Estimating and simulating the degree of serpentinization of peridotites using hyperspectral remotely sensed imagery

Abstract

The mineral products resulting from the process of serpentinization, by which primary magnesium silicate minerals in peridotites are replaced by hydrous serpentine-group minerals, are of economic importance since Alpine-type peridotites are the host rocks for virtually all large asbestos deposits, which may be attributed mainly to the serpentine-group mineral chrysotile. Conventional field mapping of the distribution of highly serpentinized areas is time consuming and requires detailed sampling and laboratory analysis. In 0.4- to 2.5-μm reflectance spectra of serpentinized peridotites, serpentinization is responsible for a decrease in contrast of olivine-pyroxene iron absorption features and an appearance and increase in OH− absorption features near 1.4 μm and 2.3 μm associated with serpentine minerals. The degree of serpentinization is positively correlated with the depth of the 1.4-μm and 2.3-μm absorption features for samples containing more than 55 weight percent serpentine minerals. Small amounts of magnetite in a sample obscure the spectral contrast and decrease the overall brightness of weakly serpentinized samples. A methodology is used for mapping serpentine minerals in ultrabasic rocks from imaging spectrometer data, which includes (1) vegetation masking, (2) calculating the absorption band depth of the 2.3-μm absorption feature in unmasked pixels, (3) translating this value into percent serpentine minerals using an empirical linear model, and (4) estimating the degree of serpentinization at the remaining locations using conditional simulation techniques or ordinary block kriging. From the results of this study, it can be concluded that mapping the degree of serpentinization from high-spectral resolution imagery is possible within marginal statistical fluctuations. Conditional simulation reproduces the spatial and statistical variability of the data set; however, it sacrifices the local accuracy. Direct estimation using ordinary kriging provides a better local estimate but does not honor the statistics and spatial dispersion of the original data.