Cellular Autometa Matlab algorithm

Agent-Based Model to Simulate Groundwater

Remediation with Nanoscale Zero Valent Iron

Davide De March1,3, Alessandro Filisetti2, Elisabetta Sartorato1, and Emanuele Argese1

1 Department of Molecular Sciences and Nanosystems, Ca’ Foscari University of Venice, Dorsoduro 2137, 30121 Venice, Italy

2 C.I.R.I. - Energy and Environment Alma Mater Studiorum - University of Bologna via F.lli Rosselli, 107 42123 Reggio Emilia - Italy

3 EvoSolutions S.r.l., Viale Ancona 17, 30172 Venice, Italy

Abstract. Soils, air and water have been deeply contaminated by an- thropogenic activities continuously spread over time. One of the most dangerous pollutant in groundwater is represented by chlorinated organic solvents, which acts as a Dense Non Aquifer Phase Liquid (DNAPL) contaminant. Many laboratory experiments have shown that nZVI en- capsulated into micelles could treat DNAPL pollution directly into the groundwater but very few in situ experimentations have been tested. Agent-Based Model (ABM) is a powerful tool to simulate and to gain better insights in complex systems. In this paper we present an ABM sim- ulation of DNAPL contaminated groundwater remediation. The model simulates a dehalogenation process of Trichloroethylene (TCE) with the application of encapsulated nZVI, directly injected into the DNAPL con- taminant source.

1 Introduction

The chlorinated organic solvents (Trichloroethylene, Tetrachloroethylene) are very common contaminants of groundwater and aquifers. Chlorinate solvents are very low soluble and more dense than water, and when released in ground- water they do not dissolved but become Dense Non Acquifer Phase Liquid (DNAPL). However, some DNAPL components are slowly and continuously re- leased into groundwater, creating a persistent source of contamination that could last for ages. Conventional techniques to remediate contaminated groundwater are “Pump and Treat” [1] and “Permeable Reactive Barrier” (PRB)[8] but they are very inefficient when treating DNAPL. In fact, these techniques are able to remediate only a very small part of DNAPL (i.e the soluble part of the pollu- tant) and, in the case of PRB, the source of contamination is not directly treated. Moreover, the remediation of a DNAPL contaminated groundwater needs un- sustainable time and costs.

In the last few years, a very promising technique for this kind of contamination seems to be the nano-scale Zero Valent Iron (nZVI) suspension for the reductive dehalogenation of chlorinated organic solvents. The use of colloidal suspension of nano-particles of iron allows the treatment of the DNAPL in aquifer, reaching

G.C. Sirakoulis and S. Bandini (Eds.): ACRI 2012, LNCS 7495, pp. 351–359, 2012. c© Springer-Verlag Berlin Heidelberg 2012

352 D. De March et al.

also deeper contaminated areas and directly reacting into the source of contami- nation or into dissolved contaminant of the plume. Unfortunately, this technique has to face some problems when applied for in situ remediation. In particular, due to their high specific surface area and, consequently, their remarkably high reactivity, nano-particles tend to form bigger aggregates of micrometers size, giving a dimensional instability. In addition, the nano-particles tend to inter- act with non target compounds, such as nitrate, perchlorate and heavy metal, reducing their efficiency in treating the target. Therefore, it is necessary to cre- ate a formulation of nZVI able to optimise the mobility, the stability and the reactivity of the nano-particles.

The encapsulated nZVI (EZVI) into artificial membranes, obtained with the emulsification of oil and sulfactant, represents a very good compromise among stability, selectivity and reactivity [4]. The surfactant surrounds the nanoparti- cles, forming very small colloidal droplets. This particular emulsifier has many advantages among which to be miscible in DNAPL, to protect nZVI from other non target compounds and to enhance the mobility of the membranes. In the last few years many laboratory experimentations have shown good results in the reductive dehalogenation of chlorinated organic solvents with encapsulated nZVI, but large information gaps currently exist about their behaviour in en- vironmental condition such as in situ remediation. A very useful tool to study in situ remediation is the application of models which are able to characterise the environment of the contaminated area and simulate possible remediation scenarios [11].

