Estimation Of Heavy Metal Contents In Soil Using VNIR Spectra

Soil is foundermantal to agricultural produticon as well as an important aspect in land resource management. Heavy metal pollution in soil has become a critical problem in China today. Every year 12 million tons of food is polluted by heavy metal contaminations. Heavy metal pollution has already threatened national food security and the health of citizens. For a better management of limited arable land resources and a scientific land use planning, it is important to acquire the heavy metal contents in soil and present the information in an easily understandable manner.Taking the Le’an River floodplain as a case study, this thesis integrates knowledge from multi-disciplines, and focuses on the efficient estimation of heavy metal content in soil and the comprehensive mapping of spatial distribution of the pollutents.The Dexing Copper Mine, located at the upper and middle reaches of the Le’an River, is the largest outcrop copper mine in Asia. The ore prodution reached up to 60000 tons as early as in the 1990s. The other major mine along the Le’an River, the Zinc and Lead Mine also has a large-scale production of nonferrous metals with a history over 40 years. The mining activities in this area produces large amounts of alkaline and acid mine drainages, which are rich in heavy metals such as Cu, Pb and Zn. The drainages with high heavy metal concentrations enter the Le’an River floodplain by the transportation of the Le’an River, and further Poyang Lake, which could induce serious ecological and environmental impacts on the water, soil and plant in these regions.Research interests concerning the environmental drawbacks from mining and smelting activities in this area dated back to a subproject of a joint program Co-operative Ecological Research Project (CERP), namely "Ecological effects of heavy-metal pollution in the Dexing copper mine region". This subproject involved a variety of issues including organisms (animals and vegetation), water, soil and sediment. The technique used was mostly geochemichal survey. Twenty years past, the traditional geochemical survey could not meet the increasing need of monitoring the environmental status over large areas effectively. Therefore, with the development of remote sensing technique and geoinformation science, it is possible and necessary to apply these newly developed techniques in the land resource survey and soil quality monitoring, which has now become an inevitable trend.This paper consists of two topics. The first topic is made up of Chapter Three, Chpater Four and Chapter Five, discussing the feasibility of estimating heavy metal contents in soil using VNIR spectra. The second topic is about mapping the heavy metal contents in soil, which is described in Chapter Six. The state-of-art researches are reviewed, the problems are identified, the methods used and the major findings are introduced. The last part of the abstract lists the essential innovations of this study.Topic one:estimations of heavy metal contents in soil using VNIR spectra.Currently, the estimations of soil contents, such as organic carbon and total N has been well developed using VNIR spectroscopy. The estimation of soil heavy metal content, especially for those trace heavy metal content, has seldom been reported despite of the ten year history of its development. One of the major reasons is that: most trace heavy metals in soil are featureless in the VNIR wavelength region. Some recent researches show that it is feasible to estimate the heavy metal contents in soil using VNIR spectra at local scale. These researches, however, have the following deficts:(1) The soil samples are collected at local scale, with homogenious soil orders, similar land use and land cover. Research carried out at larger scales, continental scale and global scale for instance, show that the VNIR estimations of heavy metal content in soil samples from different soil types, land use and land cover remain challenging.(2) Previous studies used samples which went through a series of pretreatment. The moisture content of soil and soil particle size, however, could significantly influence the spectra of soil sample, making a potential discrepancy between models from pretreated soil samples and those without pretreatment. Finding solution to such discrepancy or even simply exploring the differences between these models may fill a gap in the VNIR estimations of heavy metal content in soil, considering that there are no such peer researches considering these issues. (3) Coupled with multiple linear regression, most current researches use one or more bands, or their combiniations as explanatory variables to estimate heavy metal content. It is easy to depict the data distribution of explanatory variables with a relatively small number, e.g.10 explanatory variables. In case of VNIR specscopy, the explanatory variables are usually in number of thousands. Therefore, current methods such as Lillifors normality test could not test and display the data distribution of thousands of variables at the same time. For this reason, most current researches consider only the data distributions of response variable(s), neglecting examing the data distributions of explanatory variables.Based on the above analyses, two statistical plots, namely Lilliefors normality test plot and the kurtosis and skewness curves plot are proposed in Chapter 3. With a case study, the utilities of these two plots are illustrated. They can simultaneously analyze the normality and distributional characteristics of thousands of wavelength variables. Taking total iron and total copper for examples, Chapter 4 and Chapter 5 illustrate how to estimate their content using VNIR spectroscopy. Spectral transformations, logarimic and Box-Cox transformations are used. The effects of soil pretreatment are also studies. More specificly, major findings of this topic are listed as follows:(1) This study proposes two novel methods for simultaneously analyzing the normality and distributional characteristics of thousands of wavelength variables:the Lilliefors normality test plot, and the kurtosis and skewness curves plot. Experiments show that in the case study, the normal assumption of wavelength variables is valid for some specific bands, but not all. Spectral transformations and soil sample pretreatment can largely change the distirbutional charactists of wavelength variables.