Optimization and Modeling of Glyphosate Removal by Nanofiltration at a Pilot Scale, Using Response Surface Methodology

Tables index

Veiw figure View Table

2.2. Acute Toxicity Tests in Fish

Glyphosate commercial formulation toxic effects were evaluated by an Acute Toxicity Test which uses fish as experimental model. It is a common technique recommended by U.S. Fish and Wildlife Service Columbia National Fisheries Research Laboratory. The tests were conducted with fish of the poecilia reticulate species reproduced in our laboratory to a size of 0.7-1 cm. For the test, 10 specimens were exposed for a 96 h period in 2 L containers with 1 L of solution. The test evaluated dead fish number in each container after the 96 h period. Dead fish were removed every 24 h. The minimum commercial formulation dose which caused total mortality (100%) contained 38.4 mg/L of glyphosate IPA salt (see Table 2). Fish mortality did not occur when used pure glyphosate acid in concentrations over 100 mg/L. Toxicity tests in NF permeate were performed following the same protocol.

Table 2. Acute toxicity test of the commercial formulation of glyphosate

Download as

Tables index

Veiw figure View Table

2.3. Experimental Design

The selected factors and their corresponding experimental levels were determined based on previous experiences ^[14]. The experimental design used for this investigation was Box-Behnken type for three levels and three factors (13 runs) with two replications in the centre of the plan, considering glyphosate rejection as a response (R_g %) ^[20]. Table 3 shows the levels of the actual and coded factors

Table 3. Factors levels and coded values used in the experimental design

Download as

Tables index

Veiw figure View Table

2.4. Nanofiltration Experiment

The experiences were carried out in a nanofiltration pilot plant described in a previous work ^[19]. The operating temperature was 293 K. The feed flow was fixed at 417 L/h high enough to prevent concentration polarization. The membrane module was operated in the cross-flow mode; both permeate and retentate were recycled. The flow rates were determined using flow meters (Cole-Parmer). Synthetic samples were prepared by adding the required amounts of commercial formulation to distilled water (pH 5.7 ± 0.2 and conductivity 1.0 S/cm). Several solutions were prepared with different concentrations (41, 132 and 221 mg/L) of glyphosate IPA salt, and then, pH was adjusted with HCI or NaOH (Sigma-Aldrich). Initially 20 L of the feed solution was charged into the feed tank and pumped for 2 h until a steady permeate flux was reached. In all runs the reduction of permeate flow was insignificant regard the permeate flow with distilled water (less than 5%). Samples of permeate and feed were collected and analyzed per triplicate and subsequently, average value was calculated. Fifteen experimental runs were performed at random order (see Table 4). At the end of the each run, an easy washing with distilled water for 1 h was enough to restore the initial permeate flux. The rejection of glyphosate (R_g %) was calculated from the following equation:

(1)

where C_p and C_f are the mean concentration of glyphosate in permeate and in the feed solution, respectively.

Glyphosate quantity present in samples was measured as phosphorus (P). The pattern was prepared with a 99% purity glyphosate acid provided by Monsanto. Phosphorus concentrations were measured by USN-ICP-OES using a Sequential Inductively Coupled Plasma Spectrometer (Baird ICP 2070) with ultrasonic nebulization.

2.5. Model Fitting

Using DOE module (STATISTICA 8.0 Stat Soft Inc.), a complete second degree model like the one shown below was adjusted:

(2)

where “b” represents regression coefficients of the model, found by the least squares method, are the factors or codified variables; and ε is the statistic error or residual.

The statistical significance test of the model was developed through Analysis of Variance (ANOVA); to perform the same test for individual coefficients Student’s t-test was used. In order to find the most suitable model for the experimental results, insignificant coefficients (p value> 0.05) were eliminated step by step, adjusting once again the model according to other parameters, until the model of minimal bias was found, where the C statistic (which is a measure of the total mean square error for the regression model) is minimum or approximately matches with the number of the p model parameters ^[21].

In order to determine the adequacy of the model, the lack-of-fit test, the graphic analysis of the residuals and the adjusted coefficient of multiple determination (R²_adj) criteria were used. The lack-of-fit test was performed by means of ANOVA, where the sum of squares due to residuals SS_R was divided in two components, sum of square due to pure error SS_PE, and sum of square due to lack-of-fit SS_LOF. A test statistic was calculated F₀ = MS_LOF/ MS_PE, which was compared to Fisher F statistic tabulated F_α_{,m-p, m-n}; if F₀< F_α_{,m-p, m-n} there was no evidence of lack-of-fit of the model, thus, it was possible to combine the mean square lack-of-fit MS_LOF and the mean square pure error MS_PEto estimate the residual variance .

