Study On Estimating Winter Wheat Yield And Fertilizer-N Use Efficiency By Ground Hyperspectral Remote Sensing Based On Fertilization Experiment

Rise of hyperspectral remote sensing is one of the major technological breakthroughs in the area of Earth observation, and its appearance greatly broadened our understanding of matter, which make the original wide-band remote sensing that can not detect the substance for people to understand and understanding. Hyperspectral remote sensing, also called imaging spectral remote sensing, in the electromagnetic spectrum ultraviolet, visible, near-infrared and mid-infrared, obtain many continuous and narrow spectral image data, which contain ample space、radiation and spectral triple information, and could better describe the "red edge " feature, distinguish different biochemical components and changes of the leaves, content and changes. Therefore, hyperspectral remote sensing is widely used in classification, condition monitoring, yield estimation of crops and plant diseases and insect pests forecasting.This study mainly discusses the critical period of canopy reflectance spectral characteristics of winter wheat yield and its production. Research on winter wheat yield was achieved by the two methods of simulation. One is using the traditional regression analysis of spectral characteristic parameters (including vegetation index, differential spectra and derivative spectrum parameters) and the quantitative relationship between crop yields. The second method is that crop yield estimation model is set up by the data mining (BP) neural network and support vector machine (SVM) and multiple linear regression. In addition this paper also discusses the relationship between crop spectrum and nitrogen utilization efficiency, which is simulated by the means of data mining. The results are as follows:1. The relationship between vegetation index, differential spectra index and derivative spectra parameters and the yield of winter wheat is close, which could accurately simulate the crop yield, through the winter wheat yield and spectral parameters of correlation analysis. In winter wheat green stage, jointing stage and heading stage, vegetation index, differential index and the derivative parameters and yield had a higher correlation coefficient (all above 0.5). The spectral index mainly includes GNDVI, NDVI, RI1DB, mSR7o5, D750-D55o)/(D750+D550), EGFN, (D725-D7oo)/(D725+D700), FD-NDNI, FD725/525 and FD (525-570)/(525+570). The single-phase models of spectrum evaluates yield mainly draws up properly by parabolic equation and exponential equation is better, and the long-phase model is better using mSR705 (R2=0.8582).2. Respectively, the composite yield estimation model is complished by root normalized index of nitrogen in reviving stage, the derivative spectra normalized FD (525-570) /(525+570) in jointing stage and normalized difference vegetation index in heading stage (R2=0.8741), its accuracy is significantly higher than the single-variable and multitemporal yield estimation model.3. Using data mining technology to establish yield predication model, the results shows that degree of fitting of artificial neural network and support vector machine (SVM) are are 0.9041 and 0.9156 respectively, which are higher than that of multitemporal composite model. This suggests that data mining technology has a broad prospect in terms of crop yield and can well proposed cooperation content yield.4. The manure application plays a very essential role in increasing and stabilizing yield, which vary with the type and proportion of organic manure and chemical fertilizer. Studies have shown that pig manure composting and organic-inorganic composite yield is more ascendant than cow dung compost and organic-inorganic composite with the same conditions. There are significant differences in nitrogen use efficiency (NUE) and partial factor productivity (PFP) under different fertilization conditions. Compared with fertilizer, the combination of straw and chemical fertilizer NUE can be significantly improved, while the application of organic and inorganic fertilizer nitrogen utilization has declined. On the basis of the relationship between crop spectrum and NUE, we have simulated the NUE of winter wheat by BP neural network and support vector machine, and have achieved a better result (Coefficient of determination of 0.5243 and 0.8714, respectively).