Inversion Of Canopy Nitrogen Content In Apple Orchards Based On Differential Fusion Of Remote Sensing Data

Nitrogen is a key element in apple growth and development,which is closely related to apple yield and quality.The nitrogen content was directly affected by the amount of nitrogen applied in apple trees.Excessive application of N fertilizer not only resulted in the decrease of N use efficiency,but also had a negative effect on yield and quality.Precise nitrogen management is beneficial to balance nitrogen application rate and apple yield and quality,and achieve reasonable nitrogen use efficiency.Dynamic monitoring of nitrogen content in apple was an effective way to optimize nitrogen supply strategy.Therefore,accurate and rapid estimation of regional apple canopy nitrogen content is of great significance for mastering the nutritional status of apple trees,carrying out apple production and management more effectively,and improving apple yield and quality.However,inversion of regional canopy nitrogen content using single remote sensing data such as satellite or UAV is still challenging in terms of accuracy and monitoring range.In this study,the nonnegative matrix factorization(NMF)method was used to fuse satellite and UAV remote sensing data to perform canopy nitrogen content inversion in apple orchards.The apple orchard in Guanli Town,Qixia City,Shandong Province was used as the research area,and the long-term red Fuji apple tree was used as the research object.Firstly,the DJI Matrice 600 PRO UAV platform is equipped with Cubert UHD 185-Firefly(UHD185)hyperspectral sensor to obtain hyperspectral remote sensing data of apple orchards,and combined with the multispectral remote sensing image data of Planet satellite in the same period,the Planet satellite image and UAV hyperspectral remote sensing image are preprocessed,and the canopy layer of apple orchard and sample orchard in Guanli Town is identified respectively,and the spectral reflectance is extracted.Secondly,the hyperspectral remote sensing data of UAV is resampled to the Planet satellite band,and the spectral variables based on the Planet satellite data and simulation data are constructed and screened respectively.Then,the NMF method was used to decompose Planet satellite variables(PSVs)and simulated spectral variables(SSVs),and the spectral characteristic base matrix and differential correction coefficient matrix of nitrogen content were extracted respectively,and the extracted base matrix and coefficient matrix were multiplied to reconstruct the fusion variables(FSVs),which were used to construct the regional apple canopy nitrogen content inversion model to realize the fusion of remote sensing data at different scales.Finally,an inversion model of canopy nitrogen content in orchards based on NMF differential fusion is constructed,and an inversion map of canopy nitrogen content in Guanli Town is generated based on satellite remote sensing data to realize regional apple tree nutrition status monitoring.The main findings are as follows:(1)the canopy spectral characteristic variables of apple trees were constructed and screened.Based on the simulated spectral data and Planet image data,9 planting quilt indices were constructed.They were Difference vegetation index(DVI),Ratio vegetation index(RVI),Normal difference vegetation index(NDV),Enhanced Vegetation Index 2(EVI2),Wide dynamic range vegetation index(WDRVI),modified simple ratio vegetation index(MSR),Soil-adjusted vegetation index(OSAVI),Modified chlorophyll absorption index(MCARI),Modified triangular vegetation index 2(MTVI2).The 11 spectral parameters include,1/(Ri+Rj),1/(Ri-Rj),(?),Ri+Rj,Ri·Rj,Ri/(Ri+Rj),Ri/(Ri-Rj,Ri·Rj·Rq,i,j and q were respectively any one of the four bands,with a total of 20 spectral variables.Then the correlation analysis with canopy nitrogen content was carried out,and the spectral variables with a correlation greater than 0.6 were screened out,a total of 8 kinds,which were R630/R545,(R545-R485)/(R545+R485),(R820-R630)/(R820+R630),OSAVI,MTVI2,R545/(R545+R485),R630/(R545+R630),R545/(R545R485).(2)Planet satellite remote sensing data and UAV remote sensing data are integrated based on NMF method.In this study,NMF method is used to fuse satellite and UAV remote sensing data.Synthesizing the index parameters such as sparsity,residual and iterative times,the offset optimization algorithm is used to decompose PSVs and SSVs alternately.The decomposition level K of NMF algorithm is determined by two important parameters:Cophenetic and sum of squares of residuals(RSS).When K is 5,the spectral characteristic base matrix(WSSVs)based on SSVs and the sample differential coefficient correction matrix(HPSVs)based on PSVs are the most similar to the target matrix.Taking the product of WSSVs and HPSVs as FSVs,the data fusion of SSVs and PSVs is realized.(3)Establishment and test of regional canopy nitrogen content inversion model based on fusion variables.According to the ratio of 2:1,the 72 samples obtained were divided into modeling data set and validation data set to establish and verify the canopy nitrogen content inversion model.The PLSR method was used to construct a regional canopy nitrogen content inversion model based on the original satellite spectral variable and the fusion spectral variable,and the accuracy was compared.The differential fusion model had a calibration R2 and RMSE of 0.706 and 0.109,respectively,and a verification R2,RMSE and RPD of 0.685,0.160 and 1.624,respectively,while the unfused model had a correction R2 of 0.640,RMSE of 0.118,a validation R2 of 0.571,an RMSE of 0.171,and an RPD of 1.502.The results show that the NMF differential fusion method can effectively improve the model inversion accuracy and the results are more stable.Compared with the ratio mean method and linear conversion method,the NMF differential fusion method has better effect in improving the accuracy of inversion of regional canopy nitrogen content based on satellite remote sensing.The results show that the NMF differential fusion method can effectively improve the accuracy of inversion of the nitrogen content of the regional canopy based on satellite images by introducing UAV hyperspectral information into satellite remote sensing data.The differential fusion model was used to combine the distribution of apple orchards in Guanli Town to perform the inversion of nitrogen content in the canopy layer of apple trees,and the distribution map of nitrogen content in the canopy layer of apple trees in Guanli Town was further obtained.Study through the integration of different scale remote sensing data,the information of the beries’ crown nitrogen content is accurate,fast,and large-scale,and effectively reflects the spatial distribution of the content of the coronine content of the orchard.Calculate the amount of fertilization of orchards,and provide data support for the precise fertilization and soil improvement of the orchard,to promote the sustainable development of the apple industry and the regional ecological environment,which is of great significance to the precise management of orchard.