Edited by: Teng Xu, Hohai University, China
Reviewed by: Ty Ferre, University of Arizona, United States; Christian Moeck, Swiss Federal Institute of Aquatic Science and Technology, Switzerland; John Molson, Laval University, Canada
This article was submitted to Environmental Informatics and Remote Sensing, a section of the journal Frontiers in Environmental Science
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
Groundwater in most urban areas around the globe is often contaminated by toxic substances. Among the various sources of contamination, industries cause the heaviest impact when toxic compounds are released underground, mainly through leaking tanks or pipelines. Some contaminants (typically chlorinated hydrocarbons) tend to persist within the underground and are hard to biodegrade. As a result, substances that leaked decades ago are still impacting groundwater. Milano and its surroundings (Functional Urban Area) is a good example of an area that has been hosting industries of all dimensions for over a century, many of them contributing to groundwater contamination from chlorinated hydrocarbons. While the position of the biggest industrial facilities is well-known, many smaller sources are hard to identify in many cases where direct surveys have not been undertaken. Furthermore, the overlapping effects of big, small, known, and unknown sources of groundwater contamination make it challenging to identify the contribution of each. In order to identify the contribution of several point sources responsible for tetrachloroethylene contamination in public water supply wells, a numerical model (MODFLOW-2005) has been implemented and calibrated using PEST in the northwestern portion of the Milano Functional Urban Area. In contaminant transport modeling, the deterministic approach is still favored over the stochastic approach because of the simplicity of its application. Nevertheless, the latter is considered by the authors as the most suitable for dealing with problems characterized by high uncertainty, such as hydrogeological parameter distributions. Adopting a Null-Space Monte Carlo analysis, 400 different sets of hydraulic conductivity fields were randomly generated of which only 336 were selected using an objective function threshold. Subsequently, particle backtracking was performed for each of the accepted hydraulic conductivity fields, by placing particles in a contaminated well. The number of particle passages is considered as being proportional to the contribution of each unknown point source to the tetrachloroethylene contamination identified in the target well. The study provides a methodology to help public authorities to locate the “more probable than not” area responsible for the tetrachloroethylene contamination detected in groundwater and to focus environmental investigations in specific sectors of Milano.
In urban areas impacted by historical industrialization, the main problem of groundwater contamination is related to chlorinated hydrocarbons (CHCs) (Menichetti and Doni,
The methodology has been developed within the AMIIGA Project (Interreg Central Europe Grant N. CE32), where inverse transport modeling was one of the tools used to assess the unknown sources in the northwestern part of the Milano Functional Urban Area (FUA). The area of 157 km2 covers 12 municipalities with high urbanization density (about 4,000 inhabitants per km2) and a large presence of industrial sites. The area is historically affected by many chlorinated hydrocarbon plumes originating from the northern outer border of the Milano municipality (Giovanardi,
Study area: FUA municipalities with black contours, PCE contamination sampled in
The model used to carry out the stochastic simulations in this work is briefly described in this section; a more detailed description is provided by Colombo et al. (
Groundwater model:
The model has been calibrated in a steady state, as described in Colombo et al. (
Hydraulic conductivity in m/s (initial and bounding values) of the parameter used under NSMC (for the geological zones description see
1 | 10−4 | 10−5 | 10−2 |
2 | 10−6 | 10−8 | 10−6 |
3 | 10−5 | 10−5 | 5 × 10−3 |
4 | 10−6 | 10−8 | 10−6 |
The parameter sets are modified in the null-space projections and are adapted from the base model. In other words, PEST decomposes the parameter space into two perpendicular sub-spaces (solution and null-space). The first space is the representation of parameter combinations that can be estimated on the current field used for the calibration. The parts not “explained” in the solution space are spanned into the null-space (i.e., the second space). Each generated random parameter is therefore projected into the calibration null-space and the solution-space component is removed, as it is replaced with the calibrated parameter field from the base model. As the groundwater model is not linear, the stochastic parameter field needs to be re-calibrated. One option is a two-iteration method calibration. The final objective target function is then compared to the desired objective function to keep only the best calibrated K-distributions. Alberti et al. (
In this analysis, following Alberti et al. (
Conceptual flowchart for the PT+NSMC methodology.
In the literature, several examples of the NSMC method with different uses have been presented: from calibration purposes (Tonkin and Doherty,
To assess the most likely source area of the PCE contamination in the San Siro pumping wells in Milano and to identify the most likely occurring flow path, the particle backtracking technique has been applied by using the code MODPATH v.5 (Pollock,
Technical layout of the San Siro pumping station:
Comparison between the initial objective function (RANDPAR) and the final objective function obtained with the two-iteration process (SVD). The red line represents the threshold objective function limit (35 m2) used to retain the models with a good misfit to track particles.
During the calibration process, the hydraulic conductivity (K) random distributions are estimated over the available prior knowledge of the geological setting.
