| At present, supervision on soils polluted by heavy metals is a development hot-spot in the field of environmental science and geoscience, GIS technology and remote-sensing technology have offered a strength way for supervising soils polluted by heavy metals in time and space. With the increasing development of remote-sensing technology, especially in hyperspectral technology, a lot of study case had solved the forecast for which the high spectrum data show the contents of the organism in soils successfully, and some explore forecasting works for the content of the microelement metal(heavy metal) have begin and make somewhat process.This article have reseached the space-distrubition character of areas polluted by heavy metals of soils in Le An river in Jiangxi province, including its branches areas caused by flood water, have studied the probability of the content of heavy metal in soils that we use the hyperspectral data to forecast. Actually, according to the content of the heavy metal only small in soils, it’s difficult for us to use reflecting-spectrum to forecast and not reliable for a big error. This article have explorated the probability of forecast the content of the heavy metal in the way of using organism of soils as a medium.The areas this article researched, is a typical polluted areas by heavy metals in Le An river areas, Jiangxi province. We study a vast areas more than 1000 square meters. In this article, we use the spectrum data from which the ASD fieldspec3 instrument measured the sample of soils, and pretreat the soil’s spectrum we collected with the differential coefficient technology to analyse and then make the spectrum data smooth, and select the band, such as band of 415,485,770 and 920 as a characteristic band, according to the need of the post-forecasting mold.It is significantly important for us to treat the data that is a key in our all research, especially in researching remote-sensing hyperspectral relevant to whether the way of sample collection is scientific and reasonable or not, whether questions need to solve in our research is relevant or not, whether the sample treatment is proper or not, whether the data we collected is true and reliable or not, whether quantitative deducting work for spectrum collected is precision. In this article, we have adopted the unfreezing way with potassium dichromate and intensity vitriol liquor to measure organic content of soils and adopted XRF instrument to measure the content of the heavy metals simultaneously.It’s the first step to forecast the organism in soils. First, we validate the forecasting mold of the research sample with the classical method to forecast the organic spectrum in soils. According to Krishnan et al.(1980) and Dalal Henry(1985)’s research, it’s not reasonable for us to the forecasting result(R2=0.245) by introducing our research data of spectrum into the classical mold to validate, but it’s suitable for the traditional way to forecast mold for the organism. We have adopted the Partial Least Squares(PLS for short) way, which is a optimization technology in mathematics to forecast the content of the organism, to calculate the square summation of the minimal tolerance so as to find a most suitable function from a group data, and used the simply way to calculate the true value that can not know absolutely and make the square summation of the error as little as possible. After establishing regression mold in this way, the result shows that the astringency of the regression coefficient of the mold is perfect, and its effects is perfect either. When the main composition became five, the variance coefficient of the mold is the least and the relativity of the mold is perfect:R2 is 0.9842 and RMSE is 0.9511 that is perfect. We have chosen 10 soil samples numbering:004,009,019,022,037,048,049,063,069 and 070 as testing sample, to calculate the precision that is perfect:test precision is 0.6184. But the offset of the testing sample and the RMSE is 11.23 and 4.56 separately that exist somewhat error and offset. It’s possible that the single testing sample chosen at random comparative with others have great difference. Actually, the precision of the whole mold and its effect make us a satisfactory. Then we compare it with the linearity mold, the logarithm mold, the cubic function mold and the exponent mold, result as follows:the linearity mold R2=0.4878, the logarithm mold R2=0.4598, the cubic function mold R2=0.5108 and exponent mold R2=0.3786. The result can’t reach the precision level that the PLS reached(R2=0.9842).Above research have discussed the relations in detail about the content between heavy metal and organism in soils. The organism in soil can absorb the large heavy metal ions, keep them fastened from been adopted by plants. So theoretically, heavy metal and soil organic matter content have a positive correlation, the conclusion of study also confirmed this.It’s the second step to establish the mold, forecasting the content of heavy metal of soils from forecasting the organism of soils. The article adopt the Artificial Neural Network(ANN for short) calculation to realize the forecasting content from organism of soils to heavy metal of soils. The Artificial Neural Network also been named Neural Network composed by large neurons widely connecting each other, which reflect basic function of the mankind’s brain with abstract simplization and simulation. The research on the Artificial Neural Network from the brain structure of the mankind to research the man’s, intellective action stimulate the man’s brain function on dealing with the information. This article consider a indirectly way to forecast. Although we can’t realize and forecast them directly form the angle that the heavy metal absorb the spectrum, the heavy metal element always be relvant to the some active component of spectrum in soils, such as organism. By this, we successfully forecast the content of the heavy metal(Cu, Zn, Pb) of soils in plain with reflecting the spectrum. By the Artificial Neural Network mold, the forecasting result of the Cu and Zn is satisfying, Zn’s forecasting precision is the highest and the Pb’s by contraries. In three forecasting mold, Zn’s effect is most perfect reaching to 0.9370 in total relativity Cu’s effect is comparatively perfect reaching to 0.7763 in total relativity, but Pb’s effect is not satisfying for a great offset, only 0.4317 in total relativity. Totally, the comparative offset take place in the samples between the highest content and lowest of heavy metal, in the medium polluted areas, the effect of the forecasting mold is comparatively perfect. On the other hand, the data of the content of heavy metals is compose by two steps which are the organism forecasting mold and heavy metals forecasting mold and then make final forecasting result offset great.Above-mentioned research have offer a method to solve the key and difficult point in forecasting work of the heavy metal spectrum of soils and make a basic ground for further research. |
PDF Full Text Request