Estimation Of Water And Nitrogen Parameters Of Cotton At Canopy Scale Based On Hyperspectral Remote Sensing

Precision agriculture is the inevitable choice of China's agricultural modernization.Hyperspectral technology is characterized by high speed,high timeliness,low cost and large information coverage,meets a variety of requirements of agricultural information management,has been widely used in the field of agricultural monitoring.How to use remote sensing data to obtain crop physical and chemical parameters accurately and stably has become a hot research topic among scholars.Focused on the problems of traditional crop water and nitrogen monitoring methods,such as cumbersome operation,low timeliness,and difficulty in using them in large areas of farmland,in this study,transplanted cotton after wheat was the research object,field and pit-test experiements were carried out in Xinxiang city of Henan province for two consecutive years.It involving three factors:growth period,irrigation gradient and nitrogen application level,the study on estimating water and nitrogen parameters of cotton at canopy scale based on hyperspectral technology was carried out.The main achievements are as follows:?1?The canopy equivalent water thickness?CEWT?of cotton can accurately distinguish the degrees of water deficit.Leaf nitrogen concentration and canopy nitrogen content can accurately distinguish the nitrogen status of cotton.?2?The spectrum of cotton canopy is affected by different water and nitrogen conditions.Canopy reflectance in the visible band around 350-700nm decreases with the increase of leaf nitrogen concentration,and increases with the increase of CNC in the near infrared band around 780-950nm.In the short-wave infrared bands of 1100-1300nm and 1550-1780nm and 2100-2350nm,the canopy reflectance decreases with the increase of CEWT.Under sufficient water and fertilizer conditions,the canopy spectral reflectance of near-infrared band gradually increases from bud emergence stage to flowering and bolls stage,while the reflectance of visible band and shortwave infrared band after 1450nm show a decreasing trend,and the performance of cotton canopy spectral is abnormal at the beginning of flowering.?3?The simulation effect of PROSAIL on the canopy spectrum of cotton in different growth periods is different.The simulation accuracy of PROSAIL in the bud stage of cotton is high?RMSE=3.46%?,while the simulation accuracy of the canopy spectrum in the flowering and bolls stage of cotton is slightly lower?RMSE=4.57%?.Chlorophyll content(C_ab)has the greatest influence on the model output of visible spectrum band,and the equivalent water thickness?EWT?has the highest contribution to Shortwave Infrared?SWIR?.The selected water-sensitive band is taken as the input item,and regression modeling of CEWT was carried out by using support vector machine?R²>0.5?.?4?The continuous scaling decomposition of leaf and canopy reflectance was carried out by using continuous wavelet analysis,and the wavelet coefficients in the sensitive regions of cotton water and nitrogen parameters were screened out.The noise band in the common sensitive area of water and nitrogen should be removed.When selecting spectral features after avoiding the wavelet coefficients in these regions artificially,the spectral characteristics were modeled by different machine learning algorithms,it was found that the method of CEWT regression modeling by BPNN is used to show more accurate and stronger stability?R²>0.8?.Using the same method,the prediction of CNC is also more accurate?R²>0.8?.?5?By summarizing the construction forms and band selection rules of 31 types of classical vegetation index,hyperspectral vegetable indexes suitable for CEWT and CNC monitoring of cotton were constructed.The water spectral index was constructed by 3 forms,ratio,normalization and angle.The nitrogen spectral index was constructed by 2 forms,ratio and normalization.According to the combination interval of sensitive bands,it was found that the linear model established by the ratio spectral index has better monitoring accuracy CEWT?R²=0.73?.The linear model established by normalized difference spectral index has the highest prediction accuracy for CNC?R²=0.68?,and it is not easily affected by the change of canopy water content;The band most suitable for constructing CEWT characteristic spectral parameters is located near 1100nm,while the band most suitable for constructing CNC characteristic spectral parameters is located near 500nm and 580nm.