Forecast And Sedimentary Environment Component Based On Sediment Characteristics Indicating High Spectrum

Lake and reservoir sediments are complex mixtures of aquatic ecosystem which were determined by their source. The sediment profile can preserve useful changing information about the past history of water body and the watershed, and thereby to help to control regional pollution. However the traditional chemical and physical laboratory analysis methods are not only time-taking and labor-consuming, the contaminating chemicals also pose threaten to environment, otherwise the uncertainty of error may rise up duing to the complicated procedure. Diffuse reflectance spectroscopy (DRS) technique is a non-destructive, rapid, convenient and environment-friendly methodology which has been used to observe the properties of soil and marine sediment.Three sediment profiles were collected at Dashiba reservoir in Dianchi Basin, Kunming, Yunnan Province. The reflectance of the samples were measured by spectrophotometer coupled with integrating sphere, ranging from350nm to2500nm. The main purposes of this paper are constructing sediment predictive models and comparing distinctive sediment profile condition by spectral index. The main results are as follows:(1)Principle component analysis (PCA) was performed on the pre-processed data, the score plot of the first two PCs showed the relation among samples. Those points lying outside the95%confidence ellipse (Hotelling T2) were regarded as outliers which were removed from the matrix. As a result,8samples from DSB3were eliminated.(2)Two kinds of linear calibration techniques namely principle component regress-ion (PCR) and partial least square regression (PLSR), and non-linear back-propagation artificial neural network (BPNN) were adopted to relate chemical content (TOC, TN, TP,δ13Corg,137Cs and210Pbex) to the pre-processed spectra. The results show that the non-linear models have a better predictive ability than the linear ones. Excellent results were obtained from BPNN and combined PC-BPNN and PLS-BPNN models. The best prediction was achieved by BPNN model for TOC (R2=0.853, RMSE=2.121, RPD=2.604, RPIQ=2.113), and PC-BPNN models had the highest predictive accuracy for TN (R2=0.847, RMSE=0.299, RPD=2.554, RPIQ=3.362) and210Pbex(R2=0.319, RMSE=167.787, RPD=1.212, RPIQ=1.541), PLS-BPNN models outperformed other techniques for TP (R2=0.970, RMSE=0.394, RPD=5.799, RPIQ=2.302),δ13Corg(R2=0.881, RMSE=0.327, RPD=2.896, RPIQ=4.533) and137Cs (R2=0.313, RMSE=1.180, RPD=1.206, RPIQ=1.390). Based on the response to spectra, primary category properties TOC and TN and second category properties TP and δ13Corg had excellent predictive result, while calibration results were less reliable for137Cs and210Pbex. Comparing predicted properties value and measured value for the three sediment profiles, it showed that they had the same changing tendency for TOC, TN, TP,δ13Corg which could be used for quantitative determination, whereas only the peaks appeared at the same depth for137Cs and210Pbex which could just use for semi-quantitative prediction.(3)Brightness and redness were calculated for retain samples from raw reflectance data. Additional second PCA analysis were performed for retained samples, the first four principle components explaining97.74%variation of the matrix. The correlation between extracted spectral index (brightness, redness and principle components) and sediment properties (TOC, TN, TP,δ13Corg), particle size. The results showed redness was negative correlated with TOC, TN and TP, while positively correlated with δ13Corg. PC1presented better positive correlation with clay content (<4μm) and negative with sand content (>64μm). The scatter plot of these two spectral indexes succeed classified sediment profiles. Samples from DSB1located in low score and high redness region, which were mainly affected by reservoir water level. DSB2had a wide spread, mainly gathered in high score value and low redness region. DSB3had the same location with DSB1, however part of samples located in the region of DSB2which attribute to the complex effected by the small valley. It showed that redness and PC1extracted from diffuse reflectance could be used for rapid distinguishing sediment source. Besides, the result shows that different sample-point in small basin represent distinctive sediment conditions which vary much.