| The estimation of plant’s heavy metal content can reflect the impending surroundings of surface, which lays theoretical foundation for using covered vegetation to monitor environment and investigate resource. The floristic remote sensing of earth chemical is a discipline which based on remote sensing technology to investigate the distribution, content and migration of chemical element. It could replace the traditional mineral investigation method in the vegetation-covered areas and heavy metal pollution monitoring in the soil gradually. At present, after describing spectra of vegetation, empirical model that employs the statistical methods and vegetation index are built up using remote sensing, which has disadvantage of low stability and universality and fail to expand the application in other areas. With the development of hyperspectral remote sensing and its data processing method, it is urgent to develop the inversion of heavy metal content using hyper-spectra remote sensing, especially the blade radiative transfer model based on physical process of inversion method, which could provide the technical support for mineral resource investigation and pollution monitoring of heavy metal under the vegetation.In order to collect the vegetation physiological and biochemical parameters as well as a complete set of spectral data set where plants are under heavy metal stress, has carried out an experiment of indoor plants’ cultivation under copper and nickel stress. The experiment chooses monocotyledonous plant wheat and dicotyledons plant oilseed rape, adding copper sulfate and nickel sulfate solution to the solid. The experiment also sets 6 concentration gradients under single copper, nickel and copper nickel compound stress and obligates five growth period of leaf spectral reflectance under copper nickel stress, the chlorophyll, moisture, dry matter, copper ion and nickel ion content data set. By observing the leaves apparent physiological state, wheat and canola plants under copper and nickel stress grows slowly, presenting symptoms of chlorotic and water shortage as well as the changes in leaves biochemical component content, but there is no significant correlation between them. Using spectral Angle to evaluate the spectral variation degree of six characteristic bands, except the first data at the beginning of the growth, the other four growing 79% stress levels in six bands is variable, which manifests that the plant spectral variation under heavy mental stress is represented in all band range, rather than just within a certain band. The correlation between position features of plant spectra under heavy metal as well as vegetation index and heavy metal content are also analyzed herein, and results show that some spectra feature parameters correlates with copper-nickel ion significantly, and the correlation of normalized chlorophyll index(NPCI) and normalized moisture index(NDWI) is best among these vegetation index.To improve the classic PRPSEPCT model for the leaf, m PROSPECT model are established. Comparing the parameters to drive the model, blade structure parameters N is consider as most uncertain parameters, which closely relates to heavy metal. Taken copper and nickel ion as example, the heavy metal ion combined with the blade material and stay in the cell, which result in obvious absorption in visible band and near-infrared band. Therefore, fitted N value and absorption coefficient of heavy metal are employed to describe the effect of heavy metal on absorption and scattering, and then the model is built up. The m PROSPECT model is tested by field measurement, and results show that average N values of wheat and oilseed rape under heavy metal is higher than that of healthy leaf, as much as 3.1.The absorption coefficients of copper and nickel agree with the field measurement. The root mean square error and correlation coefficients between the spectra simulated by m PROSPECT model and filed spectra are better than PROSPECT model.The regresion model with single variable is built up using spectra feature parameters to estimate the heavy metal content. Stepwise multiple regression(SMLR) and Partial least squares(PLSR) in all bands are established and compared. The accuracy of multi variables is better than that of single variable, but PLSR method behave poorly for unknown samples. In summary, the empirical model has rather bad robustness, for example, the SMLR model to inverse the copper ion content of oilseed grape that behave best have as high square root error as 0.113. Whereas the empirical model relies on the input database and has rather low accuracy, develop the method of retrieval the content of heavy metal. The multi parameters are calculated through all the bands. Because the specific absorption coefficients overlap, the inversion result has greater errors. Based on this, Sub-band Multilevel and Target-Updates(SMTU) are proposed for separating different bands and multi levels. On the basis of the uncertainty of model parameters is set up for three different databases. The sensitivity of parameters in different bands is analyzed by sensitivity matrix and uncertainty, and the bands that sensitive to heavy metal and insensitive to other parameters is chosen as best bands. And the effect of other parameters on heavy metal ion is analyzed, and sensitive parameters are determined. The error function of future knowledge is established, the sensitive parameters are updated as well as uncertainty of heavy metal ion to improve the retrieval accuracy. Compared with the inversion results obtained form empirical model and multi-parameters in all bands, the square root error of copper ion under copper condition and copper-nickel condition decrease by 0.0978 and 0.1094 respectively, and those of nickel on decrease by 0.0757 and 0.091.The m PROSAIL model is established for heavy metal inversion after the m PROSPECT model combines SAILH model. Compared with PROSAIL, m PROSAIL has smaller sqaure root error, close to file measurment. The sensitivity of model parameters to simulated canopy spectra under different copper and nickel content are assessed using extended Fourier amplitude sensitivity testing method. The content of copper and nickel ion is set up as maximum step, the main sensitivity parameter is set up as a larger step size, and time sensitivity parameter setting for the minimum step, and finally the vegetation canopy response spectra is established. The canopy spectra collected indoor validates the spectra, and results show that the correlation coefficients between the copper and nickel content from look-up table and field measurement is 0.049 and 0.038 respectively, which means that the look-up table database could be used for inversing the unknown area and unknown database. In addition, the relationship of canopy spectra with multi angles are found, and used for create canopy spectra with multi angles considering the structure parameters, angles and other parameters, the blade spectra is simulated using SAILH model. Also, the inversion method for SMTU is discussed. The result is validated by the indoor spectra and the square root error is as small as 0.080- 0.095. The square root error between copper-nickel ion between field measurement is 0.041 and 0.035, better than that of lookup method.Taken Huma in Heilongjiang Province as example, the copper and nickel content is detected applying lookup table and SMTU method to Hyperion image. The vegetation-pixel calculated by linear unmixing model are extracted, and used as target pixel. The error function constructed from correlation coefficient of lookup table, the content of nickel and copper could be found in the lookup table. Taken chlorophyll and water content as future knowledge from look-up table, the content of nickel and copper of the leaf is also inversed by method of SMTU function. The two methods are compared with field measurements, the Hyperion image sensitive to copper ion has lower precision, therefore copper ion content from lookup table are better than that of SMTU method while the nickel content is the opposite situation, which show that SMTU method has good accuracy for satellite image. |
PDF Full Text Request