Modelling and Multi-response Optimisation of Leachate Treatment by Electrocoagulation using Full Central Composite Design

Electrocoagulation has lately received much interest as an effective method for the treatment of landfill leachate because of its ease of use, low cost, and environmental friendliness. The present study aimed to investigate the reduction of chemical organic demand (COD), biological organic demand (BOD 5 ), and turbidity from leachate by electrocoagulation process using alumin - ium electrodes. In order to achieve the maximum removal of pollutants, highlight the key effects of the variables and their simultaneous relationships, full central composite design was applied. Moreover, the backward model selection method using Bayesian Information Criterion and Akaike Information Criterion was carried out to determine the most suitable model with relevant effects. The optimum conditions indicated by the estimated quadratic models were initial pH (5.04), current density (407 A m −2 ), reaction time (74.6 min), and stirring speed (150 rpm), adjusted R 2 of 99.82 %, 99.93 %, and 99.95 % for COD, BOD 5 , and turbidity, respectively. Electrocoagulation was also successful in achieving 90 % COD, 92.3 % BOD 5 , and 99.6 % turbidity removal efficiency. The findings indicated a satisfactory agreement between model forecasts and experimental values.


Introduction
Landfilling in most countries is currently the most common method for the disposal of municipal solid waste (DSM) due to its reduced functioning costs and simplicity of operations management. 1 Additionally, heterogeneous waste deposited in landfills undergoes many physicochemical reactions, resulting in a highly contaminated, dark, malodourous fluid called leachate, with noticeable variations in physicochemical characteristics and volumetric flow. 2 The decomposition, volume of leachate produced, stabilisation, and removal of contaminants from the waste matrix depends on several parameters, such as waste type, level of compaction, and waste absorption ability. 3 Leachate is typically highly contaminated with organic pollutants, expressed through colour, chemical organic demand (COD) and biological organic demand (BOD 5 ), and with content of ammonia, halogenated hydrocarbons, and heavy metals. 4 Unsuitable treatment of leachate causes environmental threats, in particular via the pollution of groundwater and surface water. 5 Leachate can be considered an intense wastewater pollutant, and must be adequately managed before being released to the receiving water bodies. 6 Currently, several treatment processes for landfill leachate can be used. 7 Amr et al. 8 have shown that biological treatment involves aerobic and anaerobic procedures, biological methods, as well as successful treatment of young lea-chates; meanwhile, physical and chemical treatments including a variety of techniques, such as coagulation-flocculation, adsorption, ion exchange, and oxidation, are further useful in treating stable leachates. 6 Efficient and economical methods for treating and reusing industrial wastewater have become an essential requirement for protection of the ecosystem. 9 In recent years, electrocoagulation (EC) treatments are favoured as they are cost-effective and highly efficient for the treatment of municipal wastewater compared with other treatment techniques. 10 Among these techniques, EC has been commonly employed for the treatment of municipal wastewater, 11 urban wastewater, 12 surface water, 13 domestic wastewater, 14 industrial wastewater, 15 seawater, 16 and groundwater. 17 The advantages of this technique include environmental sustainability, flexibility, selectivity, energy efficiency, mechanisation capability, protection, and cost-effectiveness. 18 In general, an EC system comprises three important stages, i.e., coagulation, flotation, and electrochemistry. 19 EC often involves electrolysis, ionisation, hydrolysis, and free hydroxyl radical processes in landfill leachate treatment, which contribute to substantial variations in the aqueous medium compositions, as well as improvement in the overall removal of contaminants. 20 Besides the relatively low cost and ease of procurement of aluminium electrodes, several studies [21][22][23] have proven that these electrodes perform better with regard to various contaminants, such as organic, inorganic, and pathogenic, in the treatments of landfill leachate. 