Optimizing Natural Organic Matter Removal from Water by UV/H₂O₂ Advanced Oxidation Using Central Composite Design

Abstract

The inevitable ubiquity of natural organic matter (NOM) in all waters presents a challenge to the proper functioning of water treatment processes. Therefore, minimizing NOM in raw water is crucial to avoid operational issues in subsequent treatment steps. In this experimental study, we aimed to maximize the degradation of NOM using UV/H₂O₂ advanced oxidation, employing design of experiments (DoE) and response surface methodology (RSM) for process optimization. Experiments were carried out on synthetic water, and the effects of dissolved organic carbon (DOC) content and hydrogen peroxide concentration on DOC removal at neutral pH were examined. NOM isolated from the Suwannee River was used as a representative model. The process was modeled and optimized using Design-Expert 14.0.7.0 software. The highest DOC removal of approximately 34% was observed at a DOC level of ~8 mg L⁻¹ and an H₂O₂ concentration just below 250 mg L⁻¹. Degradation products were analyzed by ultra-high-performance liquid chromatography coupled with hybrid quadrupole time-of-flight mass spectrometry, revealing sixteen compounds, mostly long-chain saturated fatty acids. Finally, the energy efficiency of the experimental setup was assessed and discussed.

1. Introduction

Natural organic matter (NOM) is ubiquitous in all waters and can interfere with the proper functioning of unit processes within water treatment plants. It contributes to a variety of operational challenges, including the increased complexation of heavy metals with NOM and the adsorption of emerging contaminants onto NOM. Additional issues include reduced activated carbon efficiency, membrane fouling, increased consumption of coagulants and disinfectants, and intensified fouling within the distribution system [1,2]. After the discovery of organohalogen disinfection by-products (DBPs), the role of NOM in their formation became an important focus of scientific investigation [3]. To address this problem, much of the research was focused on the removal of DBP precursors (mostly NOM), which is considered the best control strategy as it minimizes the amount of DBPs regardless of the disinfectant applied. However, achieving sufficient removal of DBP precursors is not straightforward. Drinking water providers were confronted with the dual challenge of minimizing DBP formation while still maintaining effective disinfection. Several approaches have been considered to reduce DBP formation, such as more intensive removal of DBP precursors from raw water, switching to other chemical disinfectants, and removing DBPs upon their formation in treated water. The latter option is particularly difficult as it requires additional processes, leading to significantly higher costs [4,5,6]. Improved removal of organic DBP precursors can be accomplished through enhanced coagulation, which seeks to achieve greater total organic carbon (TOC) reduction by applying higher coagulant doses and/or acid addition, or by involving additional processes such as nanofiltration or granular activated carbon treatment. While both enhanced coagulation and granular activated carbon filtration lower NOM content in water, they are inefficient in bromide removal. When chlorine is added, bromide present in water leads to the formation of brominated DBPs, which are often more hazardous than chlorinated by-products [7,8]. Nanofiltration is a method with great potential for DBP control. Despite relatively high bromide rejection, it alters the bromide-to-TOC ratio between feed and permeate water and thus favors the formation of brominated species upon chlorination [9,10]. The seemingly attractive idea of replacing chlorine with other chemical disinfectants—such as chlorine dioxide, ozone, or chloramines—does not offer a universal solution, since each disinfectant/oxidant produces an inherently distinct set of by-products [11]. For instance, ozone reduces the formation of trihalomethanes (THMs) and haloacetic acids (HAAs), but in waters with elevated bromide levels (common in coastal groundwater due to seawater intrusion), it leads to bromate formation [12].

