Introduction

Humic substances are the products of biological and chemical decomposition of living organisms. They are found in abundance in black water and swampy areas, causing dark brown and yellowish colour in water. The presence of humic substances in water can have a significant impact on the treatability of the water and the efficiency of chemical disinfection process. They are believed to be the precursors to the formation of carcinogenic disinfection by-products. A previous study revealed that water tainted by peat swamp leachate contained a higher quantity of trihalomethanes compared to water from non-peat water sources [1].

In the conventional water treatment system, humic substances are removed using the coagulation and flocculation approach [2–4] requiring excess amount of coagulant and are generally limited by high operational and material costs. Other adsorbents such as activated carbon, zeolite, and mesoporous silica have also been employed for isolation of humic substances but these adsorbents are very expensive. Considering the cost and huge quantity of water treated, these adsorbents are less attractive. Biosorption is an alternative gaining increasing attention for its advantages of low cost and abundant availability. Sim et al. [5, 6] examined the adsorption ability of a spectrum of agricultural biomass in removing humic substances, and reported coconut copra with promising adsorption capacity. The biomass is relatively richer in carboxyl functional groups. The untreated coconut copra was evidenced to remove an average of 50 % of humic substances from peat swamp runoff, higher than other selected biomasses including banana trunk, coconut husk, empty fruit bunch, groundnut shell, rice husk, sago waste, saw dust and sugarcane bagasse with removal efficiency ranging between 11 and 40 %, based on the spectrophotometric approach. Upon treatment with citric acid, the adsorption capability of coconut copra is greatly improved [6]. The heterogeneous nature of biomass and humic substances has allowed various interactions such as electrostatic linkages, cation bridging, hydrophobic interaction and hydrogen bonding to take place [7–9]. Coconut copra is also potentially viable for removal of heavy metals i.e., cadmium, chromium and nickel with encouraging removal efficiency of 85–90 % [10]. This offers an environmental friendly and economically sustainable option for water and wastewater treatment.

It is very important to optimize the process conditions, minimizing the operational cost whilst maximizing the adsorption performance. According to Sim et al. [6], the adsorption performance of coconut copra for humic substances was optimized using the conventional one-factor-at-a-time strategy; this involves changing a single factor whilst maintaining other factors at constant. This approach does not demonstrate the combined effect of all factors; in addition, it is tedious and provides insufficient information; often time, the optimization point could have been missed [11]. The traditional full factorial design can be used to examine the interaction effects nonetheless this usually involves a great number of experiments. For example, a typical three-factor-five-level full factorial design would require a total of 125 experimental runs while a corresponding central composite design (CCD) only involves 20 runs. CCD incorporates full or fractional factorial design with the centre and star points where each factor is mapped onto five-level allowing evaluation of the second-order response surface [12–15]. This design has been widely employed for process optimization involving dye removal [16–18], antibiotic production [19] and humic acid and heavy metal adsorption [20, 21]. Cerino Cordova et al. [22] conclude that it is necessary to study the interaction effects of various parameters on the optimal biosorption process. Therefore in this paper, inscribed CCD and response surface methodology were employed to examine the optimum conditions for removal of humic substances from peat swamp runoff using coconut copra. Note that this paper focuses on examining the optimum conditions for removal of humic substances when multiple variables are considered simultaneously; the information derived is particularly important for scaling up operation. The characteristics and adsorption isotherm studies of raw and modified coconut copra have been reported in Sim et al. [6].

Materials and methods

Sample preparation

Coconut copra in shredded form was collected from the local wet market and washed thoroughly under running tap water to remove coconut milk and foreign matters. The coconut copra was oven dried at 105 °C. Coconut copra was then refluxed at 100 °C for one hour, washed, oven dried and subsequently agitated with 0.5 M citric acid for one hour at room temperature. The treated coconut copra was washed to pH 3 and oven dried [6].