22,24,25 Moreover, Modelling and Multi-response Optimisation of Leachate Treatment by Electrocoagulation using Full Central Composite Design in the removal of dissolved organic pollutants assessed by colour, turbidity, and COD reduction, aluminium electrodes prove to be more effective compared to other types of electrodes. 26,27 At an adequate pH, organic compounds are removed by applying the required potential. 28,29 When the electric current begins to pass through the submerged electrodes, the cathode releases hydroxide ions (OH − ) and hydrogen gas (H 2 ), and the anode begins to dissolve and generate highly charged cations (Al 3+ ). At the surface of the anode, the contaminants are adsorbed and the colloidal particles destabilised by the formation of complex hydroxo-monomeric and polymeric aluminium species, which have high adsorption properties, forming strong aggregates with the pollutants that can be separated by filtration, flotation or sedimentation. [30][31][32][33][34] Kobya et al. 35 stated that the efficacy of EC treatment is related to several specific operating criteria, like electrode material used, electrode distance, electrolyte concentration, initial pH, conductivity, reaction time, initial concentration of the pollutants, mixing speed, and current density. 36 Like any other processing technology, EC has some disadvantages that could affect its performance, i.e., high costs of electricity and generation of secondary pollutants. [37][38] In many studies investigating wastewater treatment using conventional EC method, the authors proceeded by changing one variable at a time (OVAT) while keeping constant the other variables of the process. 39,40 This approach has the shortcoming of depending upon guesswork, chance, experience, and intuition. Furthermore, the procedure needs massive amounts of resources to obtain a limited number of method data. OVAT experiments are sometimes inaccurate, ineffective, and time-consuming, and can lead to incorrect optimum process conditions. 41 These limitations can be addressed by using response surface methodology (RSM), which allows several input variables to be varied at the same time while assessing their effects on several output responses. 41 RSM is a set of statistical tools and sequential experimentation strategies that are helpful in the development of analytical models and process optimisation. 42 By using RSM, adequate multi-response models can be generated and the operating parameters are easily optimised. As a sub-model of RSM, Central Composite Design (CCD) is a commonly applied experimental design which helps to build a second-order response surface model requiring fewer experimental sets, and to assess the effects of different input variables on the selected responses. 29,42 In the present work, full central composite design (FCCD) with orthogonal backward elimination procedure (OEB) were performed to design experiments and maximise simultaneously the COD, BOD 5 , and turbidity removal efficiency by EC treatment. In addition, analysis of variance (ANOVA) at significance level (SL) of 95 % was performed to select the significant coefficients and eliminate insignificant parameters from developed models. The main goals of the study were as follows: • Reduction of EC process operating costs by restricting the number of experiments; • Development of a mathematical model that can accurately predict the maximum removal of contaminants and optimise overall process efficiency.
• Provide a better understanding of the relationship between key process inputs and outputs. • Assessment of the effects of four quantitative factors (reaction time, initial pH, current density, and stirring speed) on the responses under study (COD, BOD 5 , and turbidity). • Promote an environmentally friendly process for municipal wastewater treatment.

Characterisation of leachate
The leachate for this study was collected from Laghouat city landfill (400 km south of the Algerian capital at coordinates 33°42'19" N and 2°54'37" E) in a closed plastic container, and stored in obscurity at T = 4 °C. All analyses were conducted in triplicate. BOD 5 was determined using the Warburg method (WTW, Expotech, USA), and COD was measured using the SpectroScan 80DV Visible-UV system (Biotech Co. Ltd., UK). Conductivity, turbidity, and pH measurements were achieved using a HANNA HI 8733 (HANNA instruments, India), WTW Turb 555 IR (Expotech, USA), and an Adwa AD 110 (Adwa Instruments, Hungary Kft.), respectively.

