Study On The Soil Heavy Metal Prediction In Tuojiang River Based On TM Data

Aiming at finding out an easy and quick monitoring method on quantifying the heavy metal pollution of soil, this study establishes a linear regression prediction model by using each band spectral reflectance of TM remote-sensing image,9 vegetation indexes,4 ground cofactors, and the measured value of heavy metal content of 80% sample soil. This model takes the factors which are not involved in the modeling but have significant correlation with heavy metal content of soil as correction factors and runs Monte Carlo simulation. The rest 20% sample soil becomes checking point to do the trend analysis and error statistics test for all kinds of models. In the end, this study inverts the spatial distribution of measured value and predicted value and contrasts their differences with ordinary kriging method.The main conclusions are as follows.Using the TM image spectrum information and other factors can predict the content of As, Hg, Cr, Cd, Cu, Pb, Available Mn, Available Fe and Available Zn, model of Available Fe is better. Modeling factors are extracted from TM remote sensing images and digital elevation model, which is easy to get, can effectively save cost and time.All the three kinds of modeling, separate modeling for each band spectral reflectivity of remote-sensing image, common modeling with ground factors, separate modeling for different landforms, can predict heavy metal content (P<0.01), R2 value, average error, total-root-mean-square error, and average relative error of the soil of research area, and their precision has a progressive increase. Therefore, it is necessary to establish prediction model based on different landforms when using remote sensing to study soil heavy metal in complex-landform areas.Monte Carlo simulation can correct the prediction models. Compared with the data of linear regression prediction model, the root-mean-square error of arsenic, mercury, chromium, cadmium, copper, lead, active manganese, active ferric, and active zinc are respectively reduced 22.7%,18.0%,37.7%,36.2%,13.5%,30.5%,5.4%,25.2%, and 11.9% after corrected by Monte Carlo simulation, which indicates that Monte Carlo simulation can improve the model precision, but the differences between correction factors and the degree of correlation of target value can make different correction factors have different effects.The trend analysis, error statistics test, and spatial distribution maps show that the prediction model corrected by the geometric method of band 2 of As and the ARVI is the best, the prediction model corrected by band 3 of Hg is the best, the prediction model corrected by NDVI of Cr is the best, the prediction model corrected by the gradient of Cd is the best, the prediction model corrected by SLAVI of Cu is the best, the prediction model corrected by the geometric method of band 5 of Pb and SCI is the best, the prediction model corrected by band 3 of Available Mn is the best, the prediction model corrected by band 1 of Available Fe is the best, the prediction model corrected by the geometric method of band 7 of Available Zn and SLAVI is the best.