Monitoring The Vertical Distribution Of Nitrogen Status At Leaf And Canopy Scales With Remote Sensing Data In Maize

Maize is an important food crop,cash crop and industrial raw material as well as one of the most widely cultivated crops in the world which has the characteristics of high yield growth potential and strong regional adaptability.Nitrogen is one of the important parameters of plant photosynthetic capacity and nutrient deficiency.Therefore,precision agricultural technology can provide real-time and rapid diagnosis of plant nutrient deficits,which helps accurate farming management and crop yield prediction,reduces environmental pollution,and improves the efficiency of resource utilizing.Remote sensing technology is the core technology of precision agriculture,which can quickly and non-destructively monitor crop target parameters and provide technical support for implementing accurate field management.The summer maize in Guanzhong Plain and spring maize in Weibei plateau of Shannxin province are the researching objects.In this study,the spectral reflectance characteristics of maize leaves and canopy under different control conditions were analyzed based on perspectives of temporal variability?different growth stages?and spatial variability?different vertical levels?,and spectral sensitive regions were screened out at different growth stages.Then,combining spectral pretreatment technology and characteristic wavelength screening method,maize nitrogen status estimation models were constructed,and prediction models were validated based on independent data;Finally,the Sentinel-2 satellite multi-spectral image data was used to invert the nitrogen status of maize at regional scales,which provided an important theoretical basis for the diagnosis of crop nitrogen status and grasp crop growth status.There are four parts in this paper:1)Estimating of leaf nitrogen concentration considering the unsynchronized maize growth stages of summer maize and spring maize with hyperspectral technique;2)Estimation of vertical nitrogen distribution in maize with leaves/canopy hyperspectral reflectance data;3)Estimation of maize nitrogen nutrition index with hyperspectral reflectance data;4)Retrieving maize nitrogen status based on Sentinel-2 satellite multi-spectral image data.The main conclusions are:?1?The objectives of this study were to determine the optimum spectral analysis method for estimating the leaf nitrogen concentration of unsynchronized growth stages in Shaanxi Guanzhong summer maize and Weibei plateau spring maize at different field conditions.The results showed that:among all of the types of spectral indices,the red-edge chlorophyll index(CI_{red edge}),the new 2-band VIs and their band combinations varied across different growth stages and were not affected by unsynchronized growth stages.The correlation between leaf nitrogen concentration and spectral indices at V9 stage was significantly higher than that at R3 stage.The performance of these vegetation indices were basically the same,with no significant differences across two sites.The best spectral index of the green or red and red-edge or near-infrared band combination based on FDR spectra had a good diagnostic effect for LNC of maize across the four growth stages,in which NDSI?D528,D756?for the V9 stage,NDSI?D523,D758?for the VT stage,NDSI?D527,D754?for the R1 stage and RSI?D614,D1112?for the R3 stage.Using the raw and FDR hyperspectral data as independent variables,the leaf nitrogen concentration estimation model were constructed by using PLS?partial least squares regression?.Compared to PLS regression based on raw full-range hyperspectral data,the PLS regression for estimating LNC across four growth stages based on FDR full-range hyperspectral data showed a higher accuracy,with an average coefficient of determination(r²_val)of 0.87 and average root mean square error(RMSE_val)of 0.18%.The selected FDR wavelength regions in PLS regression were located mainly in the visible,red-edge,and NIR regions.The average r²_val for the PLS regression based on selected FDR wavelengths increased by 2.40%and the average RMSE_val decreased by 14.8%,compared with the best performing VIs during the four growth stages.These results suggest that the best two-band VIs and the PLS regression based on selected FDR wavelengths provide a useful explorative tool for estimating LNC of maize,independent of years,ecological areas,and unsynchronized growth stages.?2?The leaf nitrogen concentration at different leaf layers were estimated based on hyperspectral techniques,especially in middle and lower leaf layers,to improve the real-time,effective and accuracy diagnosis of maize nutrient status.According to the relative height within canopy,the leaves were divided into three layers by means of artificial leaf extraction in this study.The vertical distribution of leaf nitrogen concentration of summer maize and spring maize under different growth stages was analyzed.The results showed that the combination of effective wavelengths for leaf nitrogen concentration of different leaf layers were different.Moreover,the normalized difference spectral index?NDSI?and ratio spectral index?RSI?were constructed based on raw and FDR hyperspectral data,and the combination of sensitive wavelengths are mainly green and red edge/NIR regions.The NDSI?D528,D756?for upper layer?r²=0.80?,RSI?D545,D759?for middle layer?r²=0.78?and RSI?550,720??r²=0.75?,NDSI?D700,D1150??r²=0.76?for lower layer showed superior