| Due to its wide source,certain toxicity and difficulty in degradation,heavy metal elements will eventually accumulate in the human body through the food chain and endanger human health.Heavy metal pollution has become an important research direction in ecological science.With the rapid economic development in recent decades,the ecological environment in coastal areas of Jiangsu province has been damaged diversely.Tidal flat,as a typical region of sea land interaction zones,Jiangsu coastal has a wide distribution.A large number of heavy metal elements are absorbed into the sea,making the tidal flat area as a‘sink’and‘secondary source’of heavy metals.Therefore,it has a great significance to protect and restore marine ecological environment by understanding the characteristics of heavy metal pollution and the current ecological risks in tidal plain area.Traditional heavy metal monitoring methods are time-consuming and laborious,it is difficult to carry out extensive detailed monitoring.Hyperspectral technology has the advantages of high efficiency,high speed,real time and macroscopical.In this study,the spectral information of tidal flat surface sediments was used to estimate the heavy metal content,which also can provide theoretical basis and technical support for the large-scale monitoring and prevention of heavy metal pollution.In this paper,the research area was located in the sea area around Yangkou port in Jiangsu province.Multiple tidal flat surface sediments were collected to measure the particle size,TOC,TN,heavy metal elements contents(Cu,Zn,As,Pb,Cr,Ni)and sediment spectra in the laboratory.Based on multivariate statistical analysis and multiple risk assessment methods,the degree of heavy metal pollution and ecological risk were evaluated.We also discussed the possibility of using spectroscopy to predict heavy metal content.Stepwie multiple linear regression,partial least squares regression and generalized regression neural network methods were used to estimate the sediment heavy metals concentration with hyperspectrum.The main results are as follows:(1)The main sediment types are sandy s Z and Zs.There has significant differences in the types of sediments due to the complex hydrodynamic conditions in the north,than that of in the south.The content of TOC and TN was significantly correlated with the average particle size,and the content value of the northern region was higher than that of the southern,TOC is more affected by the hydrodynamic environment than TN.(2)Except As,the other heavy metal elements in the study area were all below GB18668-2002.And Pb and Ni contents were higher than the background values of Jiangsu sea sediment.There was a spatial difference among As,Pb and Zn,which is high on the south side and low on the north,while the spatial difference of Cu,Cr and Ni is not obvious.(3)Multivariate statistical analysis showed that the six heavy metal elements,except Ni,were mainly derived from human activities,such as transportation,chemical industry,pesticides,and the use of feed additives.The five elements also had a high correlation with each other.Evaluation methods indicate that the pollution degree of Cu,Zn,Cr and Ni is‘relatively low’or‘low’pollution,but the pollution degree of As and Pb is‘medium to high’,which makes the whole area presents‘medium and high’or‘high’ecological risk.And the ecological risk danger degree of the south side is higher than that of the north side.(4)After breakpoint correction,SG smoothing,first derivative,second derivative and continuum removal,the spectral characteristics were highlight effectively.The band sensitive to heavy metal response was extracted and the correlation of the band was further improved after certain combination.Finally,we selected 7,7,8,6characteristic variables for elements As,Pb,Cu,Zn,and the absolute values of correlation coefficients were no smaller than 0.67,0.75,0.73,0.65.(5)Models of stepwie multiple linear regression,partial least squares regression and generalized regression neural network methods has a good modeling accuracy for Pb,the R~2 of modeling were more than 0.75.But the R~2 of Cu modeling was only 0.54.The prediction accuracy of the three models were different.For the prediction of As,the prediction accuracy of stepwie multiple linear regression and generalized regression neural network is better.For Pb,the generalized regression neural network has the best prediction effect,with a determination coefficient of 0.658.For Zn,the prediction accuracy of stepwie multiple linear regression was the best,with a determination coefficient of 0.638.The prediction accuracy of the three models for Cu is not good,and the optimal prediction determination coefficient of partial least squares regression is only 0.416. |
