Chloride ion sequestration by vetiver root biosorption: isotherm, kinetic, and thermodynamic analyses and ANN prediction

Water scarcity, exacerbated by increasing salinity levels in surface and groundwater, poses a significant challenge to sustainable agricultural practices, notably in regions like India. Saline water, enriched with sodium chloride, adversely impacts plant growth. However, little research focuses on mitigating water salinity for irrigation purposes (Kaushal et al. 2021). Despite efforts to remediate soil salinity, minimal attention is devoted to decreasing salinity in saline water using organic amendments (Gunarathne et al. 2020).

Groundwater, which is a primary source of freshwater for various sectors, often contains high salt concentrations. This makes it unsuitable for irrigation and can have detrimental effects on soil conditions and ecosystems (Sharma & Bhattacharya 2017). The World Health Organization prescribes a chloride limit of 250 mg/L for drinking water purposes. The maximum permissible concentrations of chlorides in irrigation water, without causing yield losses, range from 350 to 1,750 ppm, depending on the cultivation of the crop (Thu et al. 2018). Saline water's impact on crop water absorption and the consequent risk of plant wilting, even when soil moisture further appears sufficient, further worsens agricultural challenges (Krishan et al. 2020).

While desalination technologies like reverse osmosis offer promise, their limitations – high costs, biofouling, and energy consumption – hinder widespread adoption (Sahu 2021). Alternative approaches using low-cost adsorbents, which are rich in organic compounds, have gained attention for their ability to remove ions that contribute to water salinity (Dhumal & Sadgir 2023). They are also being studied for their effectiveness in removing different pollutants from water (Nikhar et al. 2023) and their potential to contribute to sustainable water treatment approaches (Waikar & Sadgir 2023). Activated carbon, renowned for effective pollutant removal, especially from liquid phases, faces challenges like cost and process complexities (Saleem et al. 2019).

This study investigates vetiver biomass conversion into activated carbon to extract chloride ions from saline water, achieving a capacity of 17.58 mg/g at 600 °C pyrolysis. It delves into material characterization, process optimization, and kinetic evaluations. It also explores machine learning algorithms to forecast chloride ion removal rates. The aim is to establish vetiver root powder as a viable, cost-effective bioadsorbent for chloride ion elimination from saline water, particularly in agricultural applications.

The adsorbents sourced in this research were collected, cleaned thoroughly with tap water, and then with deionized water to eliminate dust and dirt. The adsorbent was then dried in an air oven at 105°C for 5 h. Dried adsorbent was ground into powder using a mixer and sieved through a 180 μm sieve. The activation of vetiver root powder was accomplished by imparting heat treatment (at 600 °C) and with the use of phosphoric acid in a ratio of 1:4. The activated product was washed with water and dried, then stored in containers for the use in subsequent experiments (Harikumar et al. 2010).

A standard 500 ppm NaCl solution was prepared by dissolving 0.496 g of anhydrous sodium chloride in 1,000 mL of de-ionized water. The pH of the solution was adjusted with NaOH (0.1 M) or HCl (0.1 M) to the desired value. All chemicals used were of an analytical quality and were bought from local sources in Maharashtra, India.

The preliminary part of this study was designed to optimize the parameters that influence the adsorption efficacy. This involved the outcome of pH variation (from 4 to 9), the dosage of adsorbent (from 1 to 10 g/L), contact time (from 15 to 210 min), and initial concentration C₀ of the adsorbate present in the solution. All experiments were accomplished in triplicate and in a constant volume of 100 mL prepared brackish water (dissolving NaCl in deionized water). The optimum settings were determined to examine the influence of the concentration of NaCl solution on the adsorption capacity of the bioadsorbent. The concentration tested range was in the range of 100–800 ppm. These tests were conducted after optimizing the functional pH, contact time, and adsorbent dosage.

