Mapping Radioactivity from Monitoring Data: automating the Classical Geostatistical Approach

In the context of a comparison of spatial prediction algorithms, we applied the classical geostatistical approach to see how well it would automate, and how well it performed in case of an unexpected anomaly. In case of a test without anomaly, the method performed well. In the anomaly case, automatic variogram modelling was hindered seriously, and in terms of RMSE best results were obtained by using the variogram from the test data without the anomaly. Although the 10 days of available training data showed a strong temporally persistent spatial pattern, cokriging did not improve predictions.