Modelling of the petroleum hydrocarbons concentration variation in different depths of a contaminated soil during phytoremediation using fuzzy logic

1. Introduction
Isfahan Oil Refinery (Isfahan, Iran) is responsible for the production of huge amounts of oil waste. As the released organic compounds are highly toxic, carcinogenic, and mutagenic, they can potentially contaminate the soil and groundwater resources of the adjacent area. This is particularly important in Isfahan where arid/semi-arid climate has limited the access to adequate surface water resources. Among the various methods proposed for oil-contaminated soil remediation, phytoremediation has been identified as an efficient and cost-effective technique. Limited access to soil samples from various depths during phytoremediation along with the cost, time, and effort required for quantitative measurement of TPH necessitates the development of a mathematical model to overcome the existing obstacles. Fuzzy logic is a feasible method for modeling systems with inadequate or vague and non-specific information. The fuzzy set theory, introduced by Zadeh in 1965, allows the user to define the rules and understand the relations between parameters and the existing decision-making process. Consequent to its constant evolution, the fuzzy set theory has found various applications. While fuzzy logic techniques have not been as extensively applied in the environmental field as in other fields such as industrial control systems, their diversity and progression increase their potential to affect environmental policymaking.
Therefore, in recent years, numerous studies have evaluated the application of fuzzy logic methods to assess air quality and pollution, quality of surface waters, health of the rivers, groundwater contamination, and river water quality classification has also been investigated. In Iran, however, fuzzy logic has not been commonly practiced due to the unfamiliarity of environmental experts with the subject. The present study applied fuzzy logic to model TPH concentrations at different depths of soil during phytoremediation. Considering the inaccessibility of all soil depths, high costs of measurement, and the existing ambiguities, such a model will facilitate the evaluation and control of soil contamination.
2. Method
2.1. Determining physical and chemical properties of soil
Soil samples were collected from the contaminated lands contiguous to Isfahan Oil Refinery’s Sulfur Recovery Unit where oil waste was accumulated. The samples were air dried and ground to pass a 2-mm sieve. Soil structure, electrical conductivity, pH, organic matter, available potassium and phosphorus, cation exchange capacity (CEC), total nitrogen, the concentrations of TPH and polycyclic aromatic hydrocarbons (PAH) were measured according to standard methods.
2.2. Phytoremediation experiment
Phytoremediation experiments were conducted in 130-cm long polyvinylchloride pipes (width: 20 cm) with 20-cm sand filters on the bottom. The pipes had holes at 25, 50, 75, and 100 cm depths to make the final sampling possible. The prepared soil columns were planted with either sorghum or barley seeds or left unplanted. In order to assess the resistance and stability of the plants in contaminated soil, they were maintained for 17 weeks after seeding. TPH concentrations at 25, 50, 75, and 100 cm depths of all soil columns were measured 120 days after seeding.
2.3. Fuzzy modeling
Data modeling with fuzzy logic was performed in three phases using MATLAB.
2.3.1. Fuzzification of the inputs and the output
The inputs and the output were defined using linguistic variables and membership functions (MF). Depth was defined with four linguistic variables, i.e. very low (0-25 cm), low (25-50 cm), average (50-75 cm), and high (75-100 cm). Time was also defined through two linguistic variables, namely short (0-20 days) and long (20-120 days). The output (TPH concentration) was defined with four linguistic variables including low, average, high, and very high. While the Gaussian MF was applied on depth and TPH concentration, the triangular-shaped MF was used for time. The functions were determined following trial and error.
2.3.2. Defining fuzzy rules and application of fuzzy operators
According to the measured values, the fuzzy intersection (Min) and union (Max) functions were used to multiply the inputs and combine the outputs, respectively.
2.3.3. Defuzzification
Defuzzification involves the production of a quantifiable output. As we applied Mamdani fuzzy inference method, we used the center of gravity technique for defuzzification. All defuzzification calculations were performed using relevant software and the output was quantified for various inputs.
3. Results and Discussion
TPH concentrations in treatments with sorghum and barley and also unplanted (control) treatments demonstrates that, increasing depth was associated with higher concentrations of TPH and smaller differences between the treatments. More precise, TPH concentrations of control and planted treatments were significantly different at the 0-25 cm depth (P < 0.05). However, as both sorghum and barley spread their roots at this depth, no significant difference was observed between planted soil columns. In fact, the extensive root systems of the two species enhanced the microbial activity in the rhizosphere and accelerated the decomposition of petroleum compounds. Compared to baseline, sorghum and barley decreased TPH concentrations by 64% and 52%, respectively. These values were 23%-35% greater than those detected in the control soil. At the 25-50 cm depth, the difference between TPH concentrations of the control and planted soils was still significant (P < 0.05). Meanwhile, considering sorghum’s higher root penetration, the planted treatments were also significantly different in terms of TPH concentration at this depth. At 50-75 and 75-100 cm depths, no significant difference were detected between TPH concentrations of the treatments. In fact, since the roots of sorghum and barley could not penetrate into such great depths, the three types of treatment had almost identical conditions (Fig 1).


Figure 1. Changes in the concentrations of total petroleum hydrocarbons at different depths of planted and unplanted soil columns

After fuzzification of the inputs and the output, defining fuzzy rules and application of fuzzy operators to combine fuzzy relations and aggregate the outputs, and finally defuzzification, the output values were calculated. Comparison between the measured concentrations of total petroleum hydrocarbons and the values obtained from the fuzzy model after 120 days of phytoremediation with sorghum, barly and control treatment showed the fuzzy model was well capable of determining TPH concentrations at various depths of soil during the phytoremediation process.
4. Conclusion
The present study designed a fuzzy model to determine TPH concentrations during the phytoremediation process in lands adjacent to Isfahan Oil Refinery. The measured concentrations decreased by 52%-64% in soils planted with sorghum and barley. These rates were 23%-35% greater than the values obtained from unplanted treatments. Since even small amounts of organic contaminants can seriously threaten human health, enhanced elimination of petroleum-based contaminants in presence of sorghum and barley plays a critical role in improving soil conditions in the area. On the other hand, not only is the quantitative measurement of TPH a difficult, time-consuming, and costly task, but it also requires access to different depths of soil during phytoremediation (which is not always possible). Therefore, we determined the concentrations at different times and depths by developing a fuzzy model. The applied model was actually able to mathematically formulate the existing limitations and facilitate decision-making and inference through its simple, flexible concepts.
Considering the novelty of fuzzy logic techniques in soil and water resources studies, particularly in Iran, further, more diverse research on the application of such methods in various fields of integrated soil and water resources management can lead to improved prediction and modeling accuracy at lower cost and time. As the values calculated by our fuzzy model were consistent with the measured TPH concentrations, this model can also be utilized in other contaminated areas. Meanwhile, the model comprised 10 different MFs (four for depth, two for time, and four for the output) whose parameters could be modified by the user and thus alter the numerical value of the output. Since selecting appropriate values for the parameters is complicated, future studies are suggested to use optimization methods such as genetic algorithms determine the best parameters for MFs.