predictive capacity for estimating leaf nitrogen concentration.Comparatively,the leaf nitrogen content prediction model for the upper and middle layer performed better than the leaf nitrogen content prediction model for lower layer.In addition,the sensitive bands were screened based on VIP values in PLS method.The effective wavelengths of nitrogen in different leaf layers were mainly distributed in green,red,red edge and NIR regions,among which the red-edge region was the most sensitive region.The prediction accuracy of FDR-PLS and FDR-SVM models based on effective wavelengths were significantly higher than that of full spectrum PLS and SVM models and vegetation index prediction models.The results of this study provide theoretical support for remote sensing estimation of leaf nitrogen concentration in different leaf layers,and provide technical support for maize growth monitoring.?3?The nitrogen status in the whole canopy,middle+lower and lower layers of maize can be estimated at the maize canopy scale in this study.The vertical N distribution within spring and summer sown maize canopies followed the bell-shaped distribution curve with the highest values at the middle layer regardless of growth stage.Due to the vertical heterogeneity of nitrogen distribution in maize canopy,it is important to monitor the vertical nitrogen distribution in maize canopy with hyperspectral remote sensing.Among all of published indices investigated,mND₇₀₅ for whole canopy,G-M for middle+lower layer and MTCI for lower layer were highly correlated with leaf N density.Meanwhile,our newly developed OREA index was constructed according to red-edge absorption area characteristics and was simple formula:OREA?15(3R₇₆₀-R₅₅₀)-20(R₆₈₀+2R₇₂₀);The index significantly provided more accurate prediction of leaf N density for whole canopy,middle+lower and lower layers across growth stage,maize types and sites compared with the other spectral indices.Meanwhile,the predictive model showed high robustness?the lowest AD value?under different cultivars,growth stages and cropping systems,without extensive calibration for a number of variables.In conclusion,we have developed a novel OREA index that combines the advantages of the red-edge absorption area,and it flexibly uses red-edge parameters to estimate leaf N density at different vertical N distribution.?4?Nitrogen Nutrition Index?NNI?is an important parameter for diagnosing crop nitrogen nutrient deficiency and recommending fertilization,while hyperspectral techniques provide the possibility for quickly and non-destructively monitoring crop NNI.The summer maize of Guanzhong Plain and spring maize of Weibei plateau were the researching objects.The maize canopy hyperspectral data and corresponding nitrogen nutrition index?NNI?at different N levels were obtained from summer maize in Guanzhong and spring maize in Weibei Plateau,respectively.NNI prediction models were constructed based on spectral index,PLS and SVM methods,and compared the comprehensive performance of these methods.A comprehensive critical nitrogen concentration dilution model for summer and spring maize(Nc=3.63DM^-0.403)was constructed,and an indirect NNI monitoring model was constructed based on the best vegetation index for predicting plant nitrogen concentration and aboveground dry matter accumulation.Using direct method,the NDSI and RSI based on the raw and FDR spectral data have a significant correlation with NNI,among which RSI?825,550?predicts the best performance?r²=0.788?,with r²_val of 0.774,and RMSE_val of0.139,respectively.The effective wavelengths were selected based on the VIP?variable importance in projection,VIP?scores,which were mainly concentrated in the green,red and near-infrared regions.The FDR-PLS and FDR-SVM models based on effective wavelengths had high prediction effect,with r²_cal of 0.852 and 0.870,respectively,RMSE_cal of 0.124 and0.126,respectively,r²_val of 0.823 and 0.836,respectively,and RMSE_val of 0.130 and 0.129,respectively,which have higher prediction accuracy and lower RMSE_val than optimal RSI?825,550?.Therefore,the FDR-PLS and FDR-SVM regression models constructed based on the effective wavelengths can quickly estimate NNI in maize,and greatly reduce the spectral variables and improve the model accuracy.?5?According to the characteristics of Sentinel-2 satellite band parameters,eight representative spectral indices were constructed and correlated with nitrogen status at VT and R3 stages of maize,and the best prediction models for monitoring nitrogen status at VT and R3 stages in maize were determined.Furthermore,the corresponding thematic maps was produced to provide a basis for diagnosis crop nutrient status at regional scales.The results showed that the spectral parameters based on CI_{red edge?705,842?}could be reliably used for predicting leaf nitrogen concentration?r²=0.728?and leaf nitrogen density?r²=0.708?at VT stage,with r² of 0.70,0.71,with RMSE of 0.48 and 0.46,respectively.In Comparison,the CI_{red edge?740,842?}and CI_{red edge?705,842?}were the optimal vegetation indices for predicting leaf nitrogen concentration and leaf nitrogen density at R3 stage,respectively,with r²_val of 0.72,0.67,with RMSE_val of 0.16 and 0.13,respectively.Finally,thematic maps were produced to monitor maize leaf nitrogen concentration at VT stage and R3 stage based on the CI_{red edge}ed edge prediction model,which provided basis for diagnosis crop nutrients status and precise field management at regional scales.