Feasibility of estimating heavy metal concentrations in water column using hyperspectral data and partial least squares regression

Mining and smelting often produce acidic wastes that can cause severe biogeochemical changes downstream from these mines. Dexing copper mine, as the largest open cast mine in China, is connected to Poyang Lake by Le An river. Water and spectra samples were taken from Le An River and two of its branches, and afterward the concentrations of Cd, Cu, Pb and Zn were measured in the lab. Different spectral pre-processing methods were applied to the spectra, including Savitzky-Golay spectral smoothing, SNV, first derivative, second derivative spectral transforming. On the purpose of estimating metal concentrations from differently pre-processed spectra, partial least squares regression was then used in model calibrations. For deciding the optimal number of PLS factors included in the PLS model, the model with the lowest root mean square error of validation is chosen. The coefficient of determination (R 2 v ) between the predicted and the reference values from the test set are used as an evaluation mean. For estimating Pb concentration, R 2 v = 0.915, which is acceptable. For Cd concentration, R 2 v = 0.697 and 0.683 for Zn. PLS model seems to failed in estimating Cu concentration, for the best R 2 v for PLS model of Cu is lower than 0.5. From the aspects of spectral pre-processing methods, first derivative after Savitzky-Golay smoothing performs superior to others. In conclusion, PLS models based on carefully pre-processed hyperspectral data turn out to be a promising solution for detecting certain heavy metals concentrations in river.