CRITICAL CONSIDERATIONS WHEN USING COMPILED GEOCHEMICAL DATASETS

Compiled geochemical datasets are increasingly available. Some, such as the Alaska Geochemical Database (AGDB) stream sediment dataset, include analyses of samples collected by various programs over many decades and contain concentration data determined by multiple analytical methods for individual elements. Use of geochemical datasets that include results from different analytical methods requires careful data selection and statistical handing due to differing method sensitivities and possible bias between methods. Reanalysis of archived samples by a standardized multielement package, such as efforts by the Earth Mapping Resources Initiative (EMRI) program, generates an internally-consistent dataset but the dataset may be spatially limited compared to the original. Consequently, element mapping of large regions may still require combining reanalyzed with historic data. Using Au data for the Taylor Mountains quadrangle (TMq) from the AGDB and a combination of reanalyzed and historic As data from the Yukon-Tanana uplands (YTu), we highlight considerations for selecting, processing, and mapping geochemical data from mixed-method datasets. For the Taylor Mountains historic data with high detection limits (0.02 to 0.25 ppm Au) were excluded from the analysis. A map of the gold distribution in the TMq was generated from 771 samples. Au data was determined by one of two analytic methods. Quality control (QC) data were available for both methods and were used to estimate measurement error for the methods. The samples were uniformly distributed in the TMq but analysis by the two methods clustered spatially. Mapped highs in the Au distribution correspond with known Au mineralization location, high Au concentrations in rock samples, or visible Au grains in heavy minerals samples. Our interpretation is the mapping method accurately represents the Au distribution in TMq. The As data in the YTu is more complex. Twelve methods were used to determine As in the YTu. The historic data predate the preservation of QC data. Using paired data from samples that had As measurements by more than one method we identified some 1) bias between methods and 2) outliers to general trends. To minimize effects of bias and the potential outlaying data, we are developing regression equations for the median to level the data prior to mapping.