Utvärdering av strategier för provtagning av förorenad mark

Sammanfattning: By using statistical analysis of measured data more information can be acquired about a contaminated site in comparison to using the measured data alone. This study shows that the choice of model for the distribution of the contamination is vital in getting a reliable estimation of the contaminated area. By simulating exponentially decreasing contaminant distributions around a point source and subsequent sampling using four different patterns of sampling sites, of which two are randomized patterns (random and semiregular random) and two are regular patterns (square and herringbone, we can construct a log-linear model of the point source contamination. The log-linear model is then used to evaluate where the ground is contaminated by applying a decision rule based on an acceptable level of contamination. The regular patterns give a more consistent result, but for small sample sizes unpredictable effects can be seen. The herringbone approach is more likely to include a larger contaminated area and would in most cases be recommended based on potential environmental and health hazards of leaving the contamination in the ground. No conclusions can be drawn about the randomized approaches. A sensitivity analysis of what factors influence how well the constructed contamination model approximates the actual contaminant distribution show that the size of the contaminated area is the most influential factor on whether the contamination is detected or not. The log-linear model gives equal weight to all samples and given the large number of zeroes in the data the results tend to underestimate the contaminated area and especially the concentration close to the point source. A so-called "hurdle" model, that separates the data according to whether they are zero or not and then creates a contamination model from the non-zero data set, is suggested to improve the reliability of the models.