Experimental setup
EC studies were conducted at room temperature (28 °C) using an EC laboratory cell in batch mode with a total leachate sample volume of 400 ml. Electrodes were attached to a direct current (DC) power supply (PHYWE System GmbH & Co. KG, Germany, 0-2A, 0-12V), and a magnetic mixer (AREX, VELP Scientific Inc., USA) was used for gentle agitation. The reactor included aluminium (Al) electrodes (0.3 × 2.4 × 6.1 cm) coupled as anode and cathode in monopolar parallel mode. The overall effective area and electrode spacing was 34.5 cm 2 and 3.1 cm, respectively. The electrodes were submerged in HCl solution after each test run, and washed with deionised water. The variables studied were reaction time (x 1 ), pH (x 2 ), current density (x 3 ), and magnetic stirring speed (x 4 ). By adding an adequate amount of sodium hydroxide (NaOH) and hydrochloric acid (HCl) solutions (0.1 M), the pH of the measurements was adjusted to the targeted values. The supernatant was extracted and examined after 60 min of settling time.

Process modelling and data analysis
In this study, several regression models, including linear polynomial and Quadratic polynomial, were investigated. Full central composite design (FCCD) comprising 31 runs was applied to estimate the effects of major operational parameters, and to improve the removal efficiency of COD, BOD 5 , and turbidity from leachate. In order to determine which significant terms must be retained in the empirical models, orthogonal backward elimination procedure (OEB) at 95 % SL was performed. This approach begins with all the variables involved in the model, and the variables are then excluded one at a time. For each level, the variable that is omitted is the one that results in the lowest inflation in the residual sum of squares. This elimination proceeds until only one element is left in the model or until the stop rule is fulfilled. 43 Subsequently, ANOVA was conducted to define the most appropriate factors for the designated model. The experiments were conducted using the Minitab Software Trial (Version 19, Minitab Inc., State College, Pennsylvania). The defined values of the chosen variables (Table 1) are defined based on a significant number of tests. Each variable was coded at five levels.
Polynomial models of second order can be presented in accordance with Eq. (1) to forecast the best settings, and consider the relationship between the parameters. (1) where β 0 is the constant coefficient; β i , β ij and β ii are the linear, interactive, and quadratic regression coefficients, respectively; k is the number of parameters; x i and x j are independent variables, and is the error term. 29 The overall performance of the fitted models was measured by the coefficient of determination (R 2 ), adjusted R 2 (R 2 adj), and predicted R 2 (R 2 pred). The R 2 coefficient was determined to check the fitting of the model. The closer the values of R 2 to 1, the stronger the model and the better the prediction of the response. For good fit of models, the coefficient of correlation (R 2 ) should be at least 80 %. 8 The prediction error sum of squares (PRESS) was used to assess the predictive ability of models. Generally, the lower the PRESS value, the better the model predictive capability. Standard deviation S evaluates how well the responses are described by the model. Moreover, the Bayesian Information Criterion (BIC) and the Akaike Information Criterion (AICc) parameters were also used for comparison of several models. It is particularly appropriate when comparing mixed-effects of nonlinear models. Smaller values to these statistics are preferred. 44 Model terms were evaluated with 95 % confidence level. For small probability P-value, the corresponding model and the individual coefficient become more significant. 45 The data variation around the adjusted model is defined by the lack of fit. If the model matches the data correctly, the lack of fit must be higher than 0.05. 46 The Pareto chart was used to measure the influence of the independent variables, as well as the relationship between them, in which any effects with an absolute value greater than 2.1 are considered significant. Statistically, the red line in the graph demonstrates the level of the low relevant term. Three-dimensional plots were illustrated, and depending on the key factors in the overlay graph, the optimum area was determined. Finally, to validate models, triplicate confirmatory tests were performed using the optimal conditions generated from statistical optimisation based on the high desirability that RSM implied. By comparing the experimental data and the expected results, the model's accuracy and suitability were assessed.
