Abstract: In the present study, soil samples were prepared from Zhangjiagang City to establish the quantitative inversion models of the soil heavy metals contents. The contents of the soil heavy metals and the visible and near-infrared spectra of the soil samples were obtained in a darkroom. Firstly, the original hyperspectral data was smoothed and the spectral transformations such as first derivative, reciprocal of first derivative, logarithm of reciprocal of first derivative, square root of first derivative and continuum removal were carried out. Secondly, the characteristic bands of different transform spectra were extracted through correlation analysis. Finally, quantitative estimation models of heavy metals contents were established by stepwise regression. The results showed that Cd, Hg, Cu and Zn in the farmland soil of Zhangjiagang City exhibited certain pollution risk. The correlation coefficient within the first derivative or the continuum removal and heavy metals contents were higher than that of other transformation forms. Eight quantitative estimation models of soil heavy metals contents and hyperspectral data possessed good prediction accuracy. The fitting degrees of the actual and verified values of the estimated models for Cd, Hg, Cr, As, Cu, Zn, Ni and Pb were 0.874, 0.879, 0.800, 0.646, 0.513, 0.655, 0.603 and 0.542, respectively. Therefore, hyperspectral data could be used to predict the contents of soil heavy metals in farmland in Zhangjiagang City.

Key words: farmland soil pollution, heavy metal contents, hyperspectral remote sensing monitoring, model accuracy