Identifying remedial solutions through optimal bioremediation design under real-world field conditions

Highlights

• Coupled simulation/optimisation approach was used to optimize bioremediation design.

• Simulations were performed with a multi-component reactive transport model.

• Particle Swarm Optimization was used for optimization.

• The influence of multiple optimization objectives on remediation design is illustrated.

Abstract

Over more than a century of intense industrial production and associated accidental release, petroleum products (e.g., gasoline, diesel, fuel oil) have contaminated a significant portion of the world's groundwater resources. Groundwater remediation is generally a complex task, especially where aquifers and the associated contaminant distribution are highly heterogeneous. The ability to predict the efficiency of such remediation is of crucial importance, as the costs are strongly linked to the treatment design and duration. In this study, a coupled simulation-optimization (S/O) framework, consisting of a process-based reactive transport simulation model linked with particle swarm optimization (PSO) was developed. It was subsequently applied for the design of a real-world in situ bio-treatment of a BTEX contaminated aquifer in France. In the application, the optimization framework was used to simultaneously determine optimal well locations and their optimal injection rates, both constituting key elements of the enhanced biodegradation design problem. The optimization of the treatment efficiency was examined in terms of three different regulatory objectives, (1) minimization of the residual NAPL mass of the key contaminant, i.e., benzene, in the source zone, (2) reduction of the maximum concentration of benzene in groundwater, and (3) minimization of the time required to reduce the benzene concentration in groundwater to below a threshold value. Our analysis of potential, optimal remediation strategies showed that: (i) the complexity of the biodegradation behavior at real sites may favor very different remediation options as a result of varying remediation targets, (ii) the long term behavior of the contaminants after the end of the active treatment period, which is often neglected, showed to have a significant influence on remediation design that requires increased attention, (iii) PSO has shown to be a very efficient algorithm in the context of the present study. The insights that can be gained from such a framework will provide decision support to select the most suitable remediation strategy while facing different regulatory objectives.

Introduction

Over more than a century of intense industrial production and associated accidental release, petroleum products (e.g., gasoline, diesel, fuel oil) have contaminated a significant fraction of groundwater resources that, despite the massive research and remediation efforts over the last three decades, still remain a primary concern for groundwater quality. Water-soluble and mobile gasoline components such as benzene, toluene, ethylbenzene and xylenes isomers (BTEX) have caused thousands of contaminant plumes in groundwater, originating from spills and other sources that can persist for decades or centuries (Barker et al., 1987). The majority of the released organic compounds are toxic chemicals that often exceed drinking water standards. In the 1970s, biodegradation has been recognized as an important mechanism for reducing the concentration of organic contaminants like BTEX (Borden et al., 1997; Chapelle, 1999; Davis et al., 1999; Prommer et al., 1999), polycyclic aromatic hydrocarbons (Cerniglia, 1993; Haritash and Kaushik, 2009) and chlorinated hydrocarbons (Wiedemeier et al., 1997). The two major site remediation technologies that rely on the occurrence of biodegradation processes are monitored natural attenuation (MNA), (Wiedemeier et al., 1999; Wilson et al., 1994) and enhanced in situ biodegradation (Chen et al., 2010; Gibson et al., 1998; Höhener et al., 1998; Hunkeler et al., 1999).

Enhanced in situ biodegradation of oxidisable organic pollutants relies on the injection of electron acceptors and nutrients to stimulate microbial activity, and thus improving the rate of biodegradation by supplying growth-limiting reactants (e.g., Farhadian et al., 2008; Hyman and Dupont, 2001; Megharaj et al., 2011). The technical challenge is to inject the electron acceptors and nutrients in such a way that they are adequately dispersed throughout the contaminated area (Chapelle, 1999). The efficiency of any selected remediation scheme will depend on how effectively the amendments can be transported to the zones where particular reactants are deficient. To achieve this, several design parameters can be adjusted to control the bio-treatment process (Zou et al., 2009) by manipulating the naturally occurring flow, solute transport and biogeochemical reaction processes. The most important parameters that affect the design and thus the efficiency of the in situ bioremediation process might be

• the number and locations of wells (Akbarnejad-Nesheli et al., 2016; Bayer and Finkel, 2007; Kumar et al., 2013; Minsker and Shoemaker, 1998a),

• the extraction (Fan et al., 2014; He et al., 2010; Li et al., 2015; Mategaonkar and Eldho, 2012) and the injection rates (Zou et al., 2009) applied at each well,

• the concentrations of electron acceptors (Huang et al., 2008) and

• nutrient concentration (Hu et al., 2007; Hu and Chan, 2015).