The study extensively analyzed the physicochemical attributes of vetiver biomass to understand its fundamental properties and potential for interacting with metal ions. X-ray diffraction elucidated the biomass structure using Cu Kα radiation, scanning from 20° to 80° at a rate of 2° per minute. Fourier transform Infrared (FTIR) spectra identified functional groups interacting with chloride ions, covering the range of 400–4,000 cm⁻¹ using an ATR-equipped Shimadzu IR affinity instrument. Additionally, the study assessed structural and textural properties like pore volume, diameter, and surface area via nitrogen adsorption/desorption isotherms at 77.35 K and conducted on a Quantachrome Novae 2200 Brunauer–Emmett–Teller (BET) surface area analyzer. These analyses yielded crucial insights into the porous nature and surface characteristics of the biomass, which contribute to our understanding of its adsorption capabilities.

In batch experiments, different chloride ion concentrations from 100 to 800 mg/L were studied to determine bioadsorption equilibrium through Temkin, Freundlich, and Langmuir models, as detailed in Table 1. Kinetic tests were conducted to establish equilibrium time, employing the pseudo-first-order and pseudo-second-order (PS2) models to investigate chloride ion adsorption onto vetiver powder, with equations outlined in Table 2.

Table 2

Kinetics equations and model parameters

Kinetics model .	Equation .	Parameters .		R2 .	RMSE .	MSE .	EABS .	ARE .	ERRSQ .
Linear regression
PS1		Constant k1 (/min)	0.006	0.703	0.004	0.018	0.026	0.139	0.128
		qe cal (mg/g)	8.228
PS2		Constant k2 (/min)	0.019	0.9576	0.003	0.0099	0.014	0.095	0.069
		qe cal (mg/g)	6.728
Non-linear regression
PS1		Constant k1 (/min)	0.020	0.9632	0.004	0.013	0.019	0.113	0.090
		qe cal (mg/g)	7.359
PS2		Constant k2 (/min)	0.020	0.98132	0.003	0.008	0.007	0.073	0.053
		qe cal (mg/g)	5.960

Kinetics model .	Equation .	Parameters .		R2 .	RMSE .	MSE .	EABS .	ARE .	ERRSQ .
Linear regression
PS1		Constant k1 (/min)	0.006	0.703	0.004	0.018	0.026	0.139	0.128
		qe cal (mg/g)	8.228
PS2		Constant k2 (/min)	0.019	0.9576	0.003	0.0099	0.014	0.095	0.069
		qe cal (mg/g)	6.728
Non-linear regression
PS1		Constant k1 (/min)	0.020	0.9632	0.004	0.013	0.019	0.113	0.090
		qe cal (mg/g)	7.359
PS2		Constant k2 (/min)	0.020	0.98132	0.003	0.008	0.007	0.073	0.053
		qe cal (mg/g)	5.960

PS1, pseudo first order; PS2, pseudo second order

View Large

In the equation, ΔG represents Gibb's free energy (kJ/mol), Kf (Freundlich coefficient (L/Kg)), K stands for the adsorption distribution coefficient (g/L), q_e signifies the biosorption capacity at equilibrium (mg/g), C_e denotes the residual concentration of ions at equilibrium (g/L or mg/L), R denotes the universal gas constant (8.314 J/mol K), T represents the absolute temperature (K), ΔH indicates the change in enthalpy (kJ/mol), and ΔS signifies the change in entropy (J/mol K). Using experimental data, the Gibb's free energy (G) was calculated at various temperatures. The ln K vs. 1/T graph's slope and intercept were utilized to determine the enthalpy change (H) and entropy change (S) values. b is the Temkin constant related to heat of sorption (J/mol), and A is the equilibrium binding constant corresponding to the maximum binding energy (L/g).

The application of machine learning methods to forecast the effectiveness of contaminant removal in hydrological streams continues to develop (Nguyen et al. 2022). However, due to a number of practical and financing limitations, this study aggregated batch adsorption trial data to evaluate artificial neural network (ANN) accuracy in predicting bioadsorbent performance for ion removal. The ANN architecture involves input and output layers, with an intermediate encoded layer. This layer is made up of nodes that are connected to form these layers. Nonlinear computations are made possible by activation functions such as sigmoid and rectified linear units (ReLU). The model employs a linear transfer function (purelin) and tansigmoid transfer function (tansig) within the encoded layer. The Levenberg–Marquardt algorithm is used for network training.

