Some statistical relationships between stream sediment and soil geochemistry in northwestern Wisconsin—can stream sediment compositions be used to predict compositions of soils in glaciated terranes?

Abstract

Mean stream sediment chemical compositions from northwestern Wisconsin in the north central United States, based on more than 800 samples, differ significantly from mean A-horizon and C-horizon soil compositions, based on about 380 samples of each horizon. Differences by a factor greater than 1.5 exist for some elements (Ca, Mn, Mg, P, Ti, Ni, Pb, Se, Zn). A very large database of stream sediment geochemistry exists for the region (more than 2200 samples) and for the U.S. (roughly 400,000 samples), whereas data on the chemistry of soils is much sparser both regionally and nationally. Therefore, we have attempted to quantify trends in compositional differences between stream sediments and nearby soils to test whether the abundant stream sediment data can be used to predict soil compositions. A simple computational technique of adjusting the stream sediment compositions according to the ratio of means of soils and stream sediments was conducted. A variety of techniques of correction and interpolation of data were tested and indicate that repetitive testing of results allows an optimum correction to be achieved. Predicted soil compositions compared to analytically determined soil compositions show a range of results from relatively good correspondence for some elements to rather poor correspondence for others. In general, predictions are best at midranges of compositions. The technique does not predict well more extreme or anomalous values. Thus, this technique appears to be useful for estimating background soil compositions and delineating regional compositional trends in soils in situations where large amounts of stream sediment analyses and smaller amounts of soil analyses are available. The technique also provides probabilistic qualifications on the expected error between predicted and actual soil compositions so that individual users can judge if the technique provides data of sufficient accuracy for specific needs.