REDUCING VARIANCE IN SOIL ORGANIC-CARBON ESTIMATES - SOIL CLASSIFICATION AND GEOSTATISTICAL APPROACHES

Concern for groundwater purity is increasing. Computer modeling of pro cesses that can contribute to groundwater contamination is growing, ye t the supporting data base is meager and the available data were usual ly not collected for such uses. Often one is left to estimate soil pro perties from data on analogous soils. Soil organic carbon, a major inp ut in pesticide leaching assessments, can be extrapolated to unsampled sites by two approaches: by analogous reasoning using Soil Taxonomy a nd by regionalized varible methods (geostatistics). The 23 x 32 km Pea rl Harbor recharge area was used to compare these two methods of estim ating soil organic carbon in unsampled areas. Soil organic carbon vari ed from 0.013 to about 0.060 g C/g soil. The general variance for 53 s amples in the area was 1.04 X 10(-4) and, according to gestatistical e stimates from semivariograms, could be reduced to 1.08 X 10(-5) at a s ampling distance of 1.2 km. Based on a transect sampling study within the area, increases in sampling frequency from 4600 to 7.36 million sa mples would not further reduce estimated variance and thus would not b e advisable. Extrapolation on the basis of Soil Taxonomy, i.e. to simi lar soil series, and geostatistical extrapolation provided similar est imates of uncertainty in soil organic carbon in unsampled areas. Data from the Pearl Harbor recharge area suggest that extrapolation by Soil Taxonomy may underestimate regional varibility due to sampling assump tions. Extrapolation by geostatistics ignores some important qualitati ve information. A method of combining these two approaches is needed. Our results emphasize the effect of the purpose and assumptions of a s ampling plan on the data obtained. Large scale modeling of soil proper ties of a region, typical of many geographical information systems (GI S) analyses, requires a representative sample of the variation in the region. A three-phase sampling plan that includes a qualitative phase as well as a quantitative phase may ensure that the data obtained are appropriate.