Estimating Canopy Leaf Nitrogen Of Soybean Using Hyperspectral Data At Different Nitrogen Levels And Cropping System

Hyperspectral remote sensing is one of the most important technical means in precision agriculture. It plays key function for quantitative analysis and application in agriculture. Soybean is an important raw material for oil and protein in China and plays an important role in the people’s diet structure. However, the planting area of soybean is decreasing, the soybean industry situation is grim and many have to rely on imports. The intercropped soybean can effectively increase the soybean planting area and total yield, improve the land utilization and output ratio. Due to soybean is a legume crop with nitrogen fixation, the application of nitrogen fertilizer and time is difficult to control accurately. Therefore, the nitrogen nutrition of soybean canopy are estimated by using the hyperspectral remote sensing technology in real-time, and the estimation model of soybean leaf nitrogen accumulation (LNA) are established in intercropping and monoculture conditions. It will provide scientific basis and theoretical techniques for application of hyperspectral remote sensing technology in nitrogen diagnosis of soybean.Monoculture and relay intercropping system and different nitrogen levels were been used in this research. The soybean LNA showed a single peak trend in the growth period in different nitrogen levels under intercropping and monoculture conditions, the maximum value appeared in the pod stage (9.08g/m2 for 2013 and 8.01g/m2 for 2014). The soybean LNAs of different nitrogen treatments at the earlier growth stage in monoculture condition were higher than those in intercropping. The change trends of soybean LNA were similar in both years. The soybean canopy spectrum showed the greatest difference in the green area and near red region than those in the other spectrum region. The soybean canopy spectrum of monoculture system was higher than those of intercropping system. The soybean spectrum showed first increase and then decrease in 700-1000nm, this change trend was similar with LNA in the growth period, but there was no significant difference in 400-700nm. In the first derivative spectrum of soybean canopy, red shift and blue shift phenomenon appeared at the position of red edge and the red edge amplitude firstly increased and then decreased within the growth period. The red edge amplitude showed little difference under different nitrogen levels under monoculture and intercropping system. The red edge position showed the same trend with soybean LNA.The relationship between spectrum feature and soybean LNA in different planting patterns and different nitrogen levels were analyzed in this research. The result showed that a positive correlation and negative correlation between original spectrum and LNA in 700-1000nm and 400-700nm respectively. The maximum correlation coefficient was about 0.8 in red edge region. The correlation between LNA and spectral parameters based on the spectrum position and area reached significant level except the yellow edge area. The correlation coefficient between LNA and spectral parameters based on the edge area is better than other spectral parameters. The correlation between LNA and some vegetation index appeared significant difference, especially for the terrestrial chlorophyll index (MTCI), the maximum correlation coefficient was 0.88. System analysis of the correlation between LNA and spectral index (ratio spectral index RSI, difference spectra index DSI, normalized spectral index NDSI) which were free combination of 400-1000nm wavelength based on original and derivative spectrum. The highest correlation coefficient appeared in DSI(771,755) (r=0.92), which was higher than other spectral indexes. The spectral sensitive regions of soybean leaf nitrogen accumulation are mainly located in 500-800nm and 700-1000nm of spectrum. The highly correlation coefficient area of derivative spectrum is more than that of original spectrum, The correlation between wavelet coefficients and LNA were affected by wavelet function and decomposition scale. The wavelet decomposition of red edge and near infrared region spectrum displayed better correlation with LNA. In addition, the method of original spectral information extraction using continuous wavelet analysis was better than the method of derivative spectra and vegetation index.The nitrogen sensitive spectrum and assessing model of soybean leaf nitrogen accumulation were built using linear, nonlinear model, stepwise regression and PLS method. In this research, the spectral index DSI (771,755), stepwise regression, and PLS were the high accuracy regression model. At the same time, the original spectrum data were analyzed using the scale of 1-256 and different wavelet function. The spectral variable of sym8 was been Screened in the scale of 210 and 735nm. The spectral variable of sym8 created model has highly accuracy. The checking RPD of model by using wavelet analysis was higher than any other methods. The RPD of model made in paper were all more than 2. That means these model can better establish the soybean canopy LNA. The model can better describe the correlation between soybean canopy leaf nitrogen status and canopy spectrum, the model could accurately predict the canopy nitrogen status in soybean and provide evidence for soybean nitrogen regulation.