3 Results and discussion

Characteristics of the landfill wastewater
The main characteristics of the raw leachate, such as COD, BOD 5 , conductivity, pH, and turbidity were analysed. Leachate was heavily polluted. It had a high pH value of approximately 8.1, a high turbidity of 283 NTU, and a substantial content of dissolved substances (conductivity of 33.1 mS cm −1 ). Biodegradable organic matter reached concentrations of up to 1300 mgO 2 l −1 , and values of COD were around 20011 mgO 2 l −1 . Furthermore, leachate had a comparatively very low biodegradability ratio BOD 5 : COD of 0.064 (< 0.1).

Statistical analysis, optimisation, and model validation
ANOVA was employed to examine the connection between input variables and COD (Y 1 ), BOD 5 (Y 2 ), and turbidity (Y 3 ) removal efficiencies for the EC process. In order to achieve an appropriate fit in a given model, significant model parameters are required. The FCCD provided in Table 2 permitted the generation of mathematical equations in which expected responses were related to the four factors (x 1 , x 2 , x 3 and x 4 ). Using coded factors, Eqs. (2-4) of the estimated models are as follows: ( )    Fig. 1 shows the Pareto charts for COD, BOD 5 , and turbidity removal. A grey bar indicates a term that is not included in the model. The key variables, , and x 4 (D) along with the interaction effect of the AC, AB, BC, BD, and AD extending outside the red line are statistically significant at 95 % confidence level. Fig. 1a shows that the coefficient CD (x 3 x 4 ) was not significant on COD removal, so it was not included in the estimated model. The quadratic effect of stirring speed (CC) had the largest influence on COD removal, as can be seen from Fig. 1a. Moreover, Figs. 1b and c show that the coefficients CD (x 3 x 4 ) and AA (x 1 x 1 ) were not significant on BOD 5 and turbidity removal, so they were excluded from the corresponding models. Reaction time (A) and initial pH (B) had the most substantial effect on BOD 5 and turbidity removal, respectively.
Normal probability plots were used to identify the actual effects (Fig. 2). Based on the normal probability graphs, the points placed along the adjusted line were not found to be important, although the points located far out of the line appear to be most important. All quadratic effects are important, and those represented with the blue triangle are irrelevant. As may be seen in Fig. 2, the key factors (t, pH, I, and V) are distant from the diagonal line and were therefore considered important. Since its point appears furthest from the line, quadratic effect current density*current density (CC) had the highest impact. The next major element was time*current density (AC). As far  as interacting terms are concerned, AC, AB, and BC had an important influence on the effectiveness of COD removal. Such interaction effects showed that the percentage of removal performance was higher at higher values of time, stirring speed, current density, and low pH.
From Fig. 2b, all the main factors: time reaction (t), pH, current density (I), and stirring speed (V) on the right had a positive effect on the BOD 5 removal. Fig. 2c shows that B on the left had a negative influence, whereas A, C, and D on the right had a positive impact on turbidity removal. The quadratic term pH*pH (BB) had the greatest effect.
The predicted vs. experimental values plots for the responses under study are shown in Fig. 3a-c. It may be seen that all the points are close to the diagonal line, implying good agreement between the experimental data and the expected output data of the model. The coefficient of correlation of 0.9922, 0.9996, and 0.9998 highlights the good reliability of the developed models for COD, BOD 5 , and turbidity removal. Furthermore, a normal probability graph of residuals is considered to assess the validity of the models. As seen in Fig. 4a-c, the points in the normal probability plots of residuals for COD, BOD 5 , and turbidity removal fit well in a straight line, so the residuals are normally distributed. Low dispersion of the points from the reference line indicates high quality of the models.
In order to assess the statistical importance of the regression coefficients, the analysis of variance (ANOVA) was performed for experimental results, presented in Table 3.
