Research And Application Of Wheat And Corn Nitrogen Status Hyperspectral Diagnosis Based On Leaf And Plant

The nitrogen (N) of crops is strongly related to its grain yields and grain protein content, which is important for nutritional quality. In order to ensure crop growth and productivity, farmers always supply more than less nitrogen. Excessive N in field not only results in environment pollution, but also reduces farmers’profit. Therefore, accurate detection of crop nitrogen nutrition and appropriate supplement nutrient could ensure crops’growth in key growth period. Studies showed that nitrogen nutrition index (NNI) could be used to detect crop nitrogen nutrition accurately. However, the NNI determined by the time-consuming and laborious conventional methods needed the two biochemical parameters nitrogen concentration and biomass. NNI was difficult to guide agricultural production. Therefore, remote sensing technology was the urgent need to estimate the nitrogen concentration and biomass accurately which achieved real-time estimation of NNI. This paper was devoted to the accurate estimation of nitrogen and biomass by used of hyperspectral data surrounding the leaves scales and canopy scales. Nitrogen concentration curve established to realize remote sensing estimation of NNI provided effective means for real-time monitoring of crop nitrogen status, variable rate fertilization and production forecast.For this reason, the paper analyzed the field experiment data and the hyperspectral image data in different years, different water and nitrogen supply and different varieties in Beijing xiaotangshan precision agriculture demonstration station and the experimental base of yard in Beijing academy of agriculture and forestry. These datas included some parameters such as field experiments and two hyperspectral images of unmanned aerial vehicle which was synchronous to surface measurement during flag leaf stage and flowering period in 2014/2015.Based on these data, the research was aim to create diagnosis models for NNI around leaves scales and canopy scale used hyperspectral data. At the same time, diagnosis models of nitrogen nutrition index were applied from the ground monitoring to uav hyperspectral image in the field. These provided effective technical support for real-time monitoring and accurate diagnosis of nitrogen nutrition on a large range, variable rate fertilization and production forecast. The detail investigation achievements were as follows:(1) From the spectral characteristics of original, derivative and continuum removal, spectrum index and radiative transfer model, the research progress and shortcomings of nitrogen nutrition diagnosis were introduced. Nitrogen concentration dilution model and the research progress and shortcomings of nitrogen nutrition index to diagnosis nitrogen status were introduced. On this basis, research idea of this article was put forward.(2) The paper introduced the generality of two research area that were Beijing xiaotangshan precision agriculture demonstration station and the experimental base of yard in Beijing academy of agriculture and forestry in detail and described test scheme design of the two study areas, the determination method of field test data, the acquisition process of unmanned aerial vehicle (uav) digital images and hyperspectral images.(3) The relationship between leaf nitrogen nutrition status and the spectral characteristics of original, derivative and continuum removal, the spectrum index which is sensitive to nitrogen built by EFAST method and PROSPECT model, common spectrum index related to nitrogen was analyzed. The study determined the order of spectral index sensitive to wheat and corn.(a) EFAST method and the model of PROSPECT were introduced in detail. PROSPECT model was used to stochastic simulate leaf spectral reflectance data. EFAST method was used to analyze sensitivity of the various parameters of PROSPECT model in 400-2500 nm wavelength range of leaf reflectance spectra. The results showed that the range of sensitive wavelength to chlorophyll is from 417 to 728 nm. Referring to normalized difference vegetation index and ratio vegetation index, the vegetation index sensitive to wheat nitrogen were constructed such as NDSI(564,728), NDSI(543,728), RSI(564,728) and RSI(543,728)-Vegetation indices sensitive to maize nitrogen were constructed such as NDSI(629,649), NDSI(495,669), RSI(629,649) and RSI(495i669)-The relationship between the constructed vegetation indices and leaf nitrogen content, leaf nitrogen accumulation were analyzed. Studies had shown that except filling stage of maize, the correlation between spectral characteristics and leaf nitrogen content was higher than that between spectral characteristics and leaf nitrogen accumulation in the rest growth period. So leaf nitrogen content could do better to monitor of wheat’s and maize’s nitrogen nutrition status real-timely.