We aim to develop a method for face-forward prediction of naturally-occurring heavy metals concentrations using geostatistics, by selecting elution amount of arsenic (As) and a tunnel excavated in the Shimanto Group as a case study. The As elution amount data are revealed to be spatially correlated and the correlation structure is well fitted to the spherical model with small nugget effect, which enables to clarify general trend of the As elution amount over the tunnel interval by ordinary kriging (OK). OK’s accuracy of face-forward prediction is relatively high within 3 m ahead of face, but the accuracy decreases largely where the As elution amount changes abruptly, even though applying sequential Gaussian simulation that can reduce the smoothing effect of OK. To increase the prediction accuracy, effectiveness of normalized drilling velocity ratio (NDVR) is found as a supplementary factor, because NDVR increases in shale parts as the As elution amount does. In addition, a SEM-EDS observation of shale and sandstone samples clarifies that As is generally contained in pyrite common to the samples, but pyrite in shale has Framboidal texture, i.e. aggregation of small spherical particles. Amount and texture of pyrite may be additional significant geologic factors on the prediction accuracy.

Mineral spectra of rocks and minerals have been widely measured and published through the internet, for example, by U.S. Geological Survey (USGS) and Geological Survey of Japan (GSJ). Among them, transition metal oxide minerals such as hematite and goethite and hydro-oxide minerals such as clays and sulfates are good targets to identify for geologic mapping through the spectral analysis of satellite images. The identified results can be analyzed by QGIS processing with comprehensive expressions as GIS layers.