Study On The Quantitative Estimating Of Soil Properties With Hyper-spectrum In Some Parts Of Jiangsu Province

Soil is indispensable for the birth of human civilization and development. Excessive pollutants were input to soil due to the development of modern industry and soil productivity is reducing due to degradation besause of the development of modern agriculture. To achieve efficient, high-quality agricultural production, it is essential to determine soil properties fastly and accurately. Field sampling and chemical analysis are needed in traditional soil evaluation which is time-consuming, laborious. We must find new methods to meet the development of the modern precision agriculture.Soil spectrum reflects physical and chemical properties of sample. Spectroscopy analysis of soil is a rapid, cost-effective, non-risk, non-destructive technique which can predict some characteristics of samples simultaneously. It provides new means and methods for soil study. It is a new way to monitor soil real-time in large scale by the means of remote sensing in particular hyperspectral remote sensing.There are characteristic absorption bands in the spectrum of different matter due to electronic and vibrational processes. These absorption bands or there combination are the basis for identification and quantitative estimation of constitutes. Soil is a complex organism composed of many substances and soil spectrum is the combined result of the composition. We can determine the physical or chemical properties of soil quantitatively using soil spectral features, but still need to explore the mechanism and rule. The main conclusions are as follows:(1) Soil organic matter (SOM) and clay minerals which are important constituents of soil have significant spectral features. But some trace elements for example heavy metals and chemical properties such as pH have no spectral features. We can estimate the content of heavy metals or chemical properties quantitatively because of their correlation with iron, organic matter, clay.(2) We used partial least squares regression (PLSR) to establish a relationship between reflectance spectra in the visible-near-infrared (VNIR) region and some soil characteristics and extract latent variables. We also predicted these characteristics by the means of artificial neural network (ANN) taking latent variables as input. The results of ANN are better than PLSR. The pre-processing of first derivative (FD) can get the best result because it can remove part of the linear or nearly linear noise which affecting the target spectra. The results SOM with different resolution data indicate that the outcome of calibration, validation and test becomes worse with the rising width of sampling resolution. The method of support vector machine (SVM) can get similar but more stable results than ANN using the parameters got from three methods of searching the best parameters.(3) Smoothed spectra and first derivative spectra of soil can be used to estimate SOM, total nitrogen, pH in Dongtai city and SOM, As, Ca, Na, pH in Yixing city. However, the results of available phosphorus, available nitrogen, total salt, Na, Cl, and C/N (ratio of carton and nitrogen) were poor. Using principal component analysis (PCA) method we got the scores maps of the surface soil and average spectral in profile. We found that the average spectral of the profiles can better reflect soil category than the surface soil.(4) There is negative correlation between reflectance and SOM in the entire range of 350-2500nm in these three cities. The important range affected by SOM is in visible-near infrared. The curve of correlation coefficients between reflectance and SOM in Dongtai city is different from them in Kunshan and Yixing cities prohapes due to the difference of composition and decomposition phase of SOM. The curves of correlation coefficients between reflectance and total nitrogen in these three cities are very similar. The inter correlation between N and SOM is the key factor for obtaining the ability to predict the contnet of N. The mechanism of estimating pH by reflectance is the correlation between pH and SOM, clay minerals. But this machanism varies in different regions because the influence factors are different.(5) We used the Simulated TM and ASTER reflectance spectra to predicted 4 soil properties using partial least squares regression and stepwise multiple regression.The results show that PLSR is better than multiple regression and ASTER better than the TM.(6) Reflectance of surface in the location of sampling in ETM image is affected by vegetation. The relation between reflectance and As, SOM, pH is worse. We can resolve this problem of mixing pixel due to vegetation in certain extent by the technology of spectral unmixing, but we must measure the spectroscopy of soil and vegetation in the same while.