The choice of an agent-based description turns to be useful in order to simulate the reaction within different environments. In classical chemistry solver tools the simulation of different environments is a difficult tasks. Conversely, modelling the system by means of the characteristics of single entities allows to deal with different environments, e.g. different soil compositions, by changing only few proprieties, in terms of composition, boundaries and shape, of the environment where molecules react.

In this work we present an agent-based model (ABM) to study the evolution of the reductive dehalogenation of a DNAPL, composed of trichloroethylene (TCE), treated with emulsified nZVI. The paper aims to combine the infor- mation obtained from laboratory experimentations with a simulation of possible groundwater contaminated scenarios. We present, in Sect. 2, the chemical exper- imentation of a degradation of TCE, treated with nZVI and Palladium (Pd); in Sect. 3 we present an agent-based model, simulating the in vitro experimentation of the reductive dehalogenation of TCE treated with nano-particles of Fe and Pd and in Sect. 4 the application of the ABM to simulate a DNAPL remediation of a scenario of a contaminated area. Conclusion will follow in Sect. 5.

2 Chemical Experimentation of Trichloroethylene Remediation

The experimentation is set up in a chemical laboratory of Ca’ Foscari University. We study the dechlorination of the TCE in water solution, adding nZVI with Pd,

Agent-Based Model to Simulate Groundwater Remediation 353

in proportion of 0.1%Mol/Fe, which enhances the reactivity and functionality of Fe and consequently accelerates the rate of the reaction kinetics [3,7,14]. The concentration of pollutant is set to 900mg/L in a water solution where 10g/L of Fe/Pd are dissolved. The reductive dechlorination of TCE by nZVI follows the pseudo-first-order reaction kinetic governed by Eq. 1 since the concentration of nZVI is almost 30 times the concentration of TCE and it could be considered as a constant.

Ct C0

= e−kobs·t, (1)

where C0 is the concentration of TCE (900mg/L) at time 0 and Ct is the con- centration of the pollutant at time t. Experimental samples are taken every 5 minutes in the first hour and every 30 minutes up to the 5th hour and the con- centration of TCE is recorded, as shown in Fig. 1.

Fig. 1. Dechlorination kinetics of 900mg/L of TCE in water solution with 10g/L of Fe with 0.1% moles of Pd. The plot shows the logarithm of the concentration ratio of the pollutant at time t and at t = 0 versus time

The reaction rate, kobs, is calculated by a linear regression in the time interval from 10 to 120 minutes, when maximum reductive dechlorination happens. We show in Table 1 that nZVI is significant (p-value � 0.001) and the slope of

Table 1. Summary of the linear regression model of degradation of trichloroethylene in water solution with nZVI

Coef SE T-Value P-Value

nZVI -0.03227 0.001 -23.06 2.58 e-09

R2 = 0.98

354 D. De March et al.

the regression is negative with a value equal to 0.0323min−1. The regression coefficient of determination R2 = 0.98 shows the goodness of the model.

The laboratory experiment shows an effective performance of nZVI for reduc- tive dehalogenation of trichloroethylene.

3 The Agent-Based Model Approach to Simulate Reductive Dehalogenation of Trichloroethylene

According to the experimental settings, we initially recreate the reaction of TCE reductive dehalogenation with Fe/Pd, in order to validate our model. The re- action is simulated with Netlogo 5.0 [15] starting from the model presented by Stieff and Wilensky[12]. NetLogo is an agent-based integrated modeling frame- work with two principle types of agents that are turtles (that move around in “the virtual world”) and patches (square pieces of “ground” over which turtles can move).

NetLogo allows to couple the peculiarity of the agent-based models with the spatial configurations typical of cellular automata. Each entity in the system has its own proprieties and a lattice description allows to simply describe the interactions. The update of the system is synchronous and in order to recreate the experimental settings, we scaled the Netlogo coordinates in order to create a realistic representation of the system dimensions. Therefore, to represent a beaker containing a liter of solution ( 1dm3) we convert the “Netlogo world” to a 2D world with size of 20 × 50 patches (the third dimension is considered equal to unit) so that each patch contains a volume of 1cm3 = 1mL of solution. To simulate a realistic number of molecules in the solutions we convert the experimental concentrations to moles. The TCE has an atomic weight equal to 131.79, so the initial quantity C0 = 900mg/L is converted to 6.83mMol/L. Similarly, Fe has an atomic weight of 55.847 and its initial quantity, set to 10g/L, is converted to 179mMol/L. Figure 2a shows the initial condition of the solution characterised by an average concentration of 6830μM/L of pollutant (green coloured turtles) and of 179000μM/L of nZVI (red coloured turtles). Because the real experiment is performed in an well-stirred solution, we consider a random walk of length 1 patch for each turtle as a plausible diffusion movement in the beaker. The reaction time is calculated according to Eq. 1 with the reaction rate set to kobs = 0.0322min

