Hyperspectral Estimation For Soil Organic Matter Content Of Winter Wheat Field

Winter wheat is one of the major crops in China and soil organic matter(SOM)plays an important role in representing soil quality and maintaining the sustainable production.The traditional determination method for SOM content is time-consuming,labor-intensive and not real-time.However,considering the advantages of rapidity,nondestruction and real time,hyperspectral technology can accurately achieve the SOM estimation.In this study,70 soil samples were collected from the winter wheat fields in Wenxi county of China.On the basis of soil spectra and SOM content data,we analyzed the response mechanism between SOM and soil spectra,and extracted the SOM sensitive wavebands.In addition,the effects of soil partical size,spectral preprocessing and modeling methods on SOM spectral monitoring were also explored.Ultimately,we proved that the SOM models in this study can accurately estimate the SOM content.The main conclusions as follows:(1)Both soil partical sizes and SOM content levels presented obvious negative correlation with soil spectral reflectance,but had no evident effects on the shape of soil spectral curves and spectral characteristic position.Especially,the soil partical sizes less than 0.154 mm produced the biggest effect on soil spectral reflectance.(2)Soil spectral reflectance of 0.154 mm had the highest correlation coefficient(r=0.66)with SOM content.The SOM model based on 0.154 mm soil spectra and partial least-squares regression method performed best.By combining the correlation analysis and considering the robustness and complexity of SOM model,0.154 mm proved to be the optimal soil partical size to monitor SOM content in this study.(3)Different spectral preprocessing methods generated different effects on the correlation of soil spectra with SOM and also the PLSR model of SOM.The combination method of Savitzky-Golay smoothing and first-order derivative(SG-FD)produced the highest correlation coefficient with SOM(r=0.83).All preprocessing methods,excluding multiplicative scatter correction(MSC)and standard normal variate(SNV),improved the accuracy of SOM models.Especially,SG-FD method significantly increased the PLSR model accuracy of SOM and performed best(R2c=0.97,R2v=0.85 and RPD= 1.78).(4)SOM sensitive wavebands selected by successive projection algorithm(SPA)method changed with different transformed spectra.Combining the correlation analysis and all selected wavebands,the SOM characteristic wavebands were located in the range of 400-410 nm,490-510 nm,550-560 nm,660-700 nm,1030-1200 nm,1400-1600 nm,1840-1970 nm and 2310-2440 nm.(5)Compared with PLSR method,SPA-MLR reduced the SOM calibration models accuracy while SVR decreased the SOM validation models accuracy.Among all models,the SOM model based on SVR method and MSC spectral preprocessing method performed best(R2C=0.97,R2v=0.85 and RPD=1.78)Considering the practice and application,the SOM model based on PLSR-SG-FD method and SPA-MLR-SNV method(R2C=0.83 and 0.79,R2v =0.84 and 0.91,and RPD=1.95 and 2.81)can also employed to estimate SOM content.Therefore,the SOM models established in this study have good predictive capacity and achieve the accurately estimation of SOM content.