The ANOVA of the regression model indicates that the model's terms are highly significant with a very low P-value (probability value < 0.05). As reported in Table 3, the R 2 values indicate that the models as fitted explain 99.90, 99.96, and 99.97 % of the COD, BOD 5 and turbidity variances, respectively. The adjusted R 2 (R 2 adj) should be sufficiently close to R 2 value (R 2 -R 2 adj <5 %) to indicate that undesired input factors toned not be excluded from the equation. 47 The R 2 adj values, which are more suitable to compare models with different numbers of independent variables, were 99.82, 99.93, and 99.95 % respectively. Predicted R 2 (R 2 pred) defines how well the model predicts the output variables for a new set of data. R 2 pred values should be consistent with R 2 adj as an indicator of the regression model's predictive ability. 48 The larger the R 2 pred, the more precise the predictability of the model (R 2 -R 2 pred < 0.2). For comparing models, the latter is more relevant than R 2 adj as it is focused on observations that are not involved in the model. 49 The estimated values of R 2 pred expected the developed model to explain about 99.59, 99.89 and 99.94 % for COD (Y 1 ), BOD 5 (Y 2 ), and turbidity (Y 3 ), respectively, of variability in forecasting new findings.
In accordance with these criteria, the overall predictability appeared to be quite satisfactory compared with previous results presented in refs. 45,46 . Table 3 were applied to assess the accuracy of various models when employing the method of backward elimination. 44 Higher R 2 values and As revealed in Table 3, the standard deviation S of the data points all over the adjusted results were 0.24, 0.17, and 0.22, respectively. The multiple regression was restricted to reduce the expected error sum of squares (PRESS) for each individual process. The statistical analysis showed that the fitted quadratic models depicted in Eqs. (2-4) described fairly well the process under study. The fitness requirements for the quadratic equations as well as the alternative linear models were verified. The best description of data analysis for developed quadratic equations is reported in Table 3. The findings on R 2 , R 2 adj, R 2 pred, PRESS, S, AICc, BIC, and PLOF showed that the quadratic polynomial models were adequately adjusted to establish a significant connection between the COD, BOD 5 , and turbidity removal effective-ness and the influencing factors (i.e., current density, time, stirring speed, and pH), and therefore to predict accurately the process outputs.

The AICc and BIC values shown in
Three-dimensional (3D) surface plots of the COD, BOD 5 , and turbidity removal were generated via the RSM equations. They indicate the effects and relationship of two predictor factors on the response, while the other two factors were kept unchanged within the experimental ranges. As shown in Figs. 5-7, the maximum removal of COD, BOD 5 , and turbidity was found at a lower pH range between 4.5 and 6, and current density less than 523 A m −2 . The increase in COD and BOD 5 removal (Figs. 5 and 6) at this level of pH (acid medium) can be clarified by the transformation of Al 3+ into soluble monomeric species such as Al(OH) 2+ and . By complex precipitation kinetics, these species are converted into insoluble Al(OH) 3 flocs. The generated amorphous Al(OH) 3 can absorb ions and even soluble organic compounds and/or trap colloidal particles, which then coagulate to form particles that precipitate, usually near neutral pH. 51 With these conditions, turbidity removal had also increased dramatically, as shown in Fig. 7. It was also standardised residual  observed that the removal efficiency of pollutants had increased with increasing current density. At a higher current density, the amount of oxidised metal increased producing more hydroxide flocs to remove the contaminants. Moreover, the density of the bubbles increased and their size decreased with the increase in current density, which led to a considerable removal of contaminants, mainly: COD, BOD 5 , and turbidity. In addition, an increase in current density increases the operating voltage and exacerbates the release of oxygen. 41,52 Therefore, for current density values above 523 A m −2 , a small irrelevant decline occurs. Therefore, it is suggested that the current density should be limited in order to avoid the development of excessive oxygen and remove other side effects such as heat generation. 53 Similar observations were obtained by Holt et al. 54 Damaraju et al. 55 found that charges tend to neutralise at lower pH levels as higher metal quantities are decomposed due to higher current density, leading to floc formation and increased removal efficiency. Furthermore, this result shows that the rate of removals is enhanced when electrolysis time and stirring speed increases at the same time, which enables a longer catalytic reaction time leading to an improvement in the degradation of contaminants and a larger amount of hydroxyl radicals, which allows the production of metal-polymer complex to further improve the removal percentage. 56 20 20    Electrolysis time is another effective factor in the EC process. When the interaction effect between the initial pH and electrolysis time is examined, it is observed that the increase in electrocoagulation time increases the efficiency of COD, BOD 5 , and turbidity removal. If the electrolysis time increases, the amount of aluminium hydroxide flocs increases. Namely, the formation of flocs in appropriate and sufficient quantities is time-dependent in the electrocoagulation process specifically, the formation of flocs in appropriate and sufficient quantities depends on the time of the EC process. 49 Therefore, maximum removal rates higher than 89 % for COD, BOD 5 , and turbidity were achieved within the current density range of 350-500 A m −2 , stirring speed of 140-160 rpm, pH of 4.5-6, and electrolysis time of 60-75 min.
