Hyperspectral Characteristics And Nitrogen Content Estimation Of Apple Leaves At Different Phenological Phases

Nitrogen is one of the indispensable nutrient elements in the growth and development of plants,which plays a key role in the growth,yield and quality of apple trees.At present,most of the existing studies on nitrogen content in apple trees focus on a specific phenological stage of fruit trees,while few studies on different phenological stages of fruit trees.Based on the researches published before,the hyperspectral characteristics and nitrogen content of apple leaves in different phenological periods are monitored rapidly and non-destructively,so as to determine the best estimation period of hyperspectral characteristics and nitrogen content of leaves in different phenological periods.It is of great significance for precision fertilization of fruit trees.Shuangquan Town,Changqing District,Jinan City,Shandong Province was taken as the research area,and Fuji apple trees were taken as the research object.The ASD FieldSpec 4 portable spectral instrument was used to measure the spectral data of leaves in 2019 at four phenophases of the apple tree.Including the Full Flowering Period(late April),the Growing Period of New Shoots(mid-late May),the Stopping Period of Spring Shoots(mid-late June)and the Stopping Period of Autumn Shoots(mid-late September).And then the leaf nitrogen content of apple tree was measured indoors.Physiological changes of nitrogen content in apple leaves at different phenological stages and spectral characteristics of apple leaves with different nitrogen content were analyzed.The pretreatment method includes multiple scattering correction and fractional differential transformation of the original spectral data,and screening the sensitive wavelength related to the nitrogen content.At the same time,7 vegetation indices were selected.Finally,the nitrogen content estimation model of apple canopy at different phenological periods was established.The main research results are as follows:(1)Clarified the changes of hyperspectral characteristics and nitrogen content in apple leaves at different phenological periods.The nitrogen content of apple leaves is increasing from the Full Flowering Period to the Growing Period of New Shoots,and the nitrogen content of apple leaves has been decreasing from the Growing Period of New Shoots to the Stopping Period of Autumn Shoots.Among them,the reduction of nitrogen content from the Growing Period of New Shoots to Stopping Period of Spring Shoots is relatively small,which is relatively stable in the middle stage of the sampling period.Therefore,the Growing Period of New Shoots and Stopping Period of Spring Shoots can be used as suitable sampling periods for nitrogen analysis of apple leaves.(2)The spectral differences of apple tree leaves at different phenological periods were proved.The spectral reflectance of apple leaves in different phenological periods has a similar trend.In the range of 350-780 nm,the reflectivity of the four phenological periods differs slightly.The spectral reflectance in the range of 780-1360 nm is obviously different,and the spectral reflectance in the Full Flowering Period is obviously lower than that in the other three periods.However,the difference in reflectance was the largest among the Growing Period of New Shoots,the Stopping Period of Spring Shoots and the Stopping Period of Autumn Shoots.And the order of reflectance is Stopping Period of Spring Shoots>Stopping Period of Autumn Shoots>Full Flowering Period>Growing Period of New Shoots.In the range of 1420-2500 nm,the spectral reflectance of the Stopping Period of Spring Shoots is higher than other periods.The other spectral difference is small,and almost all the spectral curves overlap.The above provides a reference for the hyperspectral study of apple leaves,and also provides a theoretical basis and technical support for the application of apple leaf spectra for nutrition diagnosis and remote sensing inversion.(3)An optimal model for estimating the nitrogen content in apple leaves at different phenological stages was established.For the Full Flowering Period,the optimal estimation model was the Random Forest model based on fractional differential first-order processing,which the determination coefficient R2 of modeling set was 0.851,with RMSE of 0.0728 and RPD of 2.1442.And the determination coefficient R2 of the fitting result of the verification set was 0.684,with RMSE of 0.1290 and RPD of 1.6005.For the Growing Period of New Shoots,the optimal estimation model was the Random Forest model based on fractional differential second-order basic processing,which the determination coefficient R2 of modeling set was 0.907,with RMSE of 0.1547 and RPD of 2.0285.And the determination coefficient R2 of the fitting result of the verification set was 0.789,with RMSE of 0.1243 and RPD of 1.7021.During the Growing Period of New Shoots,it was the best estimation model of nitrogen content in leaves.For the Stopping Period of Spring Shoots,the optimal estimation model of leaf nitrogen content was the Random Forest model based on fractional differential second-order basic treatment.Which the determination coefficient R2 of modeling set was 0.891,with RMSE of 0.0841 and RPD of 2.1396.And the determination coefficient R2 of the fitting result of the verification set was 0.617,with RMSE of 0.1251 and RPD of 1.1705.For the Stopping Period of Autumn Shoots,the optimal estimation model of leaf nitrogen content was the Random Forest model based on fractional differential 1.5 order basic treatment.The fitting result of modeling set has a determination coefficient R2 of 0.8890,with RMSE of 0.1315 and RPD of 2.2092.And the determination coefficient R2 of the fitting result of the verification set was 0.787,with RMSE of 0.1469 and RPD of 1.3444.From the accuracy results of these four phenological periods,the optimal estimation models are all based on the Random Forest regression model under fractional differential processing.