QUEST Project: Geoscience BC Analysis of Geochemical Data

Key Researcher(s):  C. T. Barnett

Project ID:  2008-003

Key Research Organization(s):  BW Mining

Project Location:  Central interior

Focus Area:  Minerals

Summary



Geoscience BC's QUEST project was designed to stimulate mineral exploration in the Quesnellia Terrane of British Columbia. During 2007, about 2100 new lake and stream sediment samples were collected, and almost 5000 older drainage sediment pulps were re-analyzed, to improve the geochemical data base in the project area. BW Mining QUEST Geochemical Analysis ProjectOne of the programs initiated by QUEST in 2008 was to discover what might be learned from a systematic analysis and evaluation of the new multi-element geochemical data. This project is one such study.

Since the samples were collected from different media and analysed by different laboratories over nearly 30 years, it was first necessary to assemble the "best picks" from each sub-population, and then to relevel the various surveys, for each element, to provide uniform blends. Estimates were also made of missing data over some small areas. As a result, syntheses of the various surveys are now available on uniform grids, over a common area, for as many as 42 of the elements. These can form the basis for systematic analysis.

As a start, we have applied various clustering methods to the 42 element data. The results show marked correlations with geology. This leads to the idea of using a neural network to model the geochemistry in areas where the geology is known, and then to apply this model to infer the bedrock geology in the non-outcropping areas. The resulting inferred geology, wherever geochemistry is known, is then almost identical to mapped geology in areas of outcrop, and blends well with mapped geology along the margins, where there is no geochemistry. These results show that geochemistry combined with neural networks can provide a powerful tool for mapping bedrock geology concealed by a veneer of glacial overburden in the QUEST project area.

Since the samples were collected from different media and analysed by different laboratories over nearly 30 years, it was first necessary to assemble the "best picks" from each sub-population, and then to relevel the various surveys, for each element, to provide uniform blends. Estimates were also made of missing data over some small areas. As a result, syntheses of the various surveys are now available on uniform grids, over a common area, for as many as 42 of the elements. These can form the basis for systematic analysis.

As a start, we have applied various clustering methods to the 42 element data. The results show marked correlations with geology. This leads to the idea of using a neural network to model the geochemistry in areas where the geology is known, and then to apply this model to infer the bedrock geology in the non-outcropping areas. The resulting inferred geology, wherever geochemistry is known, is then almost identical to mapped geology in areas of outcrop, and blends well with mapped geology along the margins, where there is no geochemistry. These results show that geochemistry combined with neural networks can provide a powerful tool for mapping bedrock geology concealed by a veneer of glacial overburden in the QUEST project area.

Deliverables