Since the hypothesis testing procedures used to determine the statistical significance of the model parameters assumed that the were normal and independently distributed with media zero and a constant variance , it was necessary to verify that no residuals or error assumptions were disrupted. This was made by analysing the following plots: the normal probability plot of the residuals, the estimated values versus the residuals and residuals versus the order followed by the experimental runs. Finally, the tridimensional response surface and its contours (constant response curves) were drawn from the adjusted model. From this plot, the optimum point was directly located and the effect of the factors [Gly], pH and TMP on R_g% response was analysed.

3. Results and Discussion

Table 4 shows the R_g% values for each run of the Box-Behnken experimental design. The high rejections of glyphosate (≥98%) despite its molecular weight below the molecular weight cut-off (MWCO) of the membrane were related to the combined effect of Donnan Exclusion and Dielectric Exclusion ^{[22, 23, 24]}. The high hydrophilicity of glyphosate (hydration of the molecule) also contributed to retention ^[25].

Table 4. Experimental design for Rejection Glyphosate (Rg %)

Download as

Tables index

Veiw figure View Table

Table 5. Effects estimated for Rg % (complete second degree model)

Download as

Tables index

Veiw figure View Table

Table 5 shows the estimated effects for the three studied factors. The pH (X₂) had the major impact on rejection, followed by [Gly] (X₁), showing statistically significant linear and quadratic effects, as well as interactive effect (X₁.X₂). Even though, TMP (X₃) linear effect was statistically significant (p_value < 0.05), however it was very small, therefore being non-relevant from the practical point of view. On the other hand X₁.X₃; X₂.X₃ and X₃² interactive and quadratic effects were not statistically significant.

Several authors have reported that the pH has significant effects on the membrane charge and solute speciation ^{[26, 27, 28]}. The pH increase not only rose the negative charge density of the membrane by a major ionization of it, but also it produced a change in the ionic specie of glyphosate that also increased its negative charge. Figure 1 shows the fraction of the different glyphosate ionic species as a function of pH. For example, at a pH below 6, there were two ionic species present H₂G⁻² and HG⁻³ with two and three negative charges respectively and in a 50% (F = 0.5) each one. At a pH = 7.5 the present specie in a 100% had three negative charges. The increased charge density of the membrane and the negative charge of glyphosate incremented the electrostatic repulsion. All the above mentioned explained the remarkable effect of pH on rejection shown by the model. Song et al., 2013 have observed the same behavior in the recovery of glyphosate from saline wastewater using hollow fiber nanofiltration membranes ^[13].

Download as

Veiw figureFigures index

NEW
View larger figure in new window

View next figure

Figure 1. Speciation of Glyphosate in aqueous solution as a function of pH

Figure 2 shows the pH effect over Rg % for the three concentration levels that were studied. It could be seen that the rejection was greater at higher concentrations which could be attributed to prevailing dielectric exclusion mechanism on the Donnan exclusion. In a recent study, it was concluded that in the process of ion permeation through membrane, the dielectric exclusion increased when increasing electrolyte concentration ^[29].

It could also be observed two different behaviours: the first one was when the pH varied between 5 and 7.5, and the second one was when it varied between 7.5 and 10. In the first range, an antagonistic interaction between the concentration and the pH was observed (evidenced by the decline in the slope), which means that the effect of increasing the pH decreases as the concentration increases, probably due to the shielding of the fixed charge of the membrane.

Above pH 7.5 the slopes were almost similar indicating the absence of interaction which could be interpreted as the increase in the charge of the membrane and the speciation of glyphosate at such a high pH neutralized to some extent the screening effect.

The effect of TMP on the rejection of glyphosate was very small. With the increase of the TMP the retention remained nearly constant ^[13]. It is believed that the solvent flow increase was compensated with a solute flow increase, something reasonable if it is considered that the solute molecular weight is inferior to MWCO membrane.

Download as

Veiw figureFigures index

View next figure

Figure 2. Rejection % (R_g%) as a function of pH and [Gly]. Marginal means and confidence intervals (95%)

As a result of the procedure explained above (2.5. Model fitting), the following model in codified variables was ajusted:

(3)

ANOVA values shown in Table 6 indicate that the lack-of-fit of the model was not statistically significant (F₀= 9 < F_0.05,₅_,₂ = 19). The value of R² = 0.96 indicated that the 96 % of the observed variation on the response (Rg%) is explained by the model. The graphic analysis of the residuals showed that they followed approximately a normal distribution and were independent, with no pattern or rare tendency. Thus the fitted model was adequate to describe glyphosate rejection behaviour based on the factors studied.

Table 6. Analysis of Variance (ANOVA)

Download as

Tables index

Veiw figure View Table