CFSG6eng
OBJECTIVES: to provide estimates when linear geostatistics reached its limits: very asymmetric distributions (including change of support models to go from the scale of the sample to that of the block), estimates of nonlinear functions of the variable of interest (for example, the average content above a threshold), etc.
PROGRAMME
For example, we seek to estimate the quantities exceeding thresholds from data and a model, or even to improve estimation and calculation of variance estimation in the case of variables a very asymmetrical distribution. Two main themes will be discussed:
 the support effect (the samples size is different from that of the selection units);
 the information effect (selection from the estimator instead of the true values);
The following methods will be presented:
 indicator kriging (estimates the indicator above a cutoff);
 multiple indicator kriging (estimate different classes of indicator corresponding to different cutoffs);
 disjunctive kriging (cokriging of indicators, taking into account the relations between the different indicator variables);
 conditional distribution (after transformation of the variable into a Gaussian variable).
These methods provide for instance the probability that the grade exceeds a given cutoff on the sample support. However the support of interest (block, or selectivity mining unit, for instance) is generally much larger than the quasipoint sample support. This results in a lower selectivity, the distribution of block values presenting for instance less low or high values than the distribution of point support values.
Change of support models will be studied, that allow predicting the global distribution of block values within the whole domain, from the known distribution of sample values. Techniques will also be seen, that take into account change of support locally, using the previously mentioned estimators of indicators, or other specific estimators (service variables, uniform conditioning).
Prerequisite: Proficiency in basic geostatistics is required (see Linear Geostatistics training).

an application for registration with a CV 
Training location : Mines ParisTech, 35 rue Saint Honoré, 77300 Fontainebleau  France
modules list previous module next module