Experience with the replication of regional survey of soil pollution

Protecting the soil from pollution requires reliable estimates of concentrations of contaminants and, in particular, of heavy metals. Laboratory procedures have been standardized to achieve this, but field sampling and statistical procedures for regional mapping have not. In particular, we found no published experience about the precision with which estimates can be reproduced from different surveys of a region. We have compared results from two independent surveys in which the topsoil of a 10-km(2) region of the Swiss plateau was sampled for potentially toxic metals. The concentrations of copper and zinc were measured in hot 2 M HNO3 extracts in different laboratories and afterwards corrected to a common standard by linear regression. The corrected data had similar means, variances and skewness. Experimental variograms were computed and fitted by bounded models with ranges between 1400 and 2000 m. Then, the concentrations were estimated by block kriging to produce maps. The general structures revealed by the two sets of data were similar, but local details differed substantially. To test the reproducibility of local predictions the values at the locations of the second survey were estimated by ordinary kriging from the first survey and compared with the measured values. An 'uncertainty' interval of twice the kriging error was constructed so that, theoretically, 95% of the measured values should lie within it. For copper, only 85% were found in the interval, and for zinc 92% lay within it. In both cases the actual variation of the underlying process was underestimated by the kriging variances. This seems to result from underestimates of the variances in the region. (C) 1998 Elsevier Science Ltd. All rights reserved.