Adsorption of humic substances

Peat swamp runoff, rich of humic substances, was collected from Asa Jaya Vicinity. Asa Jaya, located within Kota Samarahan Division, is surrounded with peat land and river is highly coloured. It was used to demonstrate the adsorption performance of treated coconut copra as the function of dosage, contact time and temperature in removing humic substances according to CCD. The adsorption was demonstrated on the ‘real world’ humic substances as the model of standard humic acid is incomparable to the naturally occurring humic as revealed [9]. Prior to adsorption test, peat swamp runoff at pH 4–5 was filtered through 0.45 µm membrane filter to remove suspended particles and foreign matter with no pH adjustment. The adsorption efficiency was examined based on the absorbance at 465 nm [23] using a UV–Visible spectrophotometer (Jasco V-360 Spectrophotometer). Hypothetically, the absorbance increases with the concentration of humic substances. When adsorption has taken place, reduction in absorbance is anticipated. The removal efficiency is calculated as follows.

$${\text{Removal Efficiency}} = \frac{{{\text{Abs of untreated water}} - {\text{Abs of treated water}}}}{\text{Abs of untreated water}} \times 100\,\%.$$

Experimental design

The CCD of three factors involving dosage (x 1), contact time (x 2) and temperature (x 3) is shown in Fig. 1. The design consists of the factorial design with centre points and the star/axial points used to estimate the curvature. This allows the quadratic terms to interfere in the model for estimation of the optimum combination of factors. The total number of experiment is calculated based on n = 2f + 2f + n where f is the number of factors and n is the replication of centre point [24]. The α value is defined as the distance between axial points and centre point, \(\alpha = \left( {2^{f} } \right)^{\frac{1}{4}}\). As inscribed CCD is a scaled down of circumscribed CCD, the factorial design is further divided by α, demonstrating a 1/α coded level in factorial design. Hence, coded value of 0.5946 is generated from 1/1.6818. Table 1 summarizes the coded and uncoded level of the selected factors where the response is represented by the percentage removal of humic substances (Table 2). The CCD is more powerful for that the conventional fractional factorial designs can only be used to investigate the effect of a single variable on the response, unable to illustrate the optimum point.

Fig. 1
figure 1

The central composite design

Full size image
Table 1 The coded and uncoded level of three selected factors
Full size table
Table 2 The experimental and the predicted response according to linear, interaction and quadratic models
Full size table

Response surface methodology (RSM)

The design matrix involving linear, interaction and quadratic terms are employed to develop the equations for estimation of the adsorption performance. The variable coefficients were derived based on multiple linear regression: \(\hat{\varvec{y}} = \varvec{D} \cdot \varvec{b},\) where the predicted response is expressed as the product of the design matrix, D, and the corresponding coefficients, b [25]. The vector of the corresponding coefficients is the pseudo inverse of D and the experimental response, \(\varvec{b} = \left( {\varvec{D^{\prime}} \cdot \varvec{D}} \right)^{ - 1} \cdot \varvec{D^{\prime}} \cdot \varvec{y}\). The coefficients indicate the behaviour of different factors on the response where various models can be described. The linear model:

$$\hat{y} = b_{1} x_{1} + b_{2} x_{2} + b_{3} x_{3} + k.$$

The interaction model:

$$\hat{y} = b_{1} x_{1} + b_{2} x_{2} + b_{3} x_{3} + b_{12} x_{1} x_{2} + b_{13} x_{1} x_{3} + b_{23} x_{2} x_{3} + k.$$

The quadratic model:

$$\hat{y} = b_{1} x_{1} + b_{2} x_{2} + b_{3} x_{3} + b_{12} x_{1} x_{2} + b_{13} x_{1} x_{3} + b_{23} x_{2} x_{3} + b_{1} x_{1}^{2} + b_{2} x_{2}^{2} + b_{3} x_{3}^{2} + k,$$

where ŷ = predicted response; x 1, x 2 and x 3 = coded factors; b = coefficients; k = offset term.

