Monitoring Nitrogen Status At Canopy And Leaf Scales With Hyperspectral Sensing In Wheat

Quantitative monitoring of plant nitrogen status is an important research field in vegetation remote sensing. Fast and non-destructive and accurate estimation of crop nitrogen status is one of key techniques in development of precision agriculture. The primary objective of this study was to explore the optimum wavebands, spectral indices and quantitative models for estimating leaf nitrogen content (LNC) and leaf nitrogen accumulation (LNA) through systematic extraction of hyperspectral information at canopy and leaf levels, on the basis of multiple field experiments under varied N rates and cultivars in wheat. The anticipated results would provide new waveband choice for manufacturing portable N monitoring instrument and utilizing space-borne remote sensing data, and thus assist in real-time estimation and precise diagnosis of plant nitrogen status in wheat.Firstly, the change patterns of canopy leaf nitrogen status over growth progress under varied nitrogen levels were established, and then a reduced precise sampling method was adopted for comprehensive analysis of the quantitative relationships of LNC and LNA to NDSI (normalized difference spectrum index, NDSI) and RSI(ratio spectrum index, RSI) composed of any two wavebands with original reflectance and its derivative within the spectral range of 350-2500 nm, and to SASI (soil adjusted spectrum index, SASI) and selected best spectrum index with different bandwidths. From the derived core bands and sensitive parameters, the monitoring equations were constructed for LNC and LNA in wheat. The results showed that the sensitive wavebands of nitrogen status were mostly located in the visible and near-infrared regions. The LNC monitoring models developed from NDSI(R1350, R700), NDSI(FD700, FD690) and RSI(FD691, FD711) gave high estimation accuracy. And the spectral indices NDSI(R86o, R720), NDSI(FD736, FD526) and RSI(R99o, R720) could be reliably used for estimating LNA. Furthermore, analysis of L parameter in SAVI revealed that the models constructed on SASI(R1350, R700) and SASI(R860, R720) had best estimation with L parameter as 0.09 and 0.3, respectively. Comparatively, the LNA models performed better than the LNC models for wheat canopy.Further analysis was conducted on the responses of the optimum spectral indices and monitoring models to the changes of spectral resolution based on the sensitive bands. The results revealed that the impact of spectral resolution differed with spectral index. The changes for NDSI(R1350, R700) and NDSI(R860, R720) were relatively stable with the resolution of four key bands less than 20 nm,60 nm,96 nm and 26 nm, respectively. With varied wavebands, the values for RSI(R697, R1155) gradually decreased in the direction of 697 nm, but enhanced in 1155nm. The RSI(R99o, R720) exhibited fast changes within 60nm bandwidth in the direction of 990 nm, but slow changes in 720nm. In addition, it was found that the spectral indices based on varied spectral resolution of key bands generated differential prediction accuracy and stability for wheat nitrogen status. The monitoring models based on NDSI(R1350, R700) and NDSI(Rg6o, R720) displayed stable performance with the resolution of four key bands less than 74 nm,46 nm,96 nm and 26 nm, respectively. And the models from RSI(R697, R1155) and RSI(R99o, R720) were excellent with the resolution of four bands less than 4 nm,6 nm,14 nm and 14 nm, respectively.On the basis of analyzing the quantitative relationships and statistical characters between different types of red edge parameters (including red edge position, red edge derivative, red edge area and other red derivative paremeter) and canopy leaf nitrogen status, the monitoring models were developed for canopy leaf nitrogen nutrient in wheat by comparing accuracy and reliability of nitrogen estimation. The results exhibited a dual-peak or multi-peak feature with the first derivative spectra in the red edge region of 690-730nm. With increasing nitrogen levels, the red edge position (REP) moved to the longer wavelength and the red edge derivative increased, thus enlarging the red edge area. With progress of the growth stages, the REP moved first to the longer wavelength and then to shorter wavelength, and three red edge parameters all gradually increased and then decreased after booting. These change patterns were consistent among different cultivars. Of several red edge parameters, the monitoring models developed from the REP-LEM (linear extrapolation method) and the minimum RED could stably indicate LNC, and the models on the REP-LEM and the difference between left and right REA (red edge area) from IGAUS could reliably