Many countries receive shipments of bulk cereals from primary producers. There is a volume of work that is on-going that seeks to arrive at appropriate standards for the quality of the shipments and the means to assess the shipments as they are out-loaded. Of concern are mycotoxin and heavy metal levels, pesticide and herbicide residue levels, and contamination by genetically modified organisms (GMOs). As the ability to quantify these contaminants improves through improved analytical techniques, the sampling methodologies applied to the shipments must also keep pace to ensure that the uncertainties attached to the sampling procedures do not overwhelm the analytical uncertainties. There is a need to understand and quantify sampling uncertainties under varying conditions of contamination. The analysis required is statistical and is challenging as the nature of the distribution of contaminants within a shipment is not well understood; very limited data exist. Limited work has been undertaken to quantify the variability of the contaminant concentrations in the flow of grain coming from a ship and the impact that this has on the variance of sampling. Relatively recent work by Paoletti et al. in 2006 [Paoletti C, Heissenberger A, Mazzara M, Larcher S, Grazioli E, Corbisier P, Hess N, Berben G, Lubeck PS, De Loose M, et al. 2006. Kernel lot distribution assessment (KeLDA): a study on the distribution of GMO in large soybean shipments. Eur Food Res Tech. 224: 129-139] provides some insight into the variation in GMO concentrations in soybeans on cargo out-turn. Paoletti et al. analysed the data using correlogram analysis with the objective of quantifying the sampling uncertainty (variance) that attaches to the final cargo analysis, but this is only one possible means of quantifying sampling uncertainty. It is possible that in many cases the levels of contamination passing the sampler on out-loading are essentially random, negating the value of variographic quantitation of the sampling variance. GMOs and mycotoxins appear to have a highly heterogeneous distribution in a cargo depending on how the ship was loaded (the grain may have come from more than one terminal and set of storage silos) and mycotoxin growth may have occurred in transit. This paper examines a statistical model based on random contamination that can be used to calculate the sampling uncertainty arising from primary sampling of a cargo; it deals with what is thought to be a worst-case scenario. The determination of the sampling variance is treated both analytically and by Monte Carlo simulation. The latter approach provides the entire sampling distribution and not just the sampling variance. The sampling procedure is based on rules provided by the Canadian Grain Commission (CGC) and the levels of contamination considered are those relating to allowable levels of ochratoxin A (OTA) in wheat. The results of the calculations indicate that at a loading rate of 1000 tonnes h(-1), primary sample increment masses of 10.6 kg, a 2000-tonne lot and a primary composite sample mass of 1900 kg, the relative standard deviation (RSD) is about 1.05 (105%) and the distribution of the mycotoxin (MT) level in the primary composite samples is highly skewed. This result applies to a mean MT level of 2 ng g(-1). The rate of false-negative results under these conditions is estimated to be 16.2%. The corresponding contamination is based on initial average concentrations of MT of 4000 ng g(-1) within average spherical volumes of 0. 3 m diameter, which are then diluted by a factor of 2 each time they pass through a handling stage; four stages of handling are assumed. The Monte Carlo calculations allow for variation in the initial volume of the MT-bearing grain, the average concentration and the dilution factor. The Monte Carlo studies seek to show the effect of variation in the sampling frequency while maintaining a primary composite sample mass of 1900 kg. The overall results are presented in terms of operational characteristic curves that relate only to the sampling uncertainties in the primary sampling of the grain. It is concluded that cross-stream sampling is intrinsically unsuited to sampling for mycotoxins and that better sampling methods and equipment are needed to control sampling uncertainties. At the same time, it is shown that some combination of cross-cutting sampling conditions may, for a given shipment mass and MT content, yield acceptable sampling performance.