For a design with three or more levels, a curvature is observed. The response surface plots were generated using MATLAB version R2012a to enable modelling of the response over an entire range of varying factors with limited number experimental runs. Analysis of Variance (regression ANOVA) was employed to determine the model fitness and significance of various factors on the removal efficiency considering the sum of squares and residual sum of squares. ANOVA was in addition performed by differentiating the variances of linear regressions. All statistical analysis was performed using SPSS version 21.0.

Model quality

The regression coefficients, R 2 and adjusted R 2, were calculated to determine the variation between the model and the experimental data [22]. According to Weinberg and Abramowitz [26], R 2 often overestimates the model fit with excessive predictors therefore adjusted R 2 is recommended. The adjusted R 2 takes all independent predictors into consideration and fixes the R 2 more accurately according to the number of predictors present in the model. R 2 usually increases with increasing predictors; however, adjusted R 2 will only increase when that particular predictor has positive effects on the response. Thus, the latter is particularly useful in comparing models with different numbers of predictors [27]. Both R 2 and adjusted R 2 indicate the quality of the proposed models; values closer to 1.0 demonstrate a good agreement between the experimental and predicted responses [16]. The root mean square error (RMSE) and percentage root mean square error (% RMSE) are also calculated to indicate the predictive performance [25, 28]. The lower is the value of RMSE, the better is the model for prediction.

$$R^{2} = 1 - \frac{{\mathop \sum \nolimits_{i = 1}^{I} (y - \hat{y})^{2} }}{{\mathop \sum \nolimits_{i = 1}^{I} (y - \bar{y})^{2} }}$$
$${\text{Adj}} R^{2} = 1 - \frac{{\left( {1 - R^{2} } \right)(n - 1)}}{n - P - 1}$$
$${\text{RMSE}} = \sqrt {\frac{{\mathop \sum \nolimits_{i = 1}^{I} (y - \hat{y})^{2} }}{n - P}}$$
$$\% {\text{RMSE}} = \sqrt {\frac{{\mathop \sum \nolimits_{i = 1}^{I} (y - \hat{y})^{2} }}{n - P}} \times \frac{100 \times n}{{\mathop \sum \nolimits_{i = 1}^{I} y}},$$

where n = number of experiments; P = number of predictors.

Results and discussion

Dosage, contact time and temperature are three major factors governing the adsorption performance of biosorbents [17, 18]. With CCD, these factors can be systematically optimized taken into account of their interaction effects. The linear, interaction and quadratic models were derived based on CCD where the magnitudes of coefficients suggest the contribution of various factors on the response. It is evidenced that dosage exerts greater influence on the removal efficiency in all models. Linear model:

$$\hat{y} = 21.752 x_{1} + 2.331 x_{2} + 2.284 x_{3} + 71.357.$$

Interaction model:

$$\hat{y} = 21.752 x_{1} + 2.331 x_{2} + 2.284 x_{3} - 1.654 x_{1} x_{2} + 1.586 x_{1} x_{3} - 0.470 x_{2} x_{3} + 71.357.$$

Quadratic model:

$$\hat{y} = 21.752 x_{1} + 2.331 x_{2} + 2.284 x_{3} - 1.654 x_{1} x_{2} + 1.586 x_{1} x_{3} - 0.470 x_{2} x_{3} - 14.166 x_{1}^{2} - 3.889 x_{2}^{2} - 5.912 x_{3}^{2} + 79.072.$$

Table 2 summarizes the predicted response according to different models. The R 2 value calculated for linear, interaction and quadratic models are 0.825, 0.827 and 0.973, respectively. Meanwhile, the corresponding adjusted R 2 values are 0.793, 0.748 and 0.949 implying that the adsorption process is more suitably described with the quadratic model.