−1 (as calculated in the laboratory experimentation). Mimicking the dechlorination reaction of TCE due to nZVI we assume that the reaction of one particle of pollutant with one particle of nZVI creates a new product (i.e. Ethene, blu coloured in Fig. 2b) by transforming the reaction compounds [5,6].

The simulation runs until all the pollutant molecules are remediated and we observed that the degradation kinetics of the TCE, in Fig. 3, matches reason- ably well the behaviour of the real experimentation. The initial settings of the laboratory test is not suitable for in situ remediation; in fact, in field experimen- tation we expect that pollutant and reactant do not largely differs in concentra- tion and for that reason we cannot assume that the reaction in a contaminated

Agent-Based Model to Simulate Groundwater Remediation 355

(a) Initial world (b) 10 Minutes reac- tion time

(c) Final world

Fig. 2. Simulation of the reaction of TCE in 1 Liter of water solution with Fe/Pd. In panel 2a the initial conditions of the system are represented. Red turtles stand for Fe and and green turtles stand for the pollutant molecules. In panel 2c the state of the system at the end of the simulation is represented. It is possible to observe that all pollutants have been converted into non toxic compounds such as Ethene [13] . Panel 2b shows the state of the system after a time of 10 Minutes.

0 50 100 150 200 250 300

0 .0

0 0

0 .0

0 1

0 .0

0 2

0 .0

0 3

0 .0

0 4

0 .0

0 5

0 .0

0 6

0 .0

0 7

Time

T C

E c

o n c e n tr

a ti o n m

o l/ L

Experimental

Simulation

Fig. 3. simulation of degradation kinetics of 0.06Mol/L of TCE in water solution with 0.179Mol/L Fe with 0.1% of Pd with a fixed degradation constant k = 0.03min−1. Solid line represents contaminant concentration of the real experimentation, and dotted line represents the simulation of the reaction with ABM

356 D. De March et al.

groundwater is of a pseudo-first-order. The most plausible scenario is that the dehalogenation follows a second-order reaction with a second order rate constant calculated as Ks = Kobs/[Fe] = 0.03/0.179 = 0.168Mol ·min−1. We validate the model simulating a more realistic concentration of both contaminant and reactant, setting an equal concentration of both TCE and nZVI. Results are then compared with Dizzy [10] software1.

0 10000 20000 30000 40000 50000

0 .0

0 0

0 .0

0 1

0 .0

0 2

0 .0

0 3

0 .0

0 4

0 .0

0 5

0 .0

0 6

Time

T C

E c

o n c e n tr

a ti o n m

o l/ L

ABM sim

Deterministic sim

Fig. 4. Simulation of degradation reaction of 0.06Mol/L of TCE in water solution with 0.06Mol/L Fe with ABM (solid line) and Dizzy software (dotted lines) with a second order rate constant k = 0.168Molmin−1

The agent-based model is able to recreate a second-order reaction of reductive dehalogenation of TCE when treated with nZVI and this model is applied in a large scale simulation of a DNAPL groundwater contaminated area.

4 The Study of a Dense Non Aqueous Phase Liquid (DNAPL) Contaminated Groundwater

The dechlornation model presented in Sect. 3 is used in a simulative study to investigate the remediation of a DNAPL contaminated groundwaters. The con- taminated area is characterised by a TCE concentration of 6.3μg/L and the DNAPL is treated with nZVI, incapsulated into amphiphilic molecules. Previ- ous laboratory experimentations have shown that the reductive dehalogenation of TCE, treated with emulsified nZVI, is governed by a pseudo-first-order ki- netic with a kobs of about 0.01 − 0.03min−1 [2] and kobs = −0.01064min−1 is selected, according to the laboratory experimentation obtained with the same

1 Available at http://magnet.systemsbiology.net/software/Dizzy/. Although Dizzy is actually developed to perform stochastic simulations, it provides also a use- ful tool for the simulation of deterministic systems. Once that the reactions scheme has been created, the user has the possibility to choose which kind of simulation performs. With regard to these analysis we adopted a finite difference method ODE solver, specifically the 5th-order Runge-Kutta algorithm with an adaptive stepsize controller.