(2) This study is among those pilot researches which explore the feasibility of estimating heavy metal content in floodplain soil with different land use and land cover at a reginal scale using VNIR specscopy. A "fast and low cost" method for estimating heavy metal content in top soil is proposed. For the samples without pretreatment, the coefficients of determination for PLSR model of total iron content and total copper content are 55% and 42%, respectively. For the pretreated soil samples, the coefficients of determination are 66% and 43%, respectively. (3) The effects of spectral transformation, data transformation of heavy metal content, and soil sample pretreatment on the PLSR model are systematically studied. Results show that the SNV spectral transformation can effectively eliminate the spectral differences from soil particle size. The soil pretreatment can improve the PLSR model.(4) The bivariate correlations between VNIR wavelengths (350-2500nm) and total iron/copper content are analyzed. Results show the the Spearman correlation coefficient curves of total iron/copper content are similar. Therefore, the mechanism of estimating total copper content from Le’an floodplain soil lies in its courrelation with total iron content.(5) With the help of correlation coefficient map of spectral wavelength pair, the Spearman correlation between band pairs from region 350-550 nm and region 551-2500 nm decreased after soil samples pre-treatment, indicating that the collaborative changes between these two regions were weakened by the air-dried, grinding and 0.2 mm sieving process.(6) This study introduced the Spearman correlation analysis in the VNIR specscopy study. The advantages of Spearman correlation analysis are also illustrated as follows. The correlation coefficien curve of heavy metal content (e.g. total copper content) and reflectance is of mirror symmetry as that of heavy metal content and absorbance. Further study reveals that these two curves have the same p vector. Moreover, the Spearman correlation curve remains constant when the data of heavy metal content goes through Box-Cox transformation or logarithmic transformation. The above characterists show the potential advantages of Spearman correlation analysis.Topic two:mapping spatial distribution of heavy metal in soil.Effective estimations of heavy metal content in soil can provide foundermental data for human health risk (HHR) assessment, because extremely high value of heavy metal content in soil would threat human health. Traditional heavy metal pollution assessment in soil is non-spatial, which finds its deficts in source identification and HHR assessment. Recently, GIS is used in HHR assessment. Most researches, however, are carried out at local scale with simple maps showing the risks. HHR assessment in China is not a hot topic despite its improtantce, and is developing slowly. Reasons are as follows:(1) Current statistical data is usually non-spatial at a relative larger scale, handicapping the use of current HHR models.(2) Land use map of high quality is not open to the public, whereas a basic rule in HHR assessment is that the risk is related with land use types.(3) The soil environmental standard is out of date. The currently used "Soil environmental quality standard" was made in 1995, considering only eight kinds of heavy metals and is used in the assessment of farmland soil quality. There is no such standard for the soil in residential area, commercial area or industrial area. Moreover, there is no national human health based Generic Assessment Criteria (GAC) for areas with soil contaminations.Therefore, the integrations of multiple sources of data in heavy metal assessment, and the presentions of the results would be a meaningful topic to be studied.In Chapter 6, a land use map showing the agricultural area, built up area and mining area is derived from two ALOS images by visual interpretation. Inverse distance weighted interpolation is used to map the spatial distribution of total copper and lead contents in Le’an floodplain soil. The potential risk of these two metals to human health is finally evaluated with a comprehensive map which is made up of topographical data, land use map, interpolation map of heavy metal contents and other ancillary data.In summary, the presented study makes the following contributions:(1) This study proposes two novel methods for simultaneously analyzing the normality and distributional characteristics of thousands of wavelength variables:the Lilliefors normality test plot, and the kurtosis and skewness curves plot.(2) Explore the feasibility of estimating heavy metal content in floodplain soil with different land use and land cover at a reginal scale using VNIR specscopy. A "fast and low cost" method for estimating heavy metal content in top soil is proposed.(3) Systematically study the effects of spectral transformation, data transformation of heavy metal content, and soil sample pretreatment on the PLSR model.(4) Propose the mechanism of estimating total copper content from Le’an floodplain soil lies in its courrelation with total iron content.(5) Reveal that the collaborative changes between region 350-550 nm and region 551-2500 nm are weakened by the air-dried, grinding and 0.2 mm sieving process.(6) Introduced the Spearman correlation analysis in the VNIR specscopy. The correlation coefficien curve of heavy metal content (e.g. total copper content) and reflectance is of mirror symmetry as that of heavy metal content and absorbance. Further study reveals that these two curves have the same p vector. Moreover, the Spearman correlation curve remains constant when the data of heavy metal content goes through Box-Cox transformation or logarithmic transformation. The above characterists show the potential advantages of Spearman correlation analysis.(7) Propose a new method for assessing the risk of heavy metal pollution to human health. This method integrates a variety of data and technique such as remote sensing and geographical information technique, as well as hyperspectral data, remote sensing images, topography. The proposed method has proved to be highly effective, low cost, no pollutions and easy to applied. It is helpful in soil environmental investigation and risk assessment.