Applying multivariate statistics to analyse data variability in groundwater quality monitoring of contaminated sites

Groundwater quality monitoring of contaminated sites represents a fundamental step for implementing effective remediation strategies. Groundwater in hydrocarbon contaminated sites is often monitored using fully screened piezometers, obtaining concentration time-series that can be affected by a strong variability. This variability can complicate data interpretation and lead to prolonged site closures. One possible solution to compensate for this data variability is to increase the monitoring frequency to better detect contamination trends. Nonetheless, this solution can be less economically sustainable. Thus, understanding and quantifying variability of monitoring data is pivotal to support remediation strategies. According with McHugh et al. (2011), the variability of monitoring data could be due to: a) long-term trend in the contaminant source; b) time-independent factors related to both well (e.g., screen length and depth) and aquifer characteristics (e.g., hydraulic conductivity, unsaturated zone thickness); c) non-standardized sampling procedures (e.g., purging and sampling flow rates, vertical position of the sampling pump); d) frequent changes in the laboratory. 

This study presents the analysis and quantification of data variability of contaminant (total hydrocarbons and benzene) concentrations in a former oil refinery located in Northern Italy. Data variability was firstly quantified calculating the coefficient of variation (CV). Subsequently, different statistical analyses were conducted to identify and quantify the main factors affecting the data variability: Mann-Kendall test and Sen’s slope estimator, correlation analysis, factor analysis and multiple linear regression models. The working dataset refers to total hydrocarbons, benzene, redox-sensitive species (oxygen, nitrate, manganese, iron, sulfate and methane) and field parameters monitored in 41 fully screened piezometers from 2011 to 2021. Results pointed out that 11 years’ time-series of concentration do not show significant temporal trends, thus evidencing a relative stability of the contaminant plume. The CV of total hydrocarbons and benzene resulted lower in the plume core, characterized by methanogenesis and iron reduction, and higher in the plume fringe, characterized by sulfate, nitrate and/or oxygen reduction. The greater variability found in the fringe area is consistent with the vertical heterogeneity of biodegradation activities and redox states featuring the plume fringe (Meckenstock et al. 2015). Accordingly, factor analysis pointed out a positive correlation between CV and sulfate and a negative correlation between CV and methane. A multiple linear regression model of total hydrocarbons with sulfate and methane as independent variables (p-value of 0.031) obtained a r² value of 0.439. This result can indicate that vertical heterogeneity is able to explain the 43.9% of total variability in total hydrocarbons concentrations. The remaining percentage of data variability is due to unidentified factors, including the adoption of non-standardized sampling procedures, the change in analytical procedures and labs, etc. In conclusion, this works confirmed the ineffectiveness of monitoring groundwater quality through fully screened piezometers in hydrocarbon contaminated sites. The adoption of multi-depth monitoring system could reduce data variability in the studied site of, at least, the ~44%.

References:

McHugh et al. (2011) Gr Water Monit Remediat 31:92–101. https://doi.org/10.1111/j.1745-6592.2011.01337.x

Meckenstock et al. (2015) Environ Sci Technol 49:7073–7081. https://doi.org/10.1021/acs.est.5b00715