Agent-Based Model to Simulate Groundwater Remediation 357

settings presented in Sect. 2. Hence, we assume to use the same reaction model, presented in Sect. 3, to simulate a second-order reaction with a rate constant equal to Ks = Kobs/[Fe] = 0.0106/0.179 = 0.0594Mol · min−1. We assume that a slurry of encapsulated nZVI is injected into the contaminated area with a concentration equal to the pollutant to be treated. The groundwater horizontal hydraulic conductivity (Kx) is estimated to be 1.2m ·day−1 and the contami- nant area is limited into a small area. In accordance with these assumptions we set our simulator as a Netlogo world with size 100×10 patches, representing an area of 103 m3 (the third dimension is set equal to unit).

Therefore, each patch represents 1 m3 of soil. Following the same conversion used in Sect. 3, the concentration of TCE in the DNAPL is 4.79∗10−2 μMol/L and the simulation unit becomes 47.9 μMol/patches. The contaminated area is simulated to be on the top-left of the represented scenario with a size of 10 × 5 = 50 patches (gray coloured in Fig. 5a), and the soil is considered with a 30% porosity. 47.9 × 50 ≈ 2500μMol of pollutants are randomly added to the contaminated area. We, then, introduce the nZVI through a injection pump in the middle of the DNAPL with the same concentration of TCE, an initial pressure is simulated to distribute the reactant into the contaminated soil. The

(a) Initial world

(b) 10 Days reaction time

(c) Final world

Fig. 5. Simulation of degradation of a Dense Non Acqueus Phase Liquid contaminated area: Panel 5a represents the initial condition where pollutant (green coloured) is spread into the DNAPL and nZVI incapsulates (red coloured) is injected in a column into the DNAPL. Panel 5b shows the movement of the DNAPL into the porous soil after 10 days and the products (blue coloured) originated by the dechlorination reaction. Panel 5c shows the simulation after 30 days where all the contaminants, nZVI and products are deposited on the bottom of the groundwater

358 D. De March et al.

agent movement is due to two forces, the horizontal hydraulic conductivity (Kx) and the infiltration conductivity (Ky), so that the direction (α) of the movement is calculated as:

α = arctan Kx Ky

(2)

Each step of the simulation (“tick” in Netlogo vocabulary) is defined as one hour and consequently Kx = 1.2/24 = 5∗ 10−2m ·h−1; the infiltration conduc- tivity, Ky, is set half of the horizontal hydraulic conductivity. The simulation runs for one month and the evolution of the DNAPL reduction is evaluated. Figure 5b shows the movement of the DNAPL after 10 days of simulation and a reduction of TCE and the formation of new products (blue coloured) due to the dechlorination reaction with nZVi emulsified with oil and surfactant is observed. The concentration of pollutant after 10 days is reduced of about 50%. At the end of the simulation (after 30 days in Fig. 5c) DNAPL reaches the bottom of the groundwater and the dechlorination of the TCE is almost 90%.

5 Conclusion

The use of agent-based model to simulate DNAPL reduction could represent an useful tool when investigating possible in situ remediation. A combination of laboratory experimentations and simulative models seems to be a promising approach in order to characterise the contaminant remediation dynamics. This work is an initial application in the contaminant remediation simulation and many parameters should be deeply investigate in order to give a more realis- tic simulation of real contaminated areas. In fact, soil characterization is very important to understand the dynamics of the system and it should be tested into some pilot scale experimentations. Moreover, in this simulation we assume optimal condition for the DNAPL in which only one contaminant, the TCE, is present; conversely, it is quite common the presence of many other pollu- tants which the nZVI reacts with. Results presented in this preliminary work are comparable to previous studies [9], nevertheless the usage of an agent-based description allows to adapt a simple model to different environmental conditions. Although we present here only an initial investigation, our model can deal with different environmental scenarios allowing an in-depth characterisation of the remediation phenomena with respect to different soils, groundwater conditions and different pollutants characteristics.

