Forecasting the number of soil samples required to reduce remediation cost uncertainty.

Sampling scheme design is an important step in the management of polluted sites. It largely controls the accuracy of remediation cost estimates. In practice, however, sampling is seldom designed to comply with a given level of remediation cost uncertainty. In this paper, we present a new technique that allows one to estimate of the number of samples that should be taken at a given stage of investigation to reach a forecasted level of accuracy. The uncertainty is expressed both in terms of volume of polluted soil and overall cost of remediation. This technique provides a flexible tool for decision makers to define the amount of investigation worth conducting from an environmental and financial perspective. The technique is based on nonlinear geostatistics (conditional simulations) to estimate the volume of soil that requires remediation and excavation and on a function allowing estimation of the total cost of remediation (including investigations). The geostatistical estimation accounts for support effect, information effect, and sampling errors. The cost calculation includes mainly investigation, excavation, remediation, and transportation. The application of the technique on a former smelting work site (lead pollution) demonstrates how the tool can be used. In this example, the forecasted volumetric uncertainty decreases rapidly for a relatively small number of samples (20-50) and then reaches a plateau (after 100 samples). The uncertainty related to the total remediation cost decreases while the expected total cost increases. Based on these forecasts, we show how a risk-prone decision maker would probably decide to take 50 additional samples while a risk-averse decision maker would take 100 samples.