Given the complexity of the interactions that can result from variations in the setup of the bioremediation scheme it is not surprising that the first attempts to systematically optimize bioremediation schemes by coupling numerical process models with optimization algorithms have already been made over 30 years ago. Gorelick et al. (1984) and Minsker and Shoemaker (1998b), for example, developed initial approaches to support decision makers in the design and operation of groundwater in situ bioremediation schemes. They used non-linear optimization combined with finite element groundwater flow and contaminant transport simulations to determine optimal well locations as well as extraction and injection rates. Later, Minsker and Shoemaker, 1998b, Minsker and Shoemaker, 1998a presented a coupled optimal control and simulation model to determine the location of extraction and injection wells and the pumping rates in cost-effective in situ bioremediation. Yoon and Shoemaker (1999) studied eight optimization algorithms to design an in situ bioremediation system with least cost and identified an algorithm with the best combination of efficiency and accuracy. Shieh and Peralta (2005) proposed a simulation-optimization model for the selection of a cost-effective remediation strategy. They combined a genetic algorithm and simulated annealing algorithm to search for optimal control of in situ bioremediation and applied the BIOPLUME II model for simulating the transport and bioremediation processes. Yadav et al. (2016) also used BIOPLUME III on a synthetic case study, where they compared the performance of different surrogate modeling techniques to reduce the computational cost of in situ bioremediation design. They employed particle swarm optimization (PSO) to solve a simple hypothetical problem. The proposed simulation-optimization model was later used, on the same synthetic case study, to design an optimal in situ bioremediation system where biological clogging is accounted for (Yadav et al., 2018). Hu et al. (2007) developed a control system for in situ bioremediation of groundwater using a genetic algorithm for the multi-objective optimization method under uncertainty. Their objectives included minimizing the overall cost and the cleanup duration. Prasad and Mathur (2008) applied a neural-network-embedded Monte Carlo approach to determine the potential well locations for in situ bioremediation of contaminated groundwater in order to minimize the total remediation cost. Kumar et al. (2013) proposed a methodology based on support vector machine and PSO techniques to determine well locations and pumping rates to achieve an optimal cost of an in situ bioremediation design. Subsequently, Kumar et al. (2015) proposed a multi-objective optimization approach to design an efficient in situ bioremediation system by using a hybrid metaheuristic technique based on differential evolution, genetic algorithms and simulated annealing. Most recently Akbarnejad-Nesheli et al. (2016) and Raei et al. (2017) developed a multi-objective simulation-optimization approach to determine the optimal design of an in situ bioremediation design for hydrocarbon contaminated groundwater. Both studies employed BIOPLUME to simulate the in situ bioremediation processes while a genetic algorithm was used to optimize the multi-objective function. The proposed approaches minimized the design and operational costs along with a pre-defined water-quality objective. Table 1 provides a comprehensive summary of previous studies employing reactive transport models to optimize bioremediation. It lists the employed objective function(s), the type of process model, the key biogeochemical reactions and the optimisation method, respectively, for each of the references. It can be seen that a diverse range of approaches have been explored, with numerous combinations of reaction types/complexity and optimisation methods. However, interestingly all of the compiled studies have only investigated synthetic cases and none of the studies has used real field data. It is also important to note that only two studies explored the impact of aquifer heterogeneity, i.e., Hu et al. (2007) and Zou et al. (2009). However, none of the investigated cases encapsulated the level of complexity that can occur in the field, although the spatial distributions of hydraulic conductivity (heterogeneity) can play a substantial role for bioremediation success.

Therefore we suggest that while significant progress has been made in the theoretical development of the simulation/optimization (S/O) approach, the uptake of this approach for solving realistic field-scale problems has remained extremely limited, despite the enormous potential for cost-saving and environmental benefits (Compernolle et al., 2013). Several factors may have contributed to this lack of practical applications. Firstly, the use of combined S/O methods for in situ bioremediation problems comes at substantial computational costs, which makes many complex reactive transport field-scale problems intractable. Secondly, there are currently few general-purpose and easy-to-use S/O codes available to practitioners at the field project level. Overall, the advantages of the S/O approach in solving real-world problems have not been adequately demonstrated because most, if not all studies presented in the literature have (see Table 1) used simplified hypothetical examples. Finally, and perhaps most importantly, under typical real-world conditions the optimization objectives have to consider stakeholders who have complex problems to solve, which clearly involve strict environmental regulations, but may also involve additional considerations that the stakeholders may or may not wish to consider. However, as outlined by Loucks (2012), “decision-makers don't know what they want until they know what they can get”. In this context, focusing solely on finding an optimal solution, with the definition of a unique objective function or, the definition of a unique complex multi-objective function, may not be easily adapted to a decision-making process. Optimization should be viewed as a complementary tool allowing to explore and better understand what option stakeholders have and what impact may result (Maier et al., 2014). An often-overlooked point is that applications to real field sites should also consider site-specific operational constraints, such as possible and excluded well positions, inter-well distances, or simply that the rate of injection cannot be larger than the pumped volumes.

Addressing some of the research gaps associated with real-world applications involving well placement, among other considerations, we develop in this new study a S/O framework for determining the optimal design of an in situ bio-treatment process. The framework consists of two main parts: a coupled flow and reactive transport simulation model and a PSO algorithm. The present study utilizes existing PSO methodologies within the S/O model framework for decision-making in a complex real-world context, and explores the challenges of achieving an optimal design for in situ bioremediation under real world constraints. We examine how the definition of specific objective functions impacts optimization and ultimately bioremediation treatment efficiency. This illustrates how stakeholders can explore possible management alternatives and then learn, by witnessing in this virtual environment, how specific optimization objectives trigger different ‘optimal’ management responses. Eventually, we use the results obtained from a field site in France to assess the applicability and usefulness of our proposed S/O approach as an environmental decision-making tool.