| Nitrogen is an essential nutrient element in the growth and development of apple trees.It not only affects the growth and development of apple trees,but also affects yield and fruit quality.Therefore,it is of great significance to accurately monitor the nitrogen content of apple trees and scientific fertilization.Traditionally,apple tree nitrogen monitoring mainly relies on farmer experience or field investigation and sampling by scientific and technical personnel.Although this method is highly accurate,it is time-consuming and labor-intensive,and it is difficult to meet the needs of large-scale apple orchard nitrogen real-time monitoring.The abundance and deficiency of nutrients such as nitrogen will cause changes in the physicochemical status of the plant,which will cause the plant's spectral response to change,which provides a possibility for rapid extraction of vegetation information using remote sensing technology.In recent years,researches on remote sensing monitoring of vegetation nutrients are mostly based on near-Earth platforms.Their small data scale and high cost determine that the research results are difficult to apply to large-scale monitoring.At the same time,nutrient estimation models constructed based on hyperspectral remote sensing information are also difficult to directly applied to multispectral images.Therefore,it is of great significance to use near-ground measured hyperspectral data to simulate multispectral satellite remote sensing data and to monitor the nitrogen content of apple trees in a timely and large-scale manner.In this study,Apple Orchard in Yantai City,Shandong Province was used as the research area.And red Fuji apple trees in the full fruit period were used as the research objects.Data were collected during the New Shoots Flourishing Stage,Spring Shoots Stop Growing Stage,and Autumn Shoots Stop Growing Stage,respectively.Landsat-8,Sentinel-2A and GF-6satellite sensors were used to estimate the accuracy difference of apple tree nitrogen content.Because the study involved multiple phenological periods,there was a lack of synchronized image data from the study area during the sampling date,and it was necessary to avoid variations in the accuracy of nitrogen estimates for each band due to differences in spatial resolution of the satellite sensors.Therefore,based on the response function of Landsat-8,Sentinel-2A,and GF-6 satellite frequency bands,the ground-level spectral data were re-sampled,and the simulation data of these satellites were obtained.Using different multispectral satellite remote sensing simulation data,the normalized vegetation index of different band combinations was constructed,and the correlation analysis was performed with the nitrogen content of each phenological period.The optimal combination of three satellite simulation data in different phenological periods was selected.The correlation between the optimal band combination of different satellite sensors and nitrogen content was analyzed.Based on the selected optimal band combination,the model for estimating the canopy nitrogen content of the apple tree canopy was established based on Landsat-8,Sentinel-2A and GF-6 satellite simulation data.Finally,the nitrogen content estimation models of apple tree were established by using the full bands of different multi-spectral satellite remote sensing simulation data,and the differences of nitrogen estimation accuracy between Landsat-8,Sentinel-2A,and GF-6 satellite sensors were discussed.The main research conclusions are as follows:(1)The effects of phenological period on the nitrogen content and spectral reflectance of apple trees were found.The maximum,minimum,and average values of nitrogen content in different phenological stages were compared.The study found that the nitrogen content of the apple canopy showed a gradual decrease from the peak of the new shoot,the stop of the spring shoot to the stop of the autumn shoot.The difference of nitrogen content between the peak and the long periods of spring shoots is relatively small,and the nitrogen content of apple trees will reach a relatively stable state during the period of Spring Shoots Stop Growing Stage.The comparison and analysis of the canopy hyperspectral curves of these three phenological stages revealed that the trends and characteristics of the canopy spectral curves of these three phenological stages were the same.But the canopy reflectance was different for each phenological phase.And the spectral reflectance of the canopy was the highest in the new peak season,which was slightly higher than that of the spring shoots stop growing period,while the reflectance of the autumn shoots stop period was the lowest.The results verified the significant effect of phenological period on the nitrogen content and spectral reflectance of apple trees,indicating that the estimation of nitrogen content in apple canopy needs to consider the important factor of phenological period.(2)Obtained satellite simulation data of ground canopy spectral data resampling.Resampling the measured ground canopy hyperspectral data using the satellite sensor's band response function to obtain the satellite simulation data of Landsat-8,Sentinel-2A and GF-6.The Landsat-8 satellite simulation data includes coastal,blue,green,red and near-infrared bands.Sentinel-2A satellite simulation data includes coastal,blue,green,red,red edge and near-infrared bands.GF-6 satellite simulation data mainly includes purple,blue,green,yellow,and red light bands,as well as red edge and near infrared bands.(3)The optimal frequency band combination and correlation coefficient of different satellite simulation data were analyzed.Based on the Landsat-8,Sentinel-2A and GF-6 satellite simulation data,the optimal frequency band combinations at different phenological stages were screened.As a result,the optimal band combinations and their correlation coefficients were significantly different in different phenological periods,further proving the phenological period impacted on nitrogen content estimates.Considering the impact of the phenological period comprehensively,comparing the correlation between different sensor band combinations and nitrogen content,the results showed that,compared with the correlation between Landsat-8 satellite simulation data and nitrogen,The correlation between band combinations based on Sentinel-2A and GF6satellite simulation data and nitrogen content has been significantly improved.(4)The accuracy of the nitrogen estimation model based on the optimum band combination of different sensors was compared.Taking full account of the influencing factors of phenological period,the average coefficient of the nitrogen content estimation model for the optimal band combination of Landsat-8,Sentinel-2A and GF-6 satellite simulation data in different phenological periods was calculated more objectively.The differences in nitrogen estimates from three satellite sensors were accurately compared and analyzed.The results showed that,the average determination coefficient of the nitrogen estimation model for the optimal band combination of Landsat-8 satellite simulation data was 0.49,compared with 0.58 for Sentinel-2A satellite simulation data,and which for GF-6 satellite simulation data was 0.57.The precision of the nitrogen content estimation model based on the optimal band combination of Sentinel-2A and GF-6 satellite simulation data was significantly better than that of Landsat-8 satellite simulation data.(5)The satellite with the highest estimation accuracy in the Landsat-8,Sentinel-2A,and GF-6 satellite simulation dataA nitrogen content estimation model was established using the full band of Landsat-8,Sentinel-2A,and GF-6 satellite simulation data.To ensure a more objective and accurate evaluation of the nitrogen estimation performance of these three satellite sensors,R~2,RMSE and RPD were averaged.Through analysis,it could be found that the estimation accuracy of the estimation model based on the Landsat-8 satellite simulation data was the lowest.The support vector machine model based on the Sentinel-2A and GF-6 satellite simulation data was significantly better than that based on the Landat-8 data.In the support vector machine model,the model built using the Sentinel-2A was slightly better than the model built using the GF-6.According to the R~2 and RMSE of the model testing,it was found that the results of the model testing and the model are consistent.Rough estimates of nitrogen content can only be made using landsat-8 satellite simulation data.The relative analysis error is less than 2.00.Both the Sentinel-2A and GF-6 satellite simulation data model testing have a R~2 of 0.61,and their RPD were greater than 2.50,indicating that both Sentinel-2A and GF-6 satellite simulation data can better estimate the nitrogen content of apple trees.However,the RMSE for the GF-6 satellite simulation data model testing was larger than that for the Sentinel-2A model.In general,the nitrogen estimation accuracy of Sentinel-2A and GF-6 satellite simulation data was better than that of the estimation model established by Landsat-8 satellite simulation data. |
PDF Full Text Request