Abstract: Abstract: Heavy metal pollution exists in many mining sites, and heavy metal in soils poses a great potential threat to the environment and human health. Therefore, it is urgent to estimate heavy metals in farmland in tailing areas of mining sites. The goal of this research was to estimate copper content in farmland of a tailing area based on visible-near infrared reflectance spectroscopy. This research took Jinduicheng mine tailings in Shaanxi as the study area. A total number of 288 soil samples were collected at the mining areas. The soil samples were divided into two groups, a training/calibration set (n=252) and an external validation set (n=36) for the Cu estimation model. The soil samples were air dried and passed through a 2 mm sieve. The Cu concentrations in soil were determined through chemical analysis in the laboratory by graphite furnace atomic absorption spectrometry (GB/T17141-1997). The visible-near infrared reflectance spectral measurements of soil Cu concentration were collected using an ASD field spectrometer for the solar reflective wavelengths (350-2500 nm) in the laboratory. The 8 angle probe was used, the distance from the contact probe to the surface of soil samples was set to 1.35 m in order to get the soil spectral in the range of 1 m2, and each soil sample was achieved 10 spectral measurements. The original reflectance was transformed with a db6 wavelet analysis. The Isomap (Isometrio Mapping) and LLE (Locally Linear Embedding) manifold learning methods were applied to the hyperspectral data of soil for dimension reduction, parameter of k and d was 10 to 50 and 8-15, respectively. Copper concentration in the mine tailing soil was estimated by the method of random forests. The estimated results were compared with the original hyperspectral data and the dimension reduction spectral data. The results showed that the spectral characteristics of the most important values were at the wavelength of 475 802, and 868 nm. The estimation model had a better performance on dimension reduction spectral data set than that on the original spectral data set, and the estimation model achieved coefficient of determination R2 of 0.7586 on the spectral data set after dimension reduced by Isomap, and the RMSE (root mean square error) was 30.50, the estimation accuracy was better than that on the dimension reduction by LLE, but the accuracy needed to be improved. The results provide a theoretical basis for rapid estimation copper content of farmland soil in the tailing area, and will provide theoretical basis and technological support for controls of mining tailings and mining wasteland and its ecological restoration and reconstruction.