Volume XLII-3
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-3, 335-340, 2018
https://doi.org/10.5194/isprs-archives-XLII-3-335-2018
© Author(s) 2018. This work is distributed under
the Creative Commons Attribution 4.0 License.
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-3, 335-340, 2018
https://doi.org/10.5194/isprs-archives-XLII-3-335-2018
© Author(s) 2018. This work is distributed under
the Creative Commons Attribution 4.0 License.

  30 Apr 2018

30 Apr 2018

RETRIEVAL AND MAPPING OF HEAVY METAL CONCENTRATION IN SOIL USING TIME SERIES LANDSAT 8 IMAGERY

Y. Fang1, L. Xu1, J. Peng1, H. Wang2, A. Wong3, and D. A. Clausi3 Y. Fang et al.
  • 1Dept. of Land science and technology, China University of Geosciences, Xueyuan Road, Beijing, China
  • 2China Agricultural University, Beijing, China
  • 3Dept. of System Design Engineering, University of Waterloo, Canada

Keywords: Soil heavy metal, Time series, Remote sensing imagery, Landsat 8, Retrieval, Model selection

Abstract. Heavy metal pollution is a critical global environmental problem which has always been a concern. Traditional approach to obtain heavy metal concentration relying on field sampling and lab testing is expensive and time consuming. Although many related studies use spectrometers data to build relational model between heavy metal concentration and spectra information, and then use the model to perform prediction using the hyperspectral imagery, this manner can hardly quickly and accurately map soil metal concentration of an area due to the discrepancies between spectrometers data and remote sensing imagery. Taking the advantage of easy accessibility of Landsat 8 data, this study utilizes Landsat 8 imagery to retrieve soil Cu concentration and mapping its distribution in the study area. To enlarge the spectral information for more accurate retrieval and mapping, 11 single date Landsat 8 imagery from 2013–2017 are selected to form a time series imagery. Three regression methods, partial least square regression (PLSR), artificial neural network (ANN) and support vector regression (SVR) are used to model construction. By comparing these models unbiasedly, the best model are selected to mapping Cu concentration distribution. The produced distribution map shows a good spatial autocorrelation and consistency with the mining area locations.