Vetiver root powder's surface area and pore properties can be identified by applying BET (Table 4) analysis to compute the pore size and surface area and the Barrett–Joiner–Halenda procedure for estimating the pore volume. Before the N₂ adsorption experiment, the vetiver root powder samples are degassed at 120 °C for 6 h in order to measure the surface area and remove any adsorbed species. Nitrogen adsorption isotherms, performed at an adsorption temperature of 77.35 K, provide information on the textural characteristics of vetiver root powder that has been pre- and post-activated. These isotherms are included in Table 4. Based on analysis, the material is classified as mesoporous by the IUPAC with a pore size of 3.4 nm. It has been shown that this mesoporous quality promotes enhanced dispersion of particles (Fatombi et al. 2019; Rosanti et al. 2022).

However, when pH levels exceed 6 and move toward the alkaline end of the spectrum, reaching up to 9. In this instance, the pH rise did not result in a corresponding rise in chloride ion absorption. Instead, the removal rate dropped from 75 to 71%, suggesting a subsequent decline in adsorption capacity. Chloride ions and hydroxyl ions compete for adsorption locations on the exterior of the adsorbent, which is the cause of this decline. The concentration of hydroxyl ions rises as the pH becomes more alkaline. This increase effectively competes with chloride ions for binding sites, resulting in a decrease in the total adsorption of chloride ions (Harikumar et al. 2010).

Increasing the adsorbent amount from 7 to 10 g/L showed a plateau in chloride adsorption, indicating that beyond a certain point, additional material did not significantly enhance adsorption capacity. This suggests a loss of significance in chloride removal rates with higher adsorbent dosages. The larger dose had a negative impact on adsorption capacity due to diffusional constraints and a reduced solvent ratio. This resulted in a slower approach and binding of chloride ions to active adsorption sites. Consequently, the concentration gradient driving ion diffusion diminished, hindering the efficacy of higher adsorbent doses (Jagtap et al. 2011; Shahawy et al. 2021).

Concurrently, the adsorption capacity (q_e) reached 5.73 mg/g after 120 min, indicating the amount of chloride ions adsorbed per unit amount of the bioadsorbent. However, the plateauing of chloride elimination and q_e beyond 120 min suggests an equilibrium or saturation point, where further contact time does not significantly increase chloride adsorption (Zhang et al. 2017). This pattern aligns with previous studies (Jiang et al. 2016) that observed optimal efficacy in the early contact phases due to the availability of open surfaces for bioadsorption, followed by a saturation of active binding sites leading to declining removal rates over extended contact periods.

The Langmuir isotherm emerged as the most suitable model for explaining Cl⁻ ion adsorption due to its stronger correlation coefficient (R²) compared to the Freundlich model. Supported by lower error values and substantial R² values, the Langmuir isotherm confirmed its applicability through nonlinear regression. It indicated that there was a monolayer adsorption of chloride ions onto activated vetiver powder with a maximum adsorption capacity (q_max) of 17.58 mg/g (Ayawei et al. 2017). Conversely, the Freundlich isotherm exhibited lower correlation coefficients, typically relevant to heterogeneous surfaces, although the ‘n’ value suggested relatively strong adsorption of chloride ions on the exterior of adsorbent (Kumar et al. 2019).

The Langmuir isotherm's superiority was reinforced by improved nonlinear regression performance, confirming its validity in representing chloride ion bioadsorption onto activated vetiver powder. The Temkin isotherm, on the other hand, was found to be inadequate in describing chloride ion adsorption onto activated vetiver powder, as evidenced by residual plots, coefficient of determination (R²) values, and errors shown in Table 1.

Additionally, the root mean square error (RMSE) values robustly supported the pseudo-second-order model's close fit to the experimental findings, affirming its superior performance in both linear and nonlinear regression analyses. Across all observations, the nonlinear versions of kinetic models consistently outperformed their linear counterparts. This was evident in the higher R² values and lower RMSE values, which signify enhanced accuracy in characterizing bioadsorption kinetics.

In summary, the study found the pseudo-second-order model to be highly effective in explaining chloride ion bioadsorption onto vetiver powder. With superior R² values, low RMSE values, and consistent performance in both linear and nonlinear regression analyses, the PS2 model emerged as the most suitable choice for accurately depicting bioadsorption kinetics, as presented in Table 2.

