Assessing statistical and spatial validity of sediment survey design and sampling densities: examples from Lake Erie

Spatial interpolation methods translate sediment contamination point data into informative area-based visualizations. Lake Erie was first sampled in 1971 based on a survey grid of 263 locations. Due to procedural costs, the 2014 survey was reduced to 34 sampling locations mostly located in deep offshore regions of the lake. Using the 1971 dataset, this study identifies the minimum sampling density at which statistically valid, and spatially accurate predictions can be made using ordinary kriging. Randomly down-sampled subsets at 10% intervals of the 1971 survey were created to include at least one set of data points with a smaller sample size than that of the 2014 dataset. Regression analyses of predicted contamination values assessed spatial autocorrelation between kriged surfaces created from the down-sampled subsets and the original dataset. Subsets at 10% and 20% of the original data density accurately predicted 51% and 75% (respectively) of the original dataset's predictions. Subsets representing 70%, 80% and 90% of the original data density accurately predicted 88%, 90% and 97% of the original dataset's predictions. Although all subsets proved to be statistically valid, sampling densities below 0.002 locations/km(2) are likely to create very generalized contamination maps from which environmental decisions might not be justified.