Spectrum Variation of Vegetation in Yanzhou Coal Mine Area and Heavy Metal Stress Characteristic
|School||Shandong University of Science and Technology|
|Keywords||heavy metal spectral characteristic stepwise regression analysis Correlation analysis artificial neural networks|
In this paper, Jining No.2 and Jining No.3 coal mine area were selected as research area, we collected Zea mays, Glycine max Triticum aestivum Linn, P. Tomentosa leaves’field spectral data and measured the content of Cu, Mn, Pb and Zn in these specific plants’leaves, telling us to what extent and type the vegetation was polluted by heavy meatal in research area. The response message between same species vegetation and different heavy metal content in leaves has great difference, and the difference also exist between same heavy metal and different species vegetation. For example, heavy metal content of Zea mays leaves in J2 coal mine area stayed in the normal range; the content of Cu, Mn and Zn in most of Zea mays leaves in J3 coal mine area were higher than normal range and stayed higher than those in J2 coal mine area. The content of Mn was the highest in the three heavy metals. The content of Mn and Cu in all Glycine max leaves, the content of Pb and Zn in most Glycine max leaves stayed in the normal range, Pb and Zn in several Glycine max leaves stayed beyond the normal range. I extracted 23 spectral parameters from the spectral curve transformed from the field spectral data and analysed the vegetation spectral characteristics. Different spectral characteristics had different response feature, the vegetation index feature were not obvious. The correlation analysis between spectral parameters and heavy metal content of vegetation leaves in coal mine area indicated that there was approximate correlation coefficient between Max reflectance of green peak, green peak, mean value of green peak reflectance and same heavy metal content in vegetation leaves in coal mine area, so the correlation between the three spectral parameters and the same heavy metal content was similar. In all the stepwise regression equations, multiple correlation coefficient of the equation between spectral parameters and Cu content of Zea mays leaves in J3 coal mine area and of the equation between spectral parameters and Pb content of P. Tomentosa were the biggest, the two equations had high accuracy. The establishment of stepwise regression equation and the correlation analysis between spectral parameters and heavy metal content of vegetation leaves in coal mine area indicated that the spectral parameters with the highest correlation coefficient between all spectral parameters and single heavy metal content in leaves were bound to enter into the stepwise regression equation, and the parameters with low or no correlation coefficient also had the possibility to enter into the equation. The simulated model by neural networks between biochemistry components in vegetation(Triticum aestivum Linn and Zea mays in J3 coal mine area) and single spectral parameter (Red valley area, TVI) indicated that Radial Basis Function neural network(RBF) is better than BP neural network on simulating these type of data.