K distribution for Aquifer A for three randomly selected parameter realizations
Similar to Alberti et al. (
Particle frequency distribution of the ensemble backtracked pathlines for
In order to better identify the contaminated sites and to clarify the contamination paths, two different maps have been prepared separating the two aquifers:
The following analysis of the particle paths in each aquifer was applied in order to narrow the areas directly impacted by the contamination:
In Aquifer A, the area with higher frequency (more than 15%) is more evident, which was used to restrict the analyses and then to find the potential responsible site. In Aquifer B, the probability is significantly less than 10%. This map has the advantage to show that the withdrawals of the Novara pumping station are high enough to affect the flow direction in Aquifer B, acting as a “hydraulic barrier”, protecting the San Siro pumping station.
In order to understand how the contamination can move and reach the screens of well N°500, it is also useful to analyze, through some cross-sections, the probability distribution of particles in the vertical direction. The represented particle frequencies have been computed summing the number of passages for each cell (the model coordinates are row and column) and dividing it by the total sum in each column (
Longitudinal cross-section B (
Transverse cross-section A (
Stochastic particle tracking (PT+NSMC) is an innovative tool that can be helpful in the search for contamination source areas and plume monitoring. Its simplicity and versatility allows several advantages such as the ability to perform multiple analyses based on the same results, indeed two possible outcomes have been presented: (1) plan map (x,y) of the particle passage frequency in each cell of the model domain, showing all the possible pathlines that the contaminant can follow starting from a specific location, reached by the contamination, considering the uncertainty related to the hydraulic conductivity. The main advantage of the maps is that they allow us to consider a global frequency (i.e., counts all the particles belonging to a specific layer or group of layers) in order to locate the more likely area to be investigated by the installation of new piezometers or the characterization of brownfield areas. On the contrary, these maps are not able to link the vertical passage of the contamination (i.e., it would be necessary to produce as many maps as the number of layers); (2) vertical cross-sections representing the distribution of the particles cell by cell as a function of the layer (z), able to represent the impact of the contamination in each model layer and to give a 3D vision if orthogonal cross-sections are used. The advantage of these analyses is the calculation of the number of particles per vertical column and therefore the frequency for each layer (i.e., accounting for the particles passing through a specific cell) in order to show the contamination impact relative to a single cell. However, the vertical cross-section is representative of a specific area and as many different cross-sections as the number of different contaminated sites should be made (i.e., a reconstruction in a greater detail of the contamination paths). The two analyses are complimentary and can be used together to identify both potential contamination sources (i.e., higher frequency in the plan map) in order to find appropriate areas to install monitoring wells and suitable positions for the well screens (i.e., in the more contaminated layer computed in the vertical cross-section). However, this methodology shows some limitations that can be further improved. For example, a preliminary linear uncertainty analysis (Moore and Doherty,
The northwestern area in the Milano Functional Urban Area is affected by a strong presence of chlorinated hydrocarbons, originating from several contaminated sites historically present in the territory. Over the last years, this contamination has posed important safety problems in the intake area of the water supply wells of Milano. For this reason, the remediation of groundwater for drinking water purposes has become a major problem for water managers; on one hand, the water treatment procedures became more onerous both in terms of management and new plants (like new active carbon filter installations), but, on the other hand, the cost of the raw water clean-up has to be imposed on the citizens within the water bill costs. In this situation, the Polluter Pay Principle becomes crucial: its application to groundwater contamination due to the “more probable than not” responsible source is able to help charge the cost of site remediation and (mainly) the treatment costs of the water extracted for public use. The support of a groundwater model is very useful to study the contamination pathways, but, especially in the presence of multiple possible sources, the deterministic simulation fails as the uncertainties strongly influence the results both in terms of groundwater flow and in terms of contaminant transport. In this context, new methodologies and models have been developed (Alberti et al.,
The data analyzed in this study is subject to the following licenses/restrictions: Data about contaminated sites are confidential due to ongoing legal procedures. Requests to access these datasets should be directed to Regione Lombardia through the online form
LC, PM, and LA developed the model and carried out the analyses. LC and PM contributed to the development of Excel Macros and scripts that enabled the automated analysis of counting particles herein. LC, PM, and LA also contributed to the higher-level workflow. LA provided expert knowledge for defining the conceptual model. LC, PM, and MA wrote the initial draft. However, all authors contributed to the writing of the manuscript.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
The authors would like to acknowledge Regione Lombardia for the data availability on the contaminated sites (information and shapefile-AGISCO DB) and contaminant concentration data (i.e., PCE concentration). We would also like to acknowledge Eng. Giovanni Formentin and John Doherty for supporting the methodology and scripts. We thank the Milano public water manager (Metropolitana Milanese S.p.A.) for the information and data about the Milano aqueduct wells. Finally, we would like to acknowledge the reviewers that help us to improve the text and the quality of the manuscript.
The Supplementary Material for this article can be found online at: