Development of a Predictive Model for Removal of Organic Matter from Leachate Landfill by Catalytic Oxidation Using Response Surface Methodology

In this work the heterogeneous Fenton catalytic experiments were arranged in a CCF experimental design and analyzed by response surface methodology, for every peroxide concentration. These response surfaces were the representation of a quadratic predictive model developed for improving of biochemical o xygen demand at five days/chemical o xygen demand (BDO5/COD) ratio, trough the optimizat ion parameters of reaction. For this proposal 18 catalytic tests were carried out evaluating three main factors: pero xide concentration, catalyst loading and the peroxide addition rate. The results in confronting with the experimental data showed that fro m 60 minutes the catalytic wet pero xide oxidation (CWPO) react ion was stationated in focusing the chemical COD removal, inclusively in any opportunities these values were decreased maybe because new co mpounds had been formed and we supposed that these new co mpounds were with biodegradable character because this behavior was perfect ly answered by the prediction model, hence with this new and enhancent ratio there are possibilities to get in to the biological process again for the refine the effluent and co mply with the actual regulation. The best conditions predicted for the model are at 4.68M pero xide concentration, catalyst loading charge (CL) (0.5 - 0.6%) and pero xide addit ion rate (PAR) (7.5 - 10 mL/h) to imp rove its biodegradability to 0.30, thus this effluent might come back to the begin of plant fo r refin ing the treatment.


Introduction
Landfill is nowadays the main way of disposal for the enormous charge of municipal solid wastes generated all around the world [1]. Ho wever, rain and liquid percolation throughout them p rovokes the side production of significant volumes of a strongly contaminant liquid, the so-called leachate of landfill. In general terms, it is a dark liquid of offensive odor, high loading of organic compounds and complex chemical co mposition, wh ich usually includes not negligible charge of bio-reluctant compounds [2]. Catalyt ic wet pero xide o xidation CWPO is a green technology that meets the characteristics required to perform efficiently in the treatment of such highly contaminant streams. In CWPO the hydrogen peroxide is the source of the powerfu l hydro xyl radicals, whose generation can be activated by using inexpensive solid catalysts like Al/Fe -pillared clays, and the process can be efficiently carried out at very mild conditions of ambient temperature and pressure. The employ ment of a solid catalyst enables both, efficient immobilizat ion of the active metal that avoids its recovery from the effluent as one extra step, as well as the catalyst reuse along several reaction catalytic cycles [3]. One of the tools that have broughtsignificant benefits when optimizing analytical processes in which several variables simu ltaneously affect the analysis, is the experimental design based on factorial design [4] and response surface methodology [5] The aim of this work was to develop a response surface methodology (RSM) to find a model ab le to predict the optimu m parameters of react ion improv ing either, COD removal, BOD 5 /COD ratio of the output stream or both, emp loying for that purpose a central co mposite face (CCF) quadratic experimental design tool.

Leachate Characteristics
The leachate employed in the catalytic runs was perfectly described in Galeano et al [3] and all data used for this development are fro m the jo int work at the Un iversity of Nariño [6]

Experi mental Design
The catalytic tests were carried out in two batches: one batch at 2.34 mo l/ L of the pero xide concentration (PC) and another one at 4.68 mo l/ L H 2 O 2 . In Table 1 it is shown the arrangement of the 18 catalytic tests carried out by duplicate using a central composite face (CCF) quadratic experimental design. The other two factors were tested at three levels each one: catalyst loading (CL) of 0.5, 1.0 and 2.0 (g NaBVAlFe3 /100 mL leachate ) and the peroxide addition rate (PAR) of 7.5, 15 and 30 (mL/h).

Design of Experiments
Since all the trends till now found for every factor at a time have shown to strongly depend on the chosen conditions for the rest of parameters, their simu ltaneous observation was tried to get better insight of the experimental system. The central co mposite face design CCF constituted by a full or fractional factorial design and center points placed in the faces, is one class of response surface methodologies often used for fitting of second-order models in the design of experiments [7]. The CCF was used in this work, considering the minimu m and maximu m levels for H 2 O 2 concentration (2.34 -4.68 mo l/ L), catalyst loading (0.5 -2.0 wt./ V %), and rate of peroxide addition (7.5 -30 mL/h) and two responses: COD removal and BI (BOD 5 /COD ratio). Assuming a second-order polynomial model, at least 18 replicated runs (36 experiments) must be carried out to solve the mat rix, for which statistical software MODDE v6.0 was used. Each run was replicated to check the reproducibility and to evaluate the experimental error of the results obtained in the design of experiments (DOE). The 36 experiments are put together in the 18 replicated runs listed in Table 1. A random order was employed to perform the experiments in order to minimize systematic errors. As above mentioned, the targeted responses to be maximized were both, the COD removal and the BI value.  The probability values (p value) from the analysis of variance for models COD removal and BI were 0.053 and 0.002 respectively. It allowed us to conclude that quadratic model developed is statistically poor to predict COD removal but good for BI enhancement (with a 95% confidence level) and therefore is appropriate for predicting the BI response in 4 h of reaction (p < 0.05). Moreover, the determination coefficient (R 2 ), which is also included in the inspection of the fitting between the experimental data and the mathematical model is 0.60, indicat ing that the model can explain at least 60% of the targeted function variations. Finally, in the analysis of variance the F value (4.26) is higher than the value from Fisher tables (F 9,26 = 2.22, for a 95% confidence level), meaning that the variations in the response are related to the model and not to random variations. Then, BI was the only response found more or less acceptable to continue with a model formu lation, in order to estimate the mo re pro mising levels of the factors required to imp rove the performance of the CWPO reaction applied on the leachate of landfill. The unacceptable fitting observed for COD removal response probably arose to above discussed order kinetics change in peroxide react ion, taking place in the range o f H 2 O 2 /Fe (s) mole ratios here studied, which could not have been correctly accounted by the model. The very important role of the H 2 O 2 /Fe(II) mo le ratio on the performance of Fenton homogeneous systems for treatment of leachates of landfill has been pointed out by Zhang et al. [8] short time ago, but also underlined more recently by emp loying a multivariate approach [9]. Besides, the implicit blind effect of contaminant adsorption on the clay catalyst probably contributed too. Likewise, the more acceptable fitting for BI was observed, probably because regardless the kinetics fo llo wed by the peroxide reaction it anyway led to improve b iodegradability either, via BOD 5 enhancing in the not-activated pathway or COD removal by the radical med iated one. Thus, using the above-mentioned software the coefficients of the BI predicting quadratic model in the polynomial expression were calculated by mu ltiple nonlinear regression analysis, resulting in the following regression  Figure 1 shows the predicted BI values fro m this equation as compared to the experimental data at 4 h of reaction. Fro m this plot, it can be seen that the values predicted by the second-order model agree reasonably with the experimental data, even though the simp lified equation has been used. Obviously, the data would fit better as the complete equation obtained from M ODDE software is used. Thus, the behavior of the response factor BI was predicted by a model as follows in next paragraph. As it can be seen in the response surfaces generated by the model equation (Figure 2), there is an important influence exerted by the pero xide concentration and dosage on relevant increase in the BI ratio under different conditions. First of all, while maximu m BI enhancement obtained for low peroxide concentration is slightly over 0.20, in the case of the higher peroxide concentration is close to 0.30, showing as expected that an increased dosage of oxidizing agent may lead to higher enhancement in the output biodegradability. Moreover, more remarkable seems that under low pero xide concentration (2.34 mo l/ L) BI was enhanced at high CL exactly the opposite respect to the catalyst, with the BI rising at lower CL o f 0.5 % and high PA R of 30 mL/h. It is also here noteworthy that under high peroxide concentration the BI is enhanced at low catalyst loading, almost irrespective fro m the PA R values (Figure 2 (b)). It coincides with the explanation offered in advance fo r single effects, on the basis of differing oxidation potentials of the species responsible to carry out the organic depletion either, when catalytic activation would prevail leading to radical fo r mation (high CL and/or lo w PAR) o r when direct o xidation by mo lecular H 2 O 2 could be promoted (low CL and/or high PAR). In other words, the model anticipates that when high PC and low CL values are employed the BI enhancement occurs almost exclusively due to direct action of the pero xide on the organ ic matter with no med iated catalytic activation. Meanwhile, under low pero xide concentration BI enhancement results promoted by a high catalyst loading, for sure because catalytic activated radicals play a mo re important ro le on biodegradability under such conditions by means of both, COD removal as well as higher BOD 5 enhancing.

Overall Performance
The overall performance of the heterogeneous CWPO reaction must be carefully analyzed taking into account not only the modeled predicted BI response but also the experimental results obtained for the COD removal [3]. It is evident that high catalyst loadings together with low peroxide addit ion rates and pero xide dosages seem to promote a better use of the oxid izing agent towards COD removal, whose direct consequence drives to lower costs of operation. Furthermore, as said in last paragraph the better predicted reaction parameters for BI enhancement are at first apparently the opposite of those needed to achieve the highest COD removal. However, in Figure 2 (b) can be seen that BI response surface at high PC exhib its another maximu m, though a bit less pronounced, for high CL and high PAR which can also be seen at low PC (see Figure 2 (a)). Run 18 in Table 1 meets such a pair of reaction parameters at high peroxide concentration and dosage, while shows a reasonable high COD removal close to 40 %. It indicates that modeled BI offers several possible sets of conditions to maximize its response, including some where moderate to high COD removals can also be achieved at the same time. It can be briefly summarized as follows. Although the commented lo wer intrinsic o xid izing power displayed by not-activated molecular hydrogen peroxide co mpared to hydroxyl radicals leads to increased biodegradability in the output leachate, it is not necessarily acco mpanied by a significant COD removal. Apparently, when the reaction is driven preferentially following such a pathway, there is an important BOD 5 increment but scarce COD depletion, which anyway conduces to higher biodegradability in the output stream (see runs 11 and 12 at Table 1). In the other hand, as high catalyst loadings are emp loyed, in fact if we speak about lower H 2 O 2 /Fe (s) mole ratios (Table 1) it could result more accurate, the role played by the catalytically-generated radical species becomes fairly more important. Under this scenario COD removal results clearly favored on BI enhancement (see runs 4 and 16 at Tab le 1) probably because the powerful radicals conduce to deeper o xidation and h igher degree of mineralizat ion, though generating lower fraction of by-products featuring enhanced biodegradability. Neverthel ess, it must be stressed that such behavior strongly depends also on peroxide concentration and dosage.

Conclusions
CCF experimental design was used to arrange the catalytic experiments, and a response surface methodology was applied to data, where an acceptable predict ing model was obtained only for BI response while poor fitting prevented to find out another one for COD removal response.