Monitoring Nitrogen Related Parameters In Rice With Hyperspectral Imaging Data

Real-time,non-destructive and accurate assessment of crop nitrogen status is critical for the precision management of nitrogen fertilizer application,the improvement of crop yield and quality and reduction of environmental pollution.Remote sensing technology has been widely used for non-destructive quantitative monitoring nitrogen status of crops,which offers important technical support for implementation of precision farming.The traditional field non-imaging spectral data is susceptible to the soil background.In contrast,the advantage of high spatial and spectral resolution for near-ground imaging spectroscopy allows us to extract pure information of plants and can further improve the accuracy and stability in monitoring nitrogen status of crops.However,previous studies used imaging data with a single extremely-high spatial resolution(millimeter-level).Moreover,most of these studies focused only on sunlit-leaf pixels.The spatial and shadow effects on the estimation of nitrogen nutrition parameters have been neglected.In this study,rice leaf and canopy images were collected from a series of paddy field experiments under different treatments using a near-ground hyperspectral imaging system.Based on the use of multiple analysis methods such as remote sensing image interpretation,feature extration from hyperspectral imagery and mathematical statistics,we built a classification decision tree for discriminating sunlit-and shaded-leaf or panicle pixels at first.Then,the response mechanism of reflectance spectra for pure vegetation pixels extracted from hyperspectral images to the nitrogen nutrient was systematically analyzed.Finally,optimal spatial resolutions and its corresponding models for monitoring nitrogen status were determined by assessing the impact of spatial effects and shadow effects on the estimation of nitrogen nutrition parameters using images with different spatial resolutions.This would help to establish the key technology for real-time monitoring rice nitrogen status with near-ground hyperspectral systems and field crop phenotyping.The performances of performing vegetation indices(VIs)and multiple regression methods for estimating pigment content of rice leaves were systematically compared using hyperspectral imaging data of leaves.The results showed that red-edge and green bands were the most important bands for estimating pigment content.The predictive performance of area-based pigment content was generally better than that of mass-based pigment content regardless of estimation approaches.All estimation approaches generally exhibited better predictions of chlorophyll content than those of carotenoid content.mNDbiue(445,547,757)and mNDblue(445,557,757)are selected as optimal VIs.Partial Least Squares Regression(PLSR)models built based on optimal bands selected from Successive Projection Algorithm(SPA)exhibited the best prediction of pigment content with R2 of 0.79 and RE of 8.78%.By using these PLSR models,pigment distributions within rice leaves can be effectivelly digitized.These results can offer not only a technological support for quantitative monitoring pigment content in rice leaves but also a technological reference for crop phenotyping.The spectral properties of rice leaves and panicles in sunlit and shaded portions of canopies and the effect of shadows on the relationships between spectral indices of leaves and foliar chlorophyll content were examined and evaluated using a near-ground imaging spectroscopy system with high spatial and spectral resolutions.The results demonstrated that the shaded components exhibited lower reflectance amplitude but stronger absorption features than their sunlit counterparts.Specifically,the reflectance spectra of panicles had unique double-peak absorption features in the blue region.Among the examined VIs,significant differences were found in the photochemical reflectance index(PRI)between leaves and panicles and further differences in the transformed chlorophyll absorption reflectance index(TCARI)between sunlit and shaded components.After an image-level separation of canopy components with these two indices,statistical analyses revealed much higher correlations between canopy chlorophyll content and both PRI and TCARI of shaded leaves than for those of sunlit leaves.The red edge chlorophyll index(CIRed-edge)exhibited the strongest correlations with canopy chlorophyll content among all vegetation indices examined regardless of shadows on leaves.To assess the impact of spatial resolution on the estimation of leaf nitrogen concentration(LNC),plant nitrogen concentration(PNC)and leaf chlorophyll concentration(LCC),we collected ground-based hyperspectral images throughout the entire growing season over two consecutive years and generated ten sets of images with spatial resolutions ranging from 1.3 mm to 450 mm.These images were used to determine the sensitivity of LNC,PNC and LCC prediction to spatial resolution with three groups of VIs and two multivariate methods Gaussian Process regression(GPR)and PLSR.The reflectance spectra of sunlit-,shaded-and all-leaf/plant pixels separated from background pixels at each spatial resolution were used to predict LNC,PNC and LCC with VIs,GPR and PLSR,respectively.The results demonstrated all-leaf/plant pixels generally exhibited more stable performance than sunlit-and shaded-leaf/plant pixels regardless of estimation approaches.The predictions of LNC,PNC,LCC required stage-specific LNC?VI,PNC?VI and LCC-VI models for each stage before booting but could be performed with a single model for all the stages after booting.Specifically,most VIs achieved stable predictive performances of LNC,PNC and LCC from all the resolutions finer than 14 mm(corresponding platform height is 13 m)for the early tillering stage but from all the resolutions finer than 56 mm(corresponding platform height is 52 m)for the other stages,except for all the resolutionsfiner than a corser resolution of 113 mm(corresponding platform height is 104 m)when estimating LCC at the jointing stage and stages after booting.In contrast,the global models for the prediction of LNC,PNC and LCC across the entire growing season were successfully established with the approaches of GPR or PLSR.In particular,GPR generally exhibited the best prediction of LNC,PNC and LCC with the optimal spatial resolution being found at 28 mm(corresponding platform height is 26 m).The impact of spatial resolution on the estimation of leaf nitrogen accumulation(LNA)and plant nitrogen accumulation(PNA)was systematically assessed using hyperspectral images with different spatial resolutions.The results demonstrated all-and shaded-leaf/plant pixels generally exhibited more stable performances than sunlit-leaf/plant pixels regardless of estimation approaches.The predictive performance of PNA?VI models was worse than that of LNA?VI models,especailly for stages after booting.CIRed-edge derived from shaded-leaf/plant pixels exhibited the best prediction of LNA with 113 mm resolution images(corresponding platform height is 104 m)and the best prediction of PNA with 225 mm resolution images(corresponding height is 207 m),respectively.For the jointing stage that is critical for the fertilizer application,most VIs exhibited the best predictive performance when using 113 mm resolution images(corresponding platform height is 104 m).The approaches of GPR or PLSR exhibited better predictions of LNA and PNA as compared to VIs,especially improved the predictive performance of PNA for stages after booting.In particular,GPR generally exhibited the best prediction of LNA and PNA with the optimal spatial resolution both being found at 225 mm(corresponding platform height is 207 m).