A sensitivity analysis of search distance and number of samples in indicator kriging

"Selection of the appropriate kriging plan during indicator kriging estimations is of a paramount importance to the precision of ore reserve estimation. However, very little is known on the sensitivy of indicator kriging results when changing the kriging plan parameters (mainly search distance and number of samples). This paper evaluates the effect of the search distance and number of samples on indicator kriging results using a gold deposit as a case study.IntroductionBlock estimation using kriging is based on the autocorrelation function (variogram) on one hand and on the kriging plan on the other. As far as the variogram is concerned it is more or less subjectively adjusted and this is reflected on its parameters (range, sill, nugget effect and anisotropy). The effect of a poor estimation of these parameters on the kriging variance of ordinary kriging has been proven very important (Brooker, 1986). Choosing a kriging plan is generally a difficult matter depending on the estimation type (global or local). Generally there is not a unique kriging plan be cause there is not a unique structural analysis (variogram). There are no rules or dogmas in choosing a kriging plan but it is rather a function of the sampling pattern (David, 1982).Structural analysis requires the hypothesis of quasistationarity but very often either the data available within the zone of quasi-stationarity are too many so that the search distance is reduced during kriging or the data are too few to satisfy a stationary kriging system (Journel and Huijbregts 1978). As a rule, the larger the neighbourhood, the more precise the estimates should be , but for large neighbourhoods the hypothesis of stationarity does not always hold (Rivoirard 1987; Armstrong and Champigny 1989)."