Large parts of Central British Columbia are covered by fluvioglacial sediment which has prompted researchers to develop ways to constrain the geology under cover. One such study has predicted the geology undercover by applying a neural network approach to regional stream and lake sediment data (Barnett and Williams, 2009a).
The report presented here, describes the results of a study aimed at ground-truthing the predicted bedrock geology map of Barnett and Williams (2009). Exposures, largely of previously undocumented bedrock, have been mapped and sampled in 33 locations in the Prince George and Fort Saint James area. These samples have been analyzed for whole rock geochemistry. In addition, stream sediments were collected from six previously unsampled locations on first order streams within 1.2 km from sampled bedrock exposures. The new field exposures were mapped and then compared to the BC Geological Survey bedrock map (Massey et al., 2005).
The results presented herein are only applicable to areas with relatively thin overburden since no areas with deeply covered bedrock could be tested. However, the data show that the neural network was able to predict the bedrock geology well (i.e., most probable predicted bedrock matches the observed unit) in about 75% of the cases. In another 10% of the cases the observed geological unit matches the second or third most probable unit.
The new field exposures correlate well with the predicted geology map, thereby validating the neural network approach for predicting bedrock geology. This is also supported by the fact that, on the basis of empiric comparisons, the local rock geochemistry is reflected in the stream sediment samples collected nearby.
There are, however, important limitations to the use of neural network technology to predict bedrock geology. For instance, the predictions of the distribution of lithologically and geochemically heterogeneous map units, such as the "Cache Creek Complex", are less trustworthy than those for geochemically more defined units like the Chilcotin Group. Similarly, if intrusive rocks are divided according to age rather than composition, the neural network approach will have problems correctly assigning intrusive units to its corresponding age group.
Three of the bedrock samples yielded geochemical results of interest when exploring for magmatic hydrothermal base or precious metal and Ni-PGE deposits. In addition, one stream sediment sample yielded elevated Au and Hg values.
An earlier version of this report was presented at the Kamloops Exploration Group Conference in April 2011.
