Characterization Of Soil Reflection And Modeling SOM Distribution Along Profile Using Imaging Spectroscopy

The research on soil properties of soil profile and different genetic horizons is significant to soil genesis, development and classification etc.. The traditional soil survey is time-consuming and effort-waste. Proximal soil sensing(PSS) using visible near infrared(VIS/NIR:400-2500nm) spectroscopy that the main band range of quantitative analysis in soil properties is a tool that provides more abounding and contiguous components and distribution information. The VIS/NIR spectroscopy of which the spectral resolution is greater than10nm can acquire fine spectral response characteristic. And imaging spectroscopy has hyperspectral resolution and high spatial resolution in simultaneously. So, using imaging spectroscopy on the soil profile research can perform the soil spectral quantitative analysis and fine mapping of soil profile properties. And using the information in research on the soil substance transport, abundance in indexes of soil classification and quality assessment has value.In this research, the Headwall imaging spectroscopy (HyperspecTMVNIR,400-1000nm and HyperspecTM NIR,900-1000nm) was using for acquiring hyperspectral images of soil profile and soil samples of different particle sizes in different genesis horizons (not-messed soil subsamples,0.28～0.90mm,0.20～0.28mm,0.15～0.20mm, and <0.15mm soil subsamples). Then, the mean reflectance spectral curves of subsamples were extracted and analyzed the characteristic of soil spectral response of different particle sizes and different genesis horizons samples. Using the mean reflectance spectrum of samples based on partial least squares regression (PLSR)/v-support vector regression (v-SVR) built the soil organic matter (SOM) estimating models. The performance of two models was comparative analyzed. Using the estimating models predicts the SOM content along profile, and then acquired the map of distribution of SOM.The conclusions of the research as follows: 1. The surface roughness of soil samples plays an important role in soil reflectance spectrum. The soil spectral reflectance increases with reduction of particle size. And the spectral reflectance of <0.15mm soil sample is much higher than other particle sizes. The overall spectral reflectanceof the not-messed soil sample increased after removing the shadow of it. And the spectral reflectance curve is closed to the curve of0.28-0.90mm.2. With depth of horizons increasing, the soil spectral reflectance along profile increases. In the paddy soil profile, the spectral reflectance curves are closed between the tillage layer (A) and the plow layer (P), and they are far from that of the gley layer (G). In the fluvo-aquic soil profile, the spectral reflectance curves are evenly distributed from the tillage layer (A) to the prototype layer (B1), and to the redox layer (B2). The change characteristic of the spectral reflectance curves of the layers is affected by the change of SOM contents and texture among layer and related to differences in soil tillage methods and fertilizer management.3. There is good correlation between soil mean spectral reflectance and SOM. The precision requirements of PLSR and u-SVR SOM estimating models using the spectral data of the different size samples is all met. The coefficient of determination (R2) is up to0.8, and the residual prediction difference (RPD) is basically up to2. The results show that in the same particle size the performance of the u-SVR estimating model is better than the performance of PLSR. On the contrasting of different particle sizes SOM estimating models, the model precision based on <0.15mm soil sample is the best, but in v-SVR models, the models of0.28-0.90mm soil sample is best. For the not-messed soil sample, the shadow removed when the mean spectrum was extracted plays a not-great hole in model precision. And then, the SOM contents along the soil profile were estimated using SOM contents estimating models. But the performance of the u-SVR estimating models is very bad. At last, the distribution of SOM contents along soil profile was estimated using <0.15mm soil sample SOM contents estimating models. The results showed that SOM contents of paddy soil profile was slowly decreased with the increase of depth, and overall trend of SOM contents along the fluvo-aquic soil profile was the as the paddy soil profile. But in depth of45cm, the SOM contents slightly increased.