期(T_(1)穗分化中期、T_(2)拔节期、T_(3)孕穗期、T_(4)齐穗期、T5蜡熟期)顶部3片全展叶(上1、上2、上3叶分别记作L1、L_(2)、L_(3))的光谱反射率,探究其变化规律以及光谱指数与叶片氮素含量的关系,并用P-k、均方根误差(RMSE)、对称平均绝对百分比误差(SMAPE)、校正均方根误差(RMSEC)、交互验证均方根误差(RMSECV)、相对预测性能(RPD)对模型精度进行验证。结果显示:提高氮肥投入量,叶片反射率在可见光区域内呈降低趋势,在近红外平台叶片反射率上升。随着生育期的推进,在可见光区域内,不同品种L1叶反射率先降低后上升,L_(2)、L_(3)叶的反射率一直上升,与叶片氮素百分含量的敏感波段为500~550和650~700nm。对光谱指标和叶片氮素百分含量进行相关分析,生育前期以下位叶片的光谱指标相关系数高,生育后期则相反,筛选出T_(1)时期L_(2)叶指标FD-NDNI、T_(2)时期L_(2)叶指标GM2、T_(3)时期L_(2)叶指标Lic2、T_(4)时期L1叶指标MRESRI以及T5时期L1叶指标Ctr1适宜作为不同时期预测叶片氮素含量的最佳指标,预测叶片氮素含量的回归方程R2分别0.54^(**)、0.60^(**)、0.66^(**)、0.62^(**)、0.51^(**),均达到极显著水平;验证指标的P-k值分别为0.00、0.04、0.06、0.01、0.04;RMSE分别为0.39、0.58、0.22、0.54、2.56;SMAPE分别为1.11、1.41、1.03、1.64、3.89;RMSEC分别为0.17、0.15、0.13、0.13、0.13;RMSECV分别为0.18、0.14、0.12、0.12、、0.14;RPD分别为2.46、2.19、3.15、1.74、3.01,其中T_(3)时期L_(2)叶指标Lic2的预测效果表现最佳。借助筛选的光谱指标能够实现快捷、无损和实时预测水稻不同生育时期的氮素营养状况,促进高产优质的寒地水稻可持续发展。

In order to realize the real-time prediction of nitrogen content in the rice leaf population by using the rice leaf spectral index,the spectral reflectance of the top three fully expanded leaves(upper 1,upper 2and upper 3leaves were recorded as L_(1),L_(2)and L_(3),respectively)at the main growth stages of rice in cold region(T_(1)mid-spike differentiation stage,T_(2)jointing stage,T_(3)booting stage,T_(4)full heading stage and T5 wax ripening stage)under different nitrogen and variety differences in different years were collected.The change rule and the relationship between spectral index and leaf nitrogen content were explored.P-k,Root Mean Square Error(RMSE),Symmetric Mean Absolute Percentage Error(SMAPE),Root Mean Square Error of Calibration(RMSEC),Root Mean Square Error of Interactive Verification(RMSECV)and Residual Prediction Deviation(RPD)were used to verify the accuracy of the model.The results showed that with the increase of nitrogen fertilizer input,the leaf reflectance decreased in the visible region,while the leaf reflectance increased in the near-infrared platform.With the advance of the growth period,in the visible light region,the reflectance of L_(1)leaves of different varieties decreased first and then increased,and the reflectance of L_(2)and L_(3)leaves increased all the time.The sensitive bands of leaf nitrogen percentage were 500~550and 650~700nm.The correlation analysis of the spectral index and leaf nitrogen percentage content showed that the correlation coefficient of the spectral index of the following leaves was high in the early stage of growth,but it was the opposite in the later stage of growth.The L_(2)leaf index FD-NDNI in the T_(21)period,L_(2)leaf index GM2in the T_(2)period,L_(2)leaf index Lic2 in the T_(3)period,L_(1)leaf index MRESRI in the T_(4)period,and L_(1)leaf index Ctr1in the T5period were selected as the best indexes to predict leaf nitrogen content in different periods.The regression equations R2 for predicting leaf nitrogen content were 0.54^(**),0.60^(**),0.66^(**),0.62^(**),and 0.51^(**),respectively,which reached extremely significant levels.The P-k values of the validation indexes were 0.00,0.04,0.06,0.01and 0.04,respectively.RMSE were 0.39,0.58,0.22,0.54,2.56,SMAPE were 1.11,1.41,1.03,1.64,3.89,RMSEC were 0.17,0.15,0.13,0.13,0.13,RMSECV were 0.18,0.14,0.12,0.12,0.14,the RPD were 2.46,2.19,3.15,1.74and 3.01,respectively.Among them,the prediction effect of the L_(2)leaf index Lic2at the T_(3)stage was the best.In summary,with the help of the selected spectral indicators,the nitrogen nutrition status of rice at different growth stages can be predicted quickly,non-destructively,and in real-time,and the sustainable development of high-yield and high-quality cold rice can be promoted.