estimate LNA, with better performance from LNA models than from LNC models in wheat canopy.By re-sampling hyperspectral data from ASD measurements, the relationships were systematically quantified of canopy leaf nitrogen status to the simulated spectral parameters including the single wavelength, ratio spectral index and normalized difference spectral index, and the capacity and stability of estimating canopy leaf nitrogen status were compared based on simulated satellite channels. The results indicated that the spectral parameters based on NDVI(MSS7, MSS5), NDVI(RBV3, RBV2), TM4, CH2, MODIS1 and MODIS2 could be reliably used for estimating LNC, and those on NDVI(PB4, PB2), NDVI(CH2, CH1), NDVI(MSS7, MSS5), RVI(MSS7, MSS5), MODIS1 and MODIS2 could be used for predicting LNA, with better performance from LNA models than from LNC models in winter wheat. In comparison, NDVI(MSS7, MSS5) and NDVI(PB4, PB2) were the best spectral indices for monitoring LNC and LNA in wheat canopy, respectively. Furthermore, MSS7, MSS5, LANDSAT and IRS-P6 should be of wider application prospect in monitoring of crop nitrogen status.By elucidating the change patterns of the single leaf hyperspectrum and canopy leaf nitrogen content at different growth stages, analysis was made on the accuracy and stability of monitoring canopy nitrogen status with the single leaf hyperspectral parameters including the new spectral indices, red parameters, and reported nitrogen indices. The results showed that the spectral indices of the 2nd and 3rd leaves had the stronger capacity to estimate the canopy leaf nitrogen content (LNC), and could be considered as the indicator of canopy LNC in wheat. The parameters NDSI(R429, R477) and RSI(R429, R498) of the 1st leaf, NDSI(R610, R480) and RSI(R610, R480) of the 2nd leaf, NDSI(R1821, R571) and RSI(R1821, R571) of the 3rd leaf, and NDSI(R654, R663) and RSI(R663, R654) of the 4th leaf could be used for reliably predicting canopy LNC. Of several red edge parameters, the minimum RED, REP-IGAUS, symmetry of REP, and position of the red vale were the best indices for evaluating canopy LNC based on the 1st,2nd,3rd and 4th leaf, respectively. With the previously reported spectral indices, PRIa, RSI(R560, R450), FD723 and FD612 were found to be the proper parameters for canopy LNC on the 1st,2nd,3rd and 4th leaf, respectively. In addition, with the spectral indices of four combined leaves, the monitoring equations based on DSI[NDSI(R1821, R571)3, NDSI(R610, R480)2], NDSI[RSI(R1821, R571)3, RSI(R610,R480)2] had better performance than all the above indices. Further analysis was made on the responses of the monitoring models to the spectral resolution of sensitive bands in the best spectral index. It was found that there was a suitable range for the spectral resolution of key bands with the individual top leaves, in which estimation accuracy of the canopy LNC did not show obvious difference.The quantitative relationships were analyzed between the leaf nitrogen contents and the new spectral indices, red edge parameters, and reported nitrogen indices at leaf level. The results revealed that the nitrogen contents in individual top leaves obviously exhibited spatial distribution pattern, markedly influencing corresponding spectral reflectance. With increasing nitrogen rates, the nitrogen contents in the top four leaves gradually increased, while the spectral reflectance decreased in the visible region but enhanced in the near-mid-infrared regions, with much higher reflectance values than in the visible region. With the growth progress, the leaf nitrogen content first increased and then decreased, while the spectral reflectance showed opposite pattern, essentially consistent in top four leaves. The spectral parameters NDSI(R510, R430), NDSI(R62o, R480), NDSI(R622, R426) and RSI(R421, R655) could be reliably used for quantifying the nitrogen contents of the 1st,2nd, 3rd and 4th leaf in the top, respectively, while the NDSI(R613, R426) could generally estimate the nitrogen contents in the top four leaves. The red edge region was greatly affected by the nitrogen rates and cultivar types, and could markedly differentiate the nitrogen levels in the spectral region of 703-742nm. The minimum RED and red edge symmetry could estimate the nitrogen contents of four combined leaves. In addition, ND705 and mND705 were more general spectral parameters for monitoring of nitrogen content at leaf level in wheat. Further analysis was conducted on the effects of the spectral resolution of key bands in the sensitive parameters on prediction accuracy for single leaf nitrogen content. It was found that there was a proper range for spectral resolution, in which estimation ability of the LNC model was relatively stable.