Statistical modeling of sediment and oyster PAH contamination data collected at a South Carolina estuary (complete and left-censored samples)

Abstract

This paper presents an analysis of polycyclic aromatic hydrocarbon (PAH) sediment and oyster contamination data collected at Murrells Inlet, South Carolina. Murrells Inlet is a high salinity estuary located in a heavily urbanized area south of Myrtle Beach, South Carolina. In the first part, lognormal and Weibull distributions are determined that best fit the data, as measured by P–P and Q–Q probability plots. The results indicate that the Weibull gives an adequate fit for almost all the PAH analytes considered. In fact, the Weibull almost always provides a better fit to the data than the lognormal distribution. The second part addresses issues associated with non‐detection points, as they are regularly encountered in environmental analyses. In statistical terms, the existence of non‐detection points corresponds to data that are left‐censored. Several statistical methods for estimating the Weibull parameters from such left‐censored data are explored. The overall result is in agreement with recent findings reported by other investigators: methods based on the underlying distribution of the data give more consistent results than those obtained by commonly used substitution methods. Copyright © 2000 John Wiley & Sons, Ltd.

A tiered approach for the human health risk assessment for consumption of vegetables from with cadmium-contaminated land in urban areas

“…Since lognormal distribution is often considered the default in environmental data analysis (e.g., Finley et al, 1994;Thompson et al, 2000) the geometric mean of the available BCFs is selected for that specific vegetable instead of a BCF based on a plant-soil Freundlich-type relation. In spite of the disadvantage of a generic BCF, a geometric mean of generic BCFs is considered better than leaving out the plant uptake characteristics of relevant vegetables.…”

Partial L-moments for the analysis of censored flood samples / Utilisation des L-moments partiels pour l’analyse d’échantillons tronqués de crues

“…A good parameter estimation technique for censored data should yield results in agreement with those obtained from the complete data (Thompson et al, 2000). To evaluate this, the sampling properties, namely bias and rmse on different quantile estimators, x(F), obtained from the method of PL-moments (involving different censoring thresholds, F 0 ), were compared with those from simple L-moments (involving uncensored data) for various sample sizes, n, and for different values of GEV shape parameter, k. Results show that the bias on quantile estimates obtained from the method of PL-moments with the level of censoring of F 0 ≤ 30% is very similar to that of simple L-moments, and for this range of censoring, the actual value of bias lies within -10% to +10%.…”

“…Thompson et al (2000) explored several statistical methods for estimating the Weibull parameters from left-censored data. Durrans & Tomic (2001) evaluated performances of five different methods which place an emphasis on fitting the lower tail of lognormal distribution for estimation of the 10-year low-flow quantile, and found no clear "winner" among them.…”

Assessment of Human Health Risks for Arsenic Bioaccumulation in Tilapia ( Oreochromis mossambicus ) and Large-Scale Mullet ( Liza macrolepis ) from Blackfoot Disease Area in Taiwan

“…USEPA (1997) in guiding principles for Monte Carlo analysis indicated that fit in the vicinity of expected values and in the tails were important criteria. Finely et al (1994) and Thompson et al (2000) indicated that the lognormal distribution is often considered the default in environmental analysis. Distributions were fit to polled BCF data and the selected lognormal distributions had the acceptable 2 fit and K-S fit in that optimizations using either statistics yielded geometric mean (gm) and geometric standard deviation (gsd) expressing as LN (gm, gsd) (Figure 1).…”