Hyperspectral sensing of heavy metals in soil and vegetation: Feasibility and challenges

摘要

Remote sensing of heavy metal contamination of soils has been widely studied. These studies concentrate heavily on the hyperspectral reflectance of typical metals in soils and in plants measured either in situ or in the laboratory. The most used wavebands lie within the visible-near infrared portion of the spectrum, especially the red edge. In comparison, mid- and far-infrared wavelengths are used far less frequently. Hyperspectral data are optimized to suppress noises and enhance the signal of the targeted metals through spectral derivatives and vegetation indexing. It is found that only subtle disparity exists in spectral responses for some metals at a sufficiently high content level. Not all metals have their own unique spectral response. Their detection has to rely on their co-variation with the spectrally responsive metals or organic matter in the soils. The closeness of the correlation dictates the accuracy of prediction. Without any theoretical grounding, this correlation is site-specific. Various analytical methods, including stepwise multi-linear regression, partial least squares regression, and neural networks have been used to model metal content level from the identified spectrally sensitive bands and/or their transformed indices. Both the model and the explanatory variables vary with the metal under detection and the area from which in situ samples are collected. Despite the amply demonstrated feasibility of estimating several metals by a large number of authors, only a few have succeeded in mapping the spatial distribution of metals from HyMAP, HJ-1A and Hyperion images to a satisfactory accuracy using complex algorithms and after taking environmental variables into account. The large number of reported failures testifies the difficulty in the detection of heavy metals in soils and plants, especially when their concentration level is low. The reasons or factors responsible for the success or failure have not been systematically analyzed, including the minimal spectral resolution required. (C) 2017 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS). Published by Elsevier B.V. All rights reserved.