The positive ΔG values confirm the feasibility of chloride ion bioadsorption onto activated vetiver powder, denoting a non-spontaneous reaction. The negative ΔH signifies an exothermic interaction between chloride ions and vetiver powder, while the negative ΔS indicates decreased uncertainty at the solid–liquid interface, suggesting increased orderliness in the course of adsorption. These negative estimates for ΔH and ΔS collectively imply a predominantly physical adsorption process without substantial structural changes in the adsorbent material (Ebelegi et al. 2020; Paranjape & Sadgir 2023).

This study developed an ANN model with two layers to assess chloride ion biosorption by activated vetiver powder. The comprehensive model details are presented in Table 6. Splitting the data into training, validation, and test subsets, 85% was used for training and 15% for testing the model's predictive capacity for new data points. This allowed us to evaluate the ANN model's general performance.

Table 6

ANN architecture, parameters, and performance

ANN architecture
Input parameters	pH, contact time (min), bioadsorbent dose (g/L), agitation speed (rpm), temperature (°C), initial metal ion concentration (mg/L), adsorption capacity (mg/g), removal efficiency (%)
No. of layers	2
No. of neurons	1–10
Training algorithm	Levenberg–Marquardt (trainlm)
Transfer function	Logsig
Max epochs	1,000
Training data and testing data	85%, 15%
R2	0.99

ANN architecture
Input parameters	pH, contact time (min), bioadsorbent dose (g/L), agitation speed (rpm), temperature (°C), initial metal ion concentration (mg/L), adsorption capacity (mg/g), removal efficiency (%)
No. of layers	2
No. of neurons	1–10
Training algorithm	Levenberg–Marquardt (trainlm)
Transfer function	Logsig
Max epochs	1,000
Training data and testing data	85%, 15%
R2	0.99

View Large

Multiple feedforward networks were trained to determine the optimal architecture by minimizing the mean squared error (MSE). The preferred configuration of the ANN model consisted of two hidden layers (10 neurons each), an output layer (one neuron), and a feedforward neural network with an input number of seven. This configuration is denoted as MLP (7:10:1:1) for inputs, hidden layers, output layer, and output, respectively. Utilizing the Levenberg–Marquardt algorithm, this network was trained to establish ideal weights and biases.

Activated vetiver grass root powder proves to be an environmentally sustainable and economically viable biosorbent for chloride ion extraction from saline media, demonstrating a maximum bioadsorption potential of 17.58 mg/g under optimized conditions. The Langmuir model effectively captured this process, establishing optimal parameters at 500 ppm concentration, pH 6, 0.7 g of biomass concentration, 120 min contact time with 150 rpm agitation speed.

Using nonlinear regression, the study validated ideal isotherms and kinetics with the pseudo-second-order model for kinetics and the Langmuir model for adsorption. Thermodynamic analysis revealed the feasibility, non-spontaneous nature, and exothermicity of chloride ion adsorption onto activated vetiver root powder, offering insights into the process.

Leveraging machine learning, specifically the ANN, enabled precise predictions of chloride ion extraction from saline solutions. These predictions exhibit strong alignment with experimental results, promising efficient bioadsorption predictions, and significant time and cost-savings. This incorporation of machine learning represents a pivotal stride in understanding bioadsorption processes, potentially revolutionizing environmental engineering practices for water and wastewater treatments.

Activating vetiver root powder provides a sustainable approach to reduce agricultural waste while creating valuable adsorbents, enhancing water quality specifically for irrigation purposes. Future research avenues should focus on enhancing or modifying the biomass for enhanced removal rates of chlorides. Examining activated vetiver biomass within intricate systems containing various co-ions in wastewater effluents is suggested to comprehensively assess its bioadsorptive traits. Moreover, exploring column studies, advancing pilot-scale commercialization, exploring adsorbent desorption and reuse, and formulating secure disposal approaches for the laden adsorbent would enhance and broaden this research domain.

The authors extend their sincere gratitude to COEP Technological University, Pune for their invaluable support and resources provided throughout this research endeavor.

R.C. Dhumal and P. Sadgir conceptualized the whole article and developed the methodology. R.C. Dhumal performed the experimentation and presented the data for the first draft of the manuscript. P. Sadgir provided guidance and reviewed the manuscript. All authors read and approved the final manuscript.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.