Research On The Synergistic Application Of X-ray Fluorescence Spectroscopy And Hyperspectral Technology In The Precise Investigation Of Soil Heavy Metal Pollutio

X-ray fluorescence spectroscopy is an efficient,non-destructive,economical and green technique for the rapid detection of heavy metals in soil,however,the accuracy of its results is easily affected by factors such as soil organic matter,moisture and mineral fraction.While the hyperspectral technology has the advantages of high spectral resolution and complete soil spectral information,and can invert soil properties such as soil organic matter,moisture,mineral fraction and soil heavy metal content.The two techniques are highly complementary in the investigation of heavy metals in soils,yet most of the synergistic applications are still used in a simple superposition.In this study,the accuracy and precision of the determination of standard substances were evaluated to compare and select a suitable laboratory method for the analysis of cadmium and arsenic in soils of a sewage irrigation area in the northwest.The main factors affecting the accuracy of soil cadmium and arsenic determination by high-precision portable X-ray fluorescence spectrometer（HDXRF）and their quantitative relationship were analyzed,so as to construct a quantitative correction model of HDXRF soil cadmium and arsenic based on single factor.The partial least squares regression（PLSR）,support vector regression（SVR）and random forest（RF）algorithms were used to construct quantitative inversion models for soil Cd and As based on characteristic bands and full bands,respectively,and to evaluate the prediction accuracy and stability of each model.The RF algorithm was used to construct a collaborative fusion model of HDXRF rapid detection data and hyperspectral data,and the prediction accuracy and stability of the model were evaluated.The results show that:（1）The best laboratory analytical method for heavy metals Cd and As in agricultural soils in the study area is a graphite digestion system of nitric acid-hydrochloric acid-perchloric acid-hydrofluoric acid,and the best testing method is inductively coupled plasma-mass spectrometry（ICP-MS）combined with inductively coupled plasma emission spectrometry（ICP-OES）.（2）The measured HDXRF values had a good linear relationship with the actual laboratory data（R²=0.929）,however,the accuracy of the statistical calibration model based on the Cd and As contamination in agricultural soils could not meet the needs of accurate investigation,especially in the low contaminated areas(R_cd²=0.810,R_AS²=0.692).decay relationship,HDXRF values showed a significant negative correlation with organic matter content,and the single-factor calibration models based on moisture and organic matter content all had good calibration accuracy（R² of 0.925～0.985）.（3）Soil spectral inversion models can predict soil Cd and As contents better,and the prediction accuracy of the optimal model reaches 0.762～0.874 for R² and 2.02～2.43 for RPD.The random forest（Full-SGFD-RF）constructed based on SGFD full band has the best prediction of soil Cd contents,and the support vector regression model constructed based on SG full band（Full-SG-SVR）predicted soil As content better.（4）The RF model based on the fusion of HDXRF fast measurement data and hyperspectral data can significantly improve the prediction accuracy of soil Cd and As,with the R² of the optimal fusion model reaching above 0.920 and the RPD reaching above 3.53,which basically meet the requirements of soil Cd and As The accuracy and reliability of the soil Cd and As pollution survey were basically met.