(b) Based on empirical statistical relations, regression models for spectrum index and leaf nitrogen content, leaf nitrogen accumulation were established. Decision coefficient (R2), root mean square error (RMSE) and relative error (RE) were used as evaluation index of the model. The results showed that leaf nitrogen content could be better to monitor nitrogen nutrition of wheat and maize. The first five spectrum indices sensitive to leaf nitrogen concentration of wheat were mND705, ND705, SR705, GMI-2, Rl-half; The first five spectrum indices sensitive to leaf nitrogen accumulation of wheat were R55o/R8oo, GMI-1, RSI(564,728), RS 1(543,728), RI-2dB; The first five spectrum indices sensitive to leaf nitrogen concentration of maize were VOGb, VOGc, NDRE, VOGa, CIred edge; The first five spectrum indices sensitive to leaf nitrogen accumulation of maize were NDRE, MTCI, RI-2dB, VOGa, VOGb.(4) The relationship between plant nitrogen nutrition status and the spectral characteristics of original, derivative and continuum removal, the spectrum index which is sensitive to nitrogen built by EFAST method and PROSPECT model, common spectrum index related to nitrogen was analyzed. The study determined the order of spectral index sensitive to wheat canopy.(a) Model of PROSPECT was introduced in detail. PROSPECT model was used to stochastic simulate canopy spectral reflectance data. EFAST method was used to analyze sensitivity of the various parameters of PROSPECT model in 400-2500 nm wavelength range of canopy reflectance spectra. The results showed that the range of sensitive wavelength to chlorophyll is from 515 to 745 nm. Referring to normalized difference vegetation index and ratio vegetation index, the vegetation index sensitive to wheat plant nitrogen content and plant nitrogen accumulation were constructed such as NDSI(546,698)、NDSI(667,685)、NDSI(539,745)、RSI(546,694)、RSI(667,684)、RSI(539,745); The relationship between the constructed vegetation indices and plant nitrogen content, plant nitrogen accumulation were analyzed. Studies had shown that plant nitrogen accumulation could do better to monitor of wheat’s nitrogen nutrition status real-timely.(b) Based on empirical statistical relations, regression models for spectrum index and plant nitrogen content, plant nitrogen accumulation were established. R2, RMSE and RE were used as evaluation index of the model. The results showed that the first five spectrum indices sensitive to wheat canopy nitrogen concentration were SRPI, NPCI, ND705, MCARI/MTVI2 and MTCI. The first five spectrum indices sensitive to wheat canopy nitrogen accumulation were SR705, RI-half, NPCI, VOGb and mSR705.(5) The relationship between leaf and plant biomass and the spectral characteristics of original, derivative and continuum removal, the spectrum index which is sensitive to biomass built by EFAST method and PROSPECT model, common spectrum index related to biomass was analyzed. The study determined the order of spectral index sensitive to wheat.(a) Model of PROSPECT(PROSAIL) were introduced in detail. PROSPECT (PTODSAIL) model was used to stochastic simulate canopy spectral reflectance data. EFAST method was used to analyze sensitivity of the various parameters of PROSPECT model in 400-2500nm wavelength range of canopy reflectance spectra. The results showed that the range of sensitive wavelength to chlorophyll is from 749 to 2410 nm. Referring to normalized difference vegetation index and ratio vegetation index, the vegetation index sensitive to wheat leaf biomass and plant biomass were constructed such as NDSI(2126,2347), NDSI(1652,1686), RSI(2126,2347) and RSI(1652,1686)·(b) Based on empirical statistical relations, regression models for spectrum index and leaf biomass and plant biomass were established R2, RMSE and RE were used as evaluation index of the model. The first five spectrum indices sensitive to leaf biomass of wheat were mSR705, RI-1dB, VOGa, GNDVI and NDCI; The first five spectrum indices sensitive to plant biomass of wheat were VOGa, mSR705, REP, NDVI705 and mNDVI705.(6) On the basis of the concept of critical nitrogen dilution curve, critical nitrogen concentration curves of wheat leaf and plant were established in xiaotangshan station. Their models were and respectively and compared with established key nitrogen concentration dilution curve.(a) In the study area, the leaf nitrogen dilution curve of winter wheat was Ncl= 4.42×WL-0.18, the plant nitrogen dilution curve of winter wheat was Ncp=5.81×W-0.541(b) On the basis of the concept of critical nitrogen dilution curve, Based on these indices and leaf (plant) critical nitrogen concentration curve, half experience and half mechanism leaf and plant model "remote sensing information-agronomic parameters-nitrogen nutrition index" was established by empirical statistical analysis method. The decision coefficient between predicted nitrogen nutrition index and measured nitrogen nutrition index was 0.77 and 0.83, respectively. The results were all significant (p<0 .01).(7) The results above were applied for uav images in the field. Results show that the R2 between predicted NNI and obtained from unmanned aerial vehicle (uav) hyperspectral images in flag leaf stage was 0.66. The R2 between predicted NNI and obtained from unmanned aerial vehicle (uav) hyperspectral images in flowering was 0.69. Both of the results were significant (p<0.01) with a better modeling and validation effect and no phenology effect. The paper provided a scientific basis for monitoring plant nitrogen status rapidly, accurately and real-timely, variable rate fertilization and yield forecast.