1. College of Resource and Environmental Sciences, Nanjing Agricultural University, Nanjing 210095, China 2. Resource and Environmental Science Data Center, Chinese Academy of Sciences, Beijing 100101, China 3. Key Laboratory of Resources Remote Sensing & Digital Agriculture, Ministry of Agriculture, Beijing 100081, China
Abstract:Soil samples in the depth from 0 to 20 cm were scooped from agricultural region beside mines and prepared for determination of As concentration, Fe concentrations and organic matter content. At the same time they were scanned by mobile hyperspectral radiometer for visible and near-infrared spectra. Savitzky-Golay filter was used to smooth noises in spectrum curve because of some low signal-to-noise ratios in some regions of visible and near-infrared light, and all the spectra were resampled with the spectral interval of 10 nm. Before principal component regression and partial least square regression models were constructed for predicting As concentration, Fe concentrations and OM content, several spectral preprocessing techniques like first/second derivative (F/SD), baseline correction (B), standard normalized variate (SNV), multiplicative scatter correction (MSC) and continuum removal (CR) were used for promotion of models’ robustness and predicting performance. For limited samples, cross validation was carried out by repeated leave-one-out procedure, and root mean square error of prediction (RMSEP) was used for validating the prediction ability of constructed models. In this study principal component regression models behave better than partial least square regression models in representing regressing ability, reducing risk of over-fitting with less factors and ensuring models’ accuracy and pertinences (relative RMSEP and R2). Preprocessing techniques of SNV, MSC and CR improve obviously the prediction ability of models for As concentration, Fe concentrations and OM content with relative RMSEP equal to 0.304 0, 0.144 3 and 0.171 2, with number of factors equal to 5, 3 and 3, respectively. The analysis of regression vectors of selected optimal PCR models shows that several important wavelengths are simultaneously taken and helpful for prediction performance: 450, 1 000, 1 400, 1 900, 2 050, 2 200, 2 250, 2 400 and 2 470 nm. Application of the calibrated models to soil contamination of croplands is promising. Concentrations of soil contaminants and contents of other matter can be determined by reflectance spectroscopy with high spectra resolution, which would provide potent reference for remote sensing monitoring of soil and environmental quality.
[1] CHEN Huai-man (陈怀满). Heavy Metal in the Soil-Plant System(土壤—植物系统中的重金属污染). Bejing: Science Press(北京: 科学出版社), 1996. 210. [2] Koppnen S M, Brezonik P L, Olmanson L G, et al. Remote Sensing of Environment, 2002, 82: 38. [3] CHI Guang-yu, LIU Xin-hui, LIU Su-hong, et al(迟光宇,刘新会,刘素红,等). Spectroscopy and Spectral Analysis(光谱学与光谱分析), 2006, 26(7): 1272. [4] Ben-Dor Inbar Y, Chen Y. Remote Sensing of Environment, 1997, 61: 1. [5] Mars J C, Crowley J K. Remote Sensing of Environment, 2003, 84: 422. [6] Kemper T, Sommer S. Environmental Science and Technology, 2002, 36: 2742. [7] Kooistra L, Wehren R, Buydens L M C, et al. Journal of Applied Geophysics, 2001, 3(4): 337. [8] LIAO Xiao-yong, CHEN Tong-bin, XIE Hua, et al. Environment International, 2005, 31: 791. [9] BAO Shi-dan(鲍士旦). Agrochemical Analysis of Soil (土壤农化分析). Beijing: Agriculture Press(北京: 农业出版社), 2000. 12. [10] WANG Ai-ping(王爱平).Chinese Journal of Soil Science(土壤通报), 2004, 35(3): 383. [11] United States Environmental Protection Agency (USEPA). Method 3050B: Acid Digestion of Sediments, Sludges and Soils(Rev.2). Washington: DC, 1996. [12] Clark R N, Roush T L J. Geophys. Res. 1984, 89: 6329. [13] WU Yun-zhao, Chen Jun, Ji Jun-feng, et al. Environmental Science and Technology, 2005, 39(3): 873. [14] JIANG Huan-yu, YING Yi-bin(蒋焕煜, 应义斌). Spectroscopy and Spectral Analysis(光谱学与光谱分析), 2007, 27(3): 499. [15] Kooistra L, Wehrens R, Leuven R S E W, et al. Analytica Chimica Acta, 2001, 446: 97.