COMPARISON OF SPATIAL INTERPOLATION MODELS CHARACTERIZING SURFICIAL DIOXIN DISTRIBUTIONS IN MIDLAND, MI

Dioxin contamination along the Tittabawassee River flood plain has been a major issue in Midland, Michigan, during the past few years. The Michigan DEQ performed river and soil sediment tests along the Tittabawassee River in 2001, and performed sediment tests at residential locations in 2003 along the flood plain. The Michigan DEQ published this data on their web site (http://www.michigan.gov/deq). In their reports summarizing dioxin distributions, no geostatistical studies were published to model the spatial variability of dioxin from soil characterization.

This study examines the data collected by the Michigan DEQ in Midland, Michigan. The study compares various geostatistical algorithms (ordinary kriging, indicator kriging, multi-Gaussian kriging and log-normal kriging) performed on the published data, using both Euclidean and non-Euclidean distance metrics. The non-Euclidean distance metrics were based on published 100-year flood plain maps, since the suspected primary mode of transport of dioxin occurred via surface runoff during floods, not through groundwater transport. This study provides the first known spatial model of dioxin in Midland, Michigan, and shows some advantages of using a priori knowledge of transport processes in modeling dioxin distributions. Saito and Goovaerts published a similar study in 2000, examining dioxin at EPA’s Piazza Road Superfund site in Missouri. This study also provides some validation of their work, at a different site.

An overview of various geostatistical algorithms will be discussed in this presentation, as well as an overview of Euclidean versus non-Euclidean distance metrics applied to geostatistics. Results of the various geostatistical comparisons will then be presented for this particular data set.