You are here:
Optimization of Sampling Design to Determine the Spatial Distributions of Emerging Contaminants in Estuaries.
Citation:
Katz, D., A. Kuhn, J. Sullivan, AND M. Cantwell. Optimization of Sampling Design to Determine the Spatial Distributions of Emerging Contaminants in Estuaries. Society of North America (SETAC) 39th Annual Meeting, Sacramento, CA, November 04 - 08, 2018.
Impact/Purpose:
Narragansett Bay waters have been intensively sampled over the past 50 years, yet no widely accepted sampling design for spatial distribution of dissolved contaminants exists. Sampling locations have often been chosen based on local knowledge alone, making comparisons among studies difficult. Such comparison would be easier if a single approach to statistically-based, probabilistic sampling designs was adopted. In an effort to harmonize future sampling efforts to assess spatial distributions of contaminants in Narragansett Bay and similar environments, a number of commonly implemented spatial models and sampling densities were evaluated. After evaluation, a randomized statistically based sampling design was developed and implemented. Results indicated that spatial models based on Empirical Bayesian Kriging were most accurate and a minimum number of samples per km2 was determined. Future comparisons of datasets both within and external to the studied estuary and would be facilitated by adoption of a sampling design similar to that proposed in this work.
Description:
Narragansett Bay (NB) is New England’s largest estuary and has been the location of numerous multi-disciplinary scientific studies over the last several decades. Most spatial research conducted within NB to date has employed deterministic sampling designs. Several studies have used probabilistic sampling designs on portions of NB; however, a randomized, probabilistic sampling design has seldom been applied to the whole estuary. In this study, a sampling design was developed to implement a probabilistic sampling approach to examine the distribution of emerging contaminants in the estuary. Existing high-density salinity data from the upper portion of NB was used for optimization of the sampling design and results from this analysis were applied to the whole estuary. Five spatial geostatistical interpolation kriging models were compared and the empirical bayesian kriging (EBK) model performed best across the estuary. The results from this analysis were used to determine an optimum number of stations (n=67) and sampling density (0.2 stations/km2) across the estuary. Stations were chosen randomly within a randomized, tessellated hexagonal grid and samples were taken to measure the concentrations of a suite of pharmaceutical compounds. Spatial models using EBK were generated at varying spatial densities from the full number of sampled stations to ¼ the number of stations. Error statistics for each spatial density were generated using cross-validation between predictions and measurements. These statistics were compared using the Derringer desirability function approach. While models were developed for each compound at each spatial density, sucralose was used as a model compound. Results of the desirability comparison indicated that the performance of the spatial model was very accurate at predicting spatial distributions with the full set of samples and that only half of the sites were sufficient to develop a spatial model for NB with only a minor decrease in accuracy compared to the full number of stations. The findings demonstrate the successful use of this geostatistical spatial modeling framework and indicate its potential suitability for use across similar environments.