| Non-destructive and quick assessment of leaf nitrogen nutrition (LNN) is important for precision management of N fertilizer. The technology of crop growth monitoring based on hyperspectral reflectance provides an effective method to obtain crop N nutrition information. Based on three years field experiments with different varieties, nitrogen rates and planting densities in rice, with expanding the impact of soil background deliberately The characteristics of hyper-spectral reflectance under different experimental conditions and their correlations with nitrogen status and growth characters in rice were investigated. A comprehensive analysis was made on the quantitative relationships between canopy spectrum and N nutrition in rice. And discussed the effective way to reduce the influence of soil background and canopy structure, finally explored the optimum spectral parameters and quantitative models for estimating leaf N status. And the result would help to provide technical support for non-destructive monitoring and precision diagnosis of rice growth.First of all, based on the features of caopy spectrum under different nitrogen levels and planting densities, a comprenhensive analysis was made on the relationships between two-band published indices and new indices and rice leaf nitrogen nutrition (LNN) based on all possible combinations by original spectrum and the first-order derivative spectrum within350-2500nm. We screened band combinations which were sensitive to leaf nitrogen concentration (LNC) and insensitive to rice leaf area index (LAI), and in all these combinations (553,537) in SR2performed the best. Baed on the parameter0which was calucated by the soil line we got modified simple index SR2(553,537). And the combined index DI(D875, D645)+SR2(5537,537) which was filted by the sensitivity of different spectrum and the technology of vegetation index combination had better performance with rice LNC than all these published vegetation indices (VIs). These two indices could estimate LNC without affected by the soil background and canopy structure whose S-R2 with LNC and LAI of0.68,0.69and0.17,0.16. The result of independent test (2009, n=120) showed that P-R2and RRMSE of the regression model based on these two new VIs were both0.70and0.14. In rice LNA, SR2(770,752) still perfoemed well. NDI(D754, D700) performed best in the first derivative indices. And the S-R2of them were0.90and0.88, resprectively. The result of independent test (2011, n=120) showed that P-R2and RRMSE of them were0.80,0.23and0.78,0.27. In general, the published spectral index performed well in estimating rice LNC, but the influence of rice LAI was large. And all of them were not as well as two new indices. Most published N/chlorophyll sensitive spectral indices could be a good estimate of LNA.A comprehensive analysis was explored on the quantitative relationships between red edge parameters and parameters of continuum removal method and rice LNN. The result showed that, the performance of all the published parameters (DD, DPS and NDPS) were poor when the influence of soil and water background was large. And after the introduction of amendment bands we got three improved red edge parameters whose performance were well. The best parameters mNDPS(A680-700, A700-724) and mDPS(A680-720, A700-720) who predicted rice LNC with its S-R2and SE of0.73and0.22, respectively. Testing of the models with independent data gave P-R2and RRMSE of0.60and0.16. In rice LNA, various parameters of two waveform analysis method could well predict the dynamic change of rice LNA in which the original red edge parameter NDPS(A680-735, A735.755) and modified parameter mNDPS(A735-755, A680-735) had the best performance with their S-R2of0.87and the independent test P-R2of0.76and0.75. the area of right absorption peak RA (560-760) based on the continuum-removal could predict rice LNA better whose S-R2and SE of0.92and0.85, respectively, and independent test result showed that P-R2and RRMSE of0.77and0.21. In general, performance of the published parameters was not ideal. These three modified red edge parameters could eliminate the influence of various types of impact factors as well as monitoring rice LNN. In addition to the selected red edge parameters, we also cited several other red edge, yellow edge and blue edge parameters, and the result showed that they were all poor in estimating rice LNN.A systematic analysis was made of the variation characteristic of spectral reflectance from different heights (50cm,100cm and150cm) and different angles of the main plane1m above the canopy (vertically, forward30°,45°and60°and backward30°,45°and60°). And a comprehensive analysis was explored on the quantitative relationships between the reflectance above and three indices (SR2, NDVI and SR). The results showed that, appropriate view zenith angle of view height and angles could somehow alleviate the influence of soil and water background. The influence from the soil and water background and canopy structure was the smallest when the position of the senor was1m above the canopy and backward60°, and could estimate rice LNC the best. Among three selected indices, SR (701,520) gave the best performance, the S-R2and SE for estimating LNC of0.62and0.39. The result of independent test showed that P-R2and RRMSE of the regression model were0.61and0.15, and its performance was stable under different vegetation coverages. In rice LNA, the relationship between canopy spectral reflectance from different heights and angles and rice LNA were almost the same, and the performance of SR (955,655) and SR (950,660) were better.Canopy spectra obtained by different years was decomposed with the linear mixed model, first select the two endmembers of vegetation and soil, and further vegetation coverage was calculated as the area ratio, and finally decomposed according to the formula of linear mixed models. A comprehensive analysis was made on the spectral variation characteristics of rice canopy spectrum before and after decomposition using linear mixed model and the quantitative relationships between5kinds of vegetation indices of rice canopy and its leaf nitrogen nutrition (LNN). The results showed that, in rice LNC, the canopy spectral took less influence from the LAI at the red and NIR band after decomposition, and the relationships between5selected VIs (SR, SR2, NDVI, SAVI and PVI) and rice LNC was better without band-shift except PVI, and SR was the best. However, the results with independent test showed that the effect of linear mixed model was not stable, and the difference between S-R2and P-R2was large.In rice LNA, the spectral variation characteristics of rice canopy spectrum before and after decomposition was small. There was limit room for improvement for the S-R2was0.88before decomposition, and there was no significant improvement before and after decomposition. In general, the linear mixed model can eliminate part of the influence of soil and water background and canopy structure, but the effect need further validated. |
PDF Full Text Request