Monitoring Nitrogen Nutrition With Hyperspectral Remote Sensing In Rice And Wheat

Real-time and non-destructive monitoring of crop nitrogen (N) status is of significant importance for precision N management in rice and wheat production. The hyperspectral remote sensing technique was recommended as an effective method for non-destructive and real-time monitoring of crop N status. The objectives of this study were to explore common indicative bands and optimum vegetation indices for monitoring leaf nitrogen status in rice and wheat, and establishing the monitoring models with strong explanation and high accuracy based on a series of field experiments in different years, cultivars, N rates and water regimes in two crops (rice and wheat). The results would help to provide technical support for designing spectral sensors in estimation of plant nitrogen status.Firstly, by taking account of canopy components and plant growth status in rice and wheat fields during different growth periods, and integrating methods of spectral analysis, crop physiological principles and statistics analysis, soil adjusted vegetation index and ratio vegetation index were constructed. In addition, the quantitative monitoring models for nitrogen were established with strong explanation and high accuracy in rice and wheat. Results indicated that the optimum spectral vegetation indices for leaf nitrogen concentration (LNC) were SAVI(R722, R815) during the growth period from jointing to booting and RVI(R722, Rsis) from heading to filling. Moreover, the optimum spectral vegetation index for estimating leaf nitrogen accumulation (LNA) was SAVI (R822, R738) during the growth period from jointing to booting, while it was RVI (R822, R738) from heading to filling.Further, a new form of three-band vegetation indices was constructed to reduce saturation in two-band vegetation indices, and the optimal common three-band vegetation index was selected to establish models for monitoring of canopy LNC and LNA in rice and wheat. The results showed that the linear models of nitrogen monitoring with three-band spectral indices were stable and accurate from jointing to filling in rice and wheat. The optimum three-band vegetation index of (R924-R703+2*R423)/(R924+R703-2*R423) for LNC was identified with coefficient of determination (R2) of0.870and0.857, and SE of0.052and0.148in rice and wheat, respectively. Testing of the models with independent dataset gave R2lager than0.86, RRMSE less than17%. The index of (R816-R732-R537)/(R816+R732+R537) was optimal for LNA estimation and represented good model performance of R2of0.803and0.862, and SE of1.244and0.942in rice and wheat, respectively. Performance testing of the models by the independent dataset indicated that the models predicted well for LNA with R2lager than0.82, RRMSE less than27%.In order to explore the influence of bandwidth change on monitoring models, the bandwidth change of central bands were further analyzed based on those optimal spectral vegetation indices in this study for LNC and LNA. Results indicated that the different band position had various responses on the bandwidth change. Furthermore, the optimal bandwidths of the best two bands spectral vegetation indices SAVI(R822, R738) and RVI(R822, R738) for LNC were24nm(722nm) and48nm(815nm); the optimum bandwidths of the best two bands spectral vegetation indices SAVI(R822, R738) and RVI(R822, R738) for LNA were33nm(822nm) and15nm(738nm); the optimum bandwidths of the best three bands spectral vegetation indices (R924-R703+2*R423)/(R924+R703-2*R423) for LNC were36nm(924nm),15nm(703nm) and21nm(423nm); the optimum bandwidths of the best three bands spectral vegetation indices (R816-R732-R537)/(R816+R732+R537) for LNA were20nm(816nm),8nm (732nm)and14nm (537nm).The change patterns of canopy original spectral reflectance and first derivative spectral reflectance of rice and wheat in the red edge region under different nitrogen levels were consistent with law. Furtehermore, compared with different methods for REP extraction techniques, this paper found that adjusted linear extrapolation had the best performance in nitrogen nutrition monitoring. Meanwhile, among these red edge shape parameters, red edge symmetry and double peak symmetry were good indicators for leaf nitrogen concentration and leaf nitrogen accumulation and had good application potentiality in nitrogen nutrition monitoring.Finally, combined with hyperspectral data processing and analysis techniques and mixed programming ideas, hyperspectral data processing and analysis system was developed according to the design methodology of modular and component based software architecture. This system can be used to explore the central bands, determine the optimal bandwidths, construct the sensitive spectral parameters and establish reliable monitoring models. Moreover, the system was analyzed by hyperspectral data within a case study on estimation of leaf nitrogen concentration. The results indicated that the system had the ability to explore the new spectral parameters with high efficiency, and was able to realize the estimation of leaf nitrogen concentration with good accuracy.