Geostatistical techniques in the improvement of spatial modeling of contaminated aquatic sediments: Evaluation and applications
Description
This research evaluated the applications of variogram based geostatistical techniques in remediation of aquatic contaminated sediments. Five issues related to spatial interpolation and mapping of aquatic sediment contaminant concentrations were focused: Was spatial continuity of a concentration contaminant specific and water body specific? Were the geostatistical models as good as the deterministic spatial models? Could secondary data improve predictions and reduce uncertainty? How did sampling size influence? What was the influence of extreme high concentration measurements? Sediment data from two river sites, two ocean sites and a lake site were used in the study. Metal concentrations from four sites and PCB concentrations from one site were spatially modeled. Two types of geostatistical approaches, univariate kriging and cokriging, and four types of deterministic interpolation techniques, inverse distance weighted, global and local polynomials, and radial basis functions were compared for different data sizes. In geostatistical approaches, three theoretical variograms were fitted: spherical, experimental and Gaussian with nugget effect. Three types of krigings were studied: simple kriging, ordinary kriging and kriging with the trend Altogether 500 spatial interpolation models were fitted. Number of factors such as water body shape, contaminant type, data density and sample size, data spread pattern in the region, and extreme concentrations were found to influence the spatial modeling of sediment contaminant concentrations. In general, geostatistical models were found better than deterministic models. Cokriging a contaminant concentration with other contaminant concentrations, soil particle percentages, water depth and previous measurements of the same attributes in the region improved the predictions. Studying extreme high concentrations showed how hotspot prediction could be improved by modifying the modeling approach, depending on the pattern and locations of the extreme concentrations. As the influence of the high concentrations was significant in the models developed with small sizes of data, a thorough comparison of root mean square error with and without the high concentrations was recommended during the model selection Future studies should address the issues such as incorporating soft data in models, cost factor in modeling and incorporating geostatistical models with mechanistic contaminant sediment transport models developed on physical, chemical and biological properties