Acknowledgement. We would thank prof. Roberto Serra, prof. Marco Villani and Prof. Irene Poli for their helpful comments and suggestion.

References

1. Bayer, P., Finkel, M.: Life Cycle Assessment of Active and Passive Groundwater Remediation Technologies. Journal of Contaminant Hydrology 83, 171–199 (2006)

Agent-Based Model to Simulate Groundwater Remediation 359

2. Bezbaruah, A., Shanbhogue, S., Simsek, S., Khan, E.: Encapsulation of iron nanoparticles in alginate biopolymer for trichloroethylene remediation. Journal of Nanoparticle Research 13(12), 6673–6681 (2011)

3. Cho, Y., Choi, S.-I.: Degradation of PCE, TCE and 1,1,1-TCA by nanosized FePd bimetallic particles under various experimental conditions. Chemosphere 80, 940– 945 (2010)

4. Kanel, S.R., Choi, H.: Transport characteristics of surface-modified nanoscale zero- valent iron in porous media. Water Science & Technology 55(1-2), 157–162 (2007)

5. Lin, C.J., Lo, S.L., Liou, Y.H.: Dechlorination of trichloroethylene in aqueous solu- tion by noble metal-modified iron. Journal of Hazardous Materials 116(3), 219–228 (2004)

6. Liu, Y., Majetich, S.A., Tilton, R.D., Sholl, D.S., Lowry, G.V.: TCE Dechlorina- tion Rates, Pathways, and Efficiency of Nanoscale Iron Particles with Different Properties. Environmental Science & Technology 39, 1338–1345 (2005)

7. Lowry, G.V., Reinhard, M.: Pd-catalyzed TCE dechlorination in water: effect of [H2] (aq) and H2-utilizing competitive solutes on the TCE dechlorination rate and product distribution. Environ. Sci. Technol. 35, 696–702 (2001)

8. Powell, R.M., Puls, R.W., Blowes, D.W., Vogan, J.L., Gillham Guelph, R.W., Pow- ell, P.D., Schultz, D., Sivavec, T., Landis, R.: Permeable Reactive Barrier Technolo- gies for Contaminant Remediation, US Environmental Protection Agency, Techni- cal Information Office, Washington, DC. USEPA/600/R-98/125 (September 1998)

9. Quinn, J., Geiger, C., Clausen, C., Brooks, K., Coon, C., O’Hara, S., Krug, T., Ma- jor, D., Yoon, W., Gavaskar, A., Holdsworth, A.: Field Demonstration of DNAPL Dehalogenation Using Emulsified Zero-Valent Iron. Environmental Science & Tech- nology 39(5), 1309–1318 (2005)

10. Ramsey, S., Orrell, D., Bolouri, H.: Dizzy: stochastic simulation of large-scale genetic regulatory networks. Journal of Bioinformatics and Computational Biol- ogy 3(2), 415–436 (2005)

11. Di Gregorio, S., Serra, R., Villani, M.: Applying cellular automata to complex environmental problems: The simulation of the bioremediation of contaminated soils. Theoretical Computer Science 217, 131–156 (1999)

12. Stieff, M., Wilensky, U.: NetLogo Enzyme Kinetics model. Center for Connected Learning and Computer-Based Modeling. Northwestern University, Evanston, IL (2001), http://ccl.northwestern.edu/netlogo/models/EnzymeKinetics

13. Zhang, W.-x.: Nanoscale iron particles for environmental remediation: An overview. Journal of Nanoparticle Research 5, 323–332 (2003)

14. Wei, J., Xu, X., Liu, Y., Wang, D.: Catalytic hydrodechlorination of 2,4- dichlorophenol over nanoscale Pd/Fe: reaction pathway and some experimental parameters. WaterRes. 40, 348–354 (2006)

15. Wilensky, U.: NetLogo. Center for Connected Learning and Computer-Based Mod- eling. Northwestern University, Evanston (1999), http://ccl.northwestern.edu/netlogo/