Sequential Updating Simulation

Abstract

This paper presents a new implementation of the sequential simulation principle, within a multi-Gaussian framework. In this approach, the local conditional distribution functions, from which simulated values are drawn by Monte-Carlo, are updated iteratively rather than re-estimated at each step. This new implementation offers several significant advantages: the local distribution functions, from which simulated values are drawn, are conditional to all hard and previously simulated data, rather than to data within a search neighbourhood only; there is no need to assign existing hard data to the nearest grid nodes; the local means and variances are estimated from the available data at their exact locations; and the updating process does not involve any longer the solving of a linear system of equations. This, in turns, relaxes the constrains on the spatial correlation models which can be used. This new approach is illustrated by a case study in soil contamination.