Advanced oxidation processes (AOPs) find applications across nearly all areas of water treatment and are therefore considered a viable strategy for NOM removal. The primary goal of each AOP is to generate a sufficiently high concentration of reactive radicals, most notably hydroxyl radicals (HO^•), which react rapidly and non-selectively with organic compounds in water [13,14], resulting in partial degradation. Among the most widely implemented AOPs is the UV/H₂O₂ process, which has been applied to remove various anthropogenic organic contaminants [15] as well as naturally occurring organic substances in water [16,17,18]. In this process, hydroxyl radicals are generated by homolytic fission of a covalent bond between two oxygen atoms in hydrogen peroxide following the absorption of UV-C quanta. Due to the relatively low molar absorption coefficient of hydrogen peroxide, i.e., 19.6 M⁻¹ cm⁻¹ at 254 nm [19], it is necessary to supply a fairly high concentration of hydrogen peroxide to generate an adequate concentration of hydroxyl radicals. However, excess H₂O₂ can reduce process efficiency because of its significant OH radical scavenging effect. Similarly, certain dissolved species in water, such as carbonate and bicarbonate, may interfere with NOM degradation by consuming HO^• radicals (k_{HO•,carbonate} = 3.9 × 10⁸ M⁻¹ s⁻¹; k_{HO•,bicarbonate} = 8.5 × 10⁶ M⁻¹ s⁻¹) [13]. Consequently, it is important to determine the optimal H₂O₂ concentration for each UV/H₂O₂ system to achieve maximum NOM removal. Beltrán et al. [20] reported that, for H₂O₂ concentrations above 0.01 M, the oxidation rate of atrazine may even be lower than that obtained by direct photolysis. Wang et al. [21] reported that the optimal H₂O₂ concentrations for humic acid degradation ranged from 0.01 to 0.05% (3.2–16.3 mM), while waters with high DOC and elevated alkalinity would require higher peroxide levels, around 0.1% (32 mM). Zhang et al. [22] studied the degradation of two cytostatic drugs (CSDs), cyclophosphamide and 5-fluorouracil, using both direct UV photolysis and the UV/H₂O₂ process. They observed that UV irradiation alone was insufficient for effective removal of these CSDs, whereas the addition of H₂O₂ markedly improved degradation efficiency. Based on the electrical energy per order (E_EO) parameter, the optimal H₂O₂ dosage was determined to be 0.2 mM for both compounds.

In experimental studies, a common approach is to vary one factor at a time (OFAT) while keeping all other factors constant. However, the limitations of the OFAT method—most notably its inability to capture potential interactions between factors—can be addressed by employing a factorial design of experiments [23]. Factorial designs are particularly effective when the effects of two or more factors need to be evaluated simultaneously. Response surface methodology (RSM) is often applied to support model development and analysis, aiming to optimize a response variable influenced by several factors [23]. Rezaee et al. [24] studied NOM removal from aqueous solutions using a UV/H₂O₂ AOP with RSM based on the Box–Behnken design. In contrast, the present research employs a central composite design (CCD), which extends the experimental space by including points where all factors are set at extreme levels. This broader coverage improves model precision at extreme settings and provides a stronger basis for optimization. The model’s quality, in terms of coefficient of determination, adequate precision, and residual analysis, was confirmed through diagnostic evaluation. Furthermore, the model’s predictive capability was validated using three randomly selected factor settings.

The objective of this study was to model and optimize NOM removal from synthetic water using the UV/H₂O₂ process under typical natural water pH conditions, employing a design of experiments (DoE) methodology. Both the main effects of the factors and their interactions, as well as the adequacy of the proposed model and the optimal factor settings, were evaluated. To enhance understanding of DOC removal from NOM-rich water via UV/H₂O₂ advanced oxidation, optimization was performed using simulated water containing Suwannee River NOM (SRNOM), which is widely regarded as a representative surrogate for natural surface water NOM [25]. SRNOM possesses a more complex and heterogeneous structure than commonly used humic acid (HA) [26], making its degradation via UV/H₂O₂ more challenging. In addition to process optimization, the study includes the identification of degradation products using advanced analytical techniques and the first reported estimate of energy efficiency via the E_EO parameter for UV/H₂O₂ degradation of SRNOM. This work also characterizes the degradation products formed during treatment, contributing to the development of more effective AOP strategies for treating NOM-rich waters with complex and undefined composition.

This study therefore addresses the current lack of reproducible and systematically optimized UV/H₂O₂ investigations on complex NOM by combining several elements that have not previously been integrated in a single study: the use of a well-characterized and structurally complex NOM standard (SRNOM) as a reproducible model matrix, the application of central composite design (CCD) for process optimization, and the simultaneous assessment of degradation products and energy efficiency (E_EO). This combined approach is intended to provide both a framework for optimizing UV/H₂O₂ performance under controlled conditions and information on the transformation of NOM during treatment.

