Multi-scale Hyperspectral Estimation Of Chlorophyll And Nitrogen Content In Crop

Food security is an important guarantee for promoting harmonious social and economic development and building a well-off society in an all-round way.Modern food crop production is developing in the direction of precise planting,efficient management,intelligent decision-making and quantitative implementation,forming an integrated precision crop cultivation technology with the features of real-time monitoring,intelligent diagnosis,and quantitative regulation.Nutrient parameters such as nitrogen and chlorophyll are closely related to the photosynthetic capacity of crop leaves,and have an important regulating effect on crop grain yield and dry matter production.The rapid development of modern remote sensing technology provides an effective way for real-time monitoring of crop nutritional status.Although chlorophyll and nitrogen monitoring based on crop reflectance spectra have made a series of research results,there are still some problems to be solved in practical applications,such as the effects of specular reflection at the leaf scale,leaf area index(LAI)and soil background at the canopy scale,and the atmospheric correction uncertainty at the satellite scale on leaf chlorophyll content(LCC)estimation as well as the unclear mechanism of mass-based leaf nitrogen concentration(LNCM)estimation in the visible and near infrared spectral region.In view of this,this thesis first conducts research on hyperspectral estimation of LCC at three scales(leaf scale,near-ground canopy scale and satellite scale),and finally transitions to LNCM estimation.The main research contents and results of this thesis are as follows:(1)To analyze the internal mechanism of the difference between leaf DHRF and BRF spectra,and mitigate the influence of the specular reflection on the estimation of LCC by the empirical model method and the physical model method.Based on the bidirectional reflection distribution function,the diffuse reflection and specular reflection components from the directional hemispheric reflection factor(DHRF)spectrum measured by the integrating sphere and the bidirectional reflection factor(BRF)spectrum measured by the leaf clip are analyzed.Mechanism connection between DHRF and BRF spectra was established by introducing the specular reflection factor.The non-negligible specular reflection in the BRF spectra is the main reason for the difference between the DHRF and BRF spectra,limiting the application of the empirical model and physical model methods based on the DHRF spectra to the BRF spectra.We evaluated the sensitivity of four types of spectral features with different mathematical combinations or calculation forms(ratio index,adjusted ratio index,double difference index and red edge position)to the specular reflection,and established a unified empirical model for LCC estimation across DHRF and BRF spectra based on a large amount of measured data.As a result,the double difference index(such as Macc01)and the red edge position(such as REPPF)are not sensitive to the specular reflection and hence can be used to establish the unified models for LCC estimation across DHRF and BRF spectra.In addition,PROCWT,a new method of physical model inversion coupled with PROSPECT and continuous wavelet transform(CWT),can also significantly eliminate the influence of specular reflection in the BRF spectra on the inversion of LCC compared to the PROSPECT standard inversion model,reducing model inversion errors.After eliminating the effect of specular reflection,both the empirical model method and the physical model method show good consistency in LCC estimation when applied to imaging hyperspectral data.(2)To propose LAI-insensitive chlorophyll index(LICI),and establish semi-empirical models of LCC-LICI based on simulated reflectance at the top of canopy and at the top of atmosphere(TOA).Based on the canopy radiation transfer model,canopy reflectance of a large number of canopy structural scenarios were simulated to analyze the sensitivity of the existing vegetation indices to LCC and LAI,and then propose LAI-Insensitive Chlorophyll Index(LICI=R735/R720-(R573-R680)/(R573+R680)).The first part of LICI(R735/R720)is positively correlated with both LAI and LCC,while the second part(R573-R680)/(R573+R680)is positively correlated with LAI and negatively correlated with LCC.The subtraction of the two parts weakened the correlation between LICI and LAI correlation,and enhanced the correlation between LICI and LCC.The LCC-LICI semi-empirical model constructed based on the RowSAIL-PD performed well in the independent ground validation data.In addition,in order to avoid the complexity and uncertainty of atmospheric correction of satellite images,this thesis constructs the LCC～LICITOA semi-empirical model based on the TOA reflectance simulated with coupled SAIL and MODTRAN.This model was successfully used to estimate global LCC based on the TROPOMI TOA reflectance.These results have important application value for high-throughput intelligent monitoring of LCC in crops and global mapping of LCC.(3)To propose a strategy of optimizing the canopy spectra observation time in the early growth stage of crop,which can effectively reduce the influence of sunlit soil background on the LCC estimation.To mitigate the effect of soil background on canopy spectrum during the early growth stage of row-planted crops,a strategy of optimizing the canopy spectra observation time is proposed,which is different from the traditional spectral observation during noon time.The SAIL and RowSAIL simulations were used to clarify the contribution of soil background and hotspot effects to the diurnal variation pattern of uniform canopies and open row-crop canopies.For open row crops,the change in the proportion of sunlit soil is the main factor leading to the daily variation in canopy reflectance.By simulating the latitude variation,seasonal variation,and diurnal variation of sunlit soil fraction,the sunlit soil fraction was highest during the noon period(10:00 h-14:00 h)for north-south orientation crop,while for off-noon period the sunlit soil fraction gradually decreased until it reached to zero.Therefore,for open crops in the north-south orientation,off-noon spectral observations can reduce the effect of the soil background on the canopy spectra.The estimation error(RMSE)of LCC was 5.01 μg/cm2 by applying the LCC～LICI semi-empirical model to the off-noon observed spectra,which was significantly lower than that of using traditional spectral index(such as MTCI)and noon observed spectra(RMSE is 11.50 μg/cm2).(4)To clarify the estimation mechanism of LNCM in the visible and near infrared spectral region,and to propose an indirect estimation model of LNCM based on nitrogen distribution theory.Based on the hypothesis that leaf total nitrogen consisted of two parts:photo synthetic nitrogen(related to chlorophyll)and non-photosynthetic nitrogen(related to dry matter),the estimation mechanism of LNCM in the visible and near infrared spectral region was clarified,and the LNCM estimation model based on nitrogen distribution theory was proposed.Using the comprehensive dataset including LNCM,LCC,and LMA of rice and wheat leaves for many years,the leaf-scale LNCM model was parameterized through a linear model fitting(R2 is 0.82).Its performance in canopy-scale LNCM estimation was evaluated using measurements in wheat and rice fields and was compared with the traditional method based on the chlorophyll related index.LMA in rice was significantly higher than that of wheat.If the role of LMA in the estimation of LNCM was not considered,the traditional LNCM monitoring model based on the chlorophyll related index(such as CI800,710)did not have a unified model between rice and wheat.The relationship between LNCM and CI800,710 was only 0.27 for the whole growth period of wheat and rice.On the contrary,the our LNCM estimation model based on LCC and LMA can be applied to the whole growth period of wheat and rice,with R2 of 0.63.The above results provide theoretical and technical support for real-time and high-precision monitoring of chlorophyll and nitrogen nutrition status of rice and wheat crops by hyperspectral remote sensing,and provide an important reference basis for the formulation and implementation of crop cultivation plans.