The deviation between observed and predicted percentage according to linear, interaction and quadratic models were 0.63–28.76, 1.53–28.76 and 0.15–7.74 % suggesting that the quadratic model has the lowest error. RMSE serves as a measure of predictive power based on difference between observed and predicted values. The RMSE calculated according to linear, interaction and quadratic models are 5.39, 5.91 and 2.62, respectively. These values can be converted into % RMSE in turn to indicate the closeness of the predicted and experimental responses. The corresponding % RMSE are 7.35 % (linear), 8.06 % (interaction) and 3.57 % (quadratic). This suggests that the quadratic model is best fitted for estimation of the removal efficiency. Noticeably, the interactions terms are relatively smaller, indicative of less influence on the adsorption.

The fitness of the model was examined using regression ANOVA [17]. Under the null hypothesis, the model has no predictive capability. The p values obtained for linear, interaction and quadratic models are consistently <0.05 (Table 3) suggesting that these models are capable of predicting the response. Table 4 shows the estimated regression coefficients and the corresponding t and p values; all linear, interaction and quadratic terms are characterized by p > 0.05 except x 1 and x 21 indicating that dosage has a significant effect. Factor x 3 demonstrates non-significant linear term but significant quadratic term. This is possibly due to the response relies more on the quadratic form rather than linear form or in other words, x 3 has a quadratic effects towards the response. Figure 2 illustrates the interaction plot of dosage, contact time and temperature on the removal efficiency of humic substances. Typically, if there is no interaction, the profile of respective factor would appear parallel.

Table 3 ANOVA regression models for removal of humic substances
Full size table
Table 4 Estimated regression coefficients and the corresponding t and p values
Full size table
Fig. 2
figure 2

The interaction plot of dosage, contact time and temperature

Full size image

Figure 3 illustrates the three-dimensional response surface plots of removal efficiency against two factors. The surface plots of (dosage vs temperature) and (dosage vs contact time) indicate that at higher dosage, the adsorption process is favoured attaining elevated removal efficiency possibly due to the increased binding sites. The optimum dosage is found at 4.56 g/100 mL. The removal efficiency is in addition improved with increased temperature and contact time, though not as significant as the dosage, corroborating the coefficient values derived for respective factors based on the quadratic model (b dosage: 21.75; b contact time: 2.33; b temperature: 2.28). According to Yasin et al. [21], higher temperature may facilitate adsorption by speeding up the collision of particles and lowering the mass transfer energy. The optimal temperature identified for maximum removal percentage is 56.8 °C/329°K, beyond which the removal efficiency begins to decline. This is possibly associated to the exothermic characteristic where desorption process is anticipated to take place [17, 29]. The adsorption process is enhanced at longer contact time nevertheless the process appears to slow down after 43 min when the equilibrium is reached [17].

Fig. 3
figure 3

The three-dimensional response surface plots of a response versus dosage and temperature; b response vs contact time and dosage; and c response vs temperature and contact time

Full size image

The maximum removal efficiency of 88.19 % is attained when 4.56 g of coconut copra is agitated in 100 mL of peat swamp runoff for 42.9 min at 56.8 °C/329°K. The optimum conditions are experimentally verified where an average of 86.54 % of removal is achieved corresponding to percentage error of 1.87 %. Yasin et al. [21] reported 73.46 % removal with humic acid standard solution using anionic clay hydrotalcite under optimum conditions.

Conclusion

Coconut copra is a potential biosorbent for water and wastewater treatment. The quadratic model involving dosage, contact time and temperature is best fitted to predict the efficiency of coconut copra in removing humic substances from peat swamp runoff. Among these factors, dosage exhibits relatively greater influence. The optimal conditions necessary to attain maximum removal efficiency (88.19 %) are dosage (4.56 g modified coconut copra per 100 mL peat swamp runoff), temperature (56.8 °C/329°K) and contact time (42.9 min). The experimental response acquired (86.54 %) based on the modelled optimum conditions is in good agreement with the predicted response. Coconut copra adsorbed with humic compounds can be potentially recycled into soil conditioner due to its beneficial agricultural value. Lee et al. [30] revealed that biomass fortified with humic compounds is a viable soil conditioner for improving plant growth.