Based on the surface response methodology and the desirability functions, numerical optimisation was conducted to determine the combination of optimal parametric settings, which concurrently increase COD, BOD 5 , and turbidity removal.
The optimal operating values of all responses can also be defined by overlaying their shapes in an overlay plot wherein the optimum region against the grey zone, which did not follow the requirements, is clearly displayed in blank (Fig. 8).
The degree of desirability varied from 0 to 1, where zero (0) implies an unacceptable response, and one (1) is the most desired response. 57 Table 4 were very close to those of the predicted results, which indicates that our models are good and reliable. Leachate is subject to clear black to light yellow discolouration following treatment.
Operating costs. The overall running costs were estimated using Eq. (5).  The electrical energy consumption (EN C ) of the process and the aluminium electrode consumption (EL C ) can be calculated using Eqs. (6) and (7). 58 where U is the applied voltage (V), I is the current (A), t is the EC reaction time (s), and V is the volume of leachate tested (m 3 ), M w is the molecular mass of the aluminium electrode (26.98 g mol −1 ), F is Faraday's constant (96487 C mol −1 ), Z is the number of electrons transferred (Z Al = 3), a is electrical energy price (0.014 US$ kW −1 h −1 , according to the Algerian market in April 2020), and b is electrode material price (1.6 US$ kg −1 Al −1 ). The dissolved amount of electrodes was determined experimentally by calculating the difference between the sacrificial anode before and after each trial. 59 The power consumption and average dissolved mass of the anode were about 52 kwh m −3 and 0.13 g l −1 , respectively. The estimated operating cost of COD, BOD 5 , and turbidity removal from leachate by EC process at optimum conditions was calculated to be 0.76 US$ m −3 leachate, which shows the viability of treatment of landfill leachates.
The findings of EC process efficiency and optimum operating conditions for leachate treatment observed in peer-reviewed literature and the current study are given in Table 5. As foreseen, these results demonstrate the good reproducibility of the proposed model in the design range.

Conclusions
In this research, FCCD with orthogonal backward elimination procedure (OEB) was applied to define optimum experimental parameters for EC leachate treatment, and to examine the effects of linear, interactive, and quadratic variables (pH, current density, stirring speed, and EC reaction time) on studied responses. Developed model results indicated a suitable relationship between experimental and expected values. ANOVA revealed a high coefficient of determination value (R 2 ˃ 95 %), indicating a reasonable adjustment of the quadratic regression equation with the actual results. Under optimal operational factor settings of electrolysis duration 74.6 min, pH 5.04, current density 407 A m −2 , and stirring speed 150 rpm, the removal Orthogonal backward elimination procedure efficiencies of COD, BOD 5 , and turbidity from leachate reached 90 %, 92.3 % and 99.6 %, respectively. The findings showed that RSM and OEB were successfully exploited to optimise the operating parameters of EC treatment using small dissolved amounts of aluminium electrodes, at low operating cost of about 0.76 USD/m 3 of leachate treated.

List of abbreviations and symbols
AICc -Akaike information criterion BIC -Bayesian information criterion