Study On Prediction Of Soil Organic Matter Based On Hyperspectral Remote Sensing

With the development of hyperspectral remote sensing technology, acquiring the soil nutrient information by using spectral techniques has been received more and more attention in recent years. Soil organic matter, as an important part of the soil, not only in the aspect of soil fertility, environmental protection but also in agricultural sustainable development plays an important role and has a great significance. Using hyperspectral remote sensing techniques to estimate the content of soil organic matter is an urgent need of the development of modern agriculture, but also the inevitable requirement for the development of precision agriculture.In this study, a total of1581soil samples of16soil types which collected from14provinces in China were firstly physico-chemical analyzed and measured using an ASD spectrometer under laboratory condition with a range of350nm to2500nm. And then, the spectral reflectance curve, first derivative curve and continuum curve were made. Moreover, correlation analysis was done between soil organic matter content and spectral reflectance, first derivative of spectral reflectance, respectively. The results show that under different organic matter level, the first derivative curve and continuum curve of soil spectral reflectance have a more obvious performance with the decrease of the organic matter content and show more prominent features in the special bands. The soil spectral reflectance was affected obviously by other factors such as iron oxide, etc. when the content of soil organic matter stayed at a low level. Moreover, this study shows that different soil types have large differences in the performance of first derivative curve and continuum curve of soil spectral reflectance in the full range of spectrum, of which, black soil, red soil, purple soil, and meadow soil due to the large differences in the content of organic matter showed the most obvious features. What’s more,600-800nm band can be used as common characteristic bands of these different soil types from different region in the study.Using Visible/near infrared spectroscopy to predict soil properties and build a model is very important in research of proximal soil sensing. It can be applied to many aspects such as rapidly access to soil information and precision management in argriculture. In this part of study, paddy soil in Zhejiang Province is treated as the research samples. The nonlinear models such as random forests (RF), supported vector machines (SVM) and artificial neural networks (ANN) were used respectively to build models to predict soil organic matter based on different selection of validation and calibration datasets. The results show that there is a certain impact on prediction results under the different division of sample modes. Compared to the linear model PLSR which has been commonly used, the nonlinear model RF and SVM have comparable prediction accuracy, the value of RDP were all more than1.4under three kinds of different calibration datasets, especially predictions by using SVM with full range of Vis-NIR wavelengths produced the smallest RMSE value and the largest R2. It shows that the model constructed by SVM method has a good predictive ability. In addition, a combined method which is PLSR-ANN (with the introduction of ANN into PLSR) significantly improves the predictive ability of PLSR which was used alone. The RDP reached2.36under mode2. Even though ANNs are "black box" systems the combination of linear model "PLSR" and nonliner modelling helps to achieve good predictions and interpretability.In the third part of study, hyperspectral data of253soil samples from Hunan, Zhejiang and Fujian provinces of different soil iron oxide (SIO) content and soil organic matter (SOM) content were analyzed to investigate the effect of SIO content on hyperspectral characteristics and quantificational inversion of SOM content. The results showed that:range of600-1400nm is the characteristic wave band of SIO content. What’s more, if SIO content>30g/kg, the hyperspectral information (HI) of SOM would be covered up by SIO;20<SIO content<30g/kg, HI of SOM in visible band would be influenced; SIO content<20g/kg, the hyperspectral information (HI) of SOM would not be affected by SIO. At the same time, when the ratio of SIO content to SOM content was more than2.21, HI of SOM content would be shadowed by SIO totally; when SIO content/SOM content ranged from1.05-2, HI of SOM in400-1300nm band would be influenced seriously, but that in1300-2400nm would be affected slightly; when the ratio of SIO content to SOM content was less than0.726, HI of SOM would not be affected by SIO. In addition, SIO content can influece the hyperspectral quantitative inversion of SOM. The inversion accuracy and stability of prediction model estimating SOM content decreased with the increase of SIO content or the ratio of SIO content to SOM content. But the effect of SIO content on quantificational inversion of SOM was slight when SIO content was less than20g/kg or SIO content/SOM content was less than2.0.