2. Materials and Methods

2.1. Synthetic Water Preparation and Reagents

All experiments were conducted using synthetic water prepared by dissolving the Suwannee River natural organic matter (SRNOM) standard (1R101N, International Humic Substances Society, St. Paul, MN, USA) in ultrapure water produced by a GenPure system (TKA Wasseraufbereitungssysteme GmbH, Niederelbert, Germany). The SRNOM stock solution was made by adding 0.1 g of SRNOM to 0.5 L of ultrapure water and stirring overnight at room temperature. The stock solution was then filtered through a prewashed 0.45 µm polyethersulfone membrane filter (FilterBio Membrane Co., Budapest, Hungary). The resulting solution contained 80 mg L⁻¹ of dissolved organic carbon (DOC), measured as non-purgeable organic carbon. Working solutions were buffered with phosphate buffer (0.02 M KH₂PO₄/Na₂HPO₄) and adjusted to neutral pH 7.0 with 1 M NaOH. All working solutions were freshly prepared daily and stored at 4 °C in amber glass bottles.

Formic acid (Suprapur) was obtained from Merck (Darmstad, Germany), ammonium fluoride from Sigma Aldrich (Steinheim, Germany), and methanol, acetone, and water (all LC/MS grade) from J.T. Baker (Deventer, Netherlands). Hydrogen peroxide (30% stock solution) was acquired from Kemika (Zagreb, Croatia).

2.2. Analytical Methods

An HP 8453 UV-Vis spectrophotometer (Hewlett Packard, Palo Alto, CA, USA) was employed for spectrophotometric measurements. Dissolved organic carbon (DOC) was measured using a TOC analyzer (model TOC-V_CPH, Shimadzu Co., Kyoto, Japan).

Degradation products were analyzed by UHPLC–QTOF–MS using a 1290 UHPLC system coupled to a 6550 iFunnel Q-TOF LC/MS (Agilent Technologies, Santa Clara, CA, USA). Separation was performed on a reversed-phase Acquity UPLC BEH C18 column (130 Å, 1.7 µm, 2.1 mm × 150 mm; Waters, Milford, MA, USA) with an injection volume of 100 µL and column temperature of 55 °C. Mobile phases consisted of 1 mM ammonium fluoride in water (A) and 100% methanol (B), with a 22 min gradient elution from 100% A to 100%B at a flow rate of 0.4 mL min⁻¹. The Dual Agilent Jet Stream ion source was used for ionization, and analyses were conducted in MS and MS/MS modes with collision energies of 10, 20, and 40 eV over a mass range of 50–1000 m/z. Data acquisition and processing were performed with Agilent MassHunter software (Qualitative Analysis version B.07.00 SP1, B7024.29, Agilent Technologies, Santa Clara, CA, USA). Non-target screening results were interpreted using MassHunter algorithms and personal compound database and libraries (PCDLs), covering approximately 85,000 organic compounds (forensic toxicology, pesticide, and Metlin metabolite PCDLs).

2.3. Experimental Methods

2.3.1. Experimental Set-Up

The experiments were conducted in a cylindrical glass photoreactor with an inner diameter of 28 mm and a height of 202 mm, equipped with a cooling fan. A low-pressure mercury UV lamp (UVP, Pen-Ray 90-0012-01, Cambridge, UK), emitting monochromatic radiation at 254 nm, was positioned at the center of the reactor to ensure uniform light distribution. The photon flux at 254 nm was determined using hydrogen peroxide actinometry following [27] and was found to be 1.37 × 10⁻⁷ einstein s⁻¹ for a reaction volume of 105 mL. When normalized per liter, this corresponds to 1.30 × 10⁻⁶ einstein s⁻¹ L⁻¹. This value is comparable to those reported in similar bench-scale UV/H₂O₂ studies. For instance, Ahn et al. [28] reported a photon flux of 1.80 × 10⁻⁶ einstein s⁻¹ L⁻¹ under similar experimental conditions. Although relatively low compared to industrial-scale UV/H₂O₂ systems (typically 10⁻⁶ to 10⁻⁵ einstein s⁻¹), the irradiation conditions established in this study provide a solid foundation for future scale-up using more advanced UV sources, such as UV-LED arrays, which offer modular design and potential for higher irradiation intensities, improved energy efficiency, and longer operational lifetimes. Moreover, the irradiation conditions used in this study enabled detailed optimization and analytical characterization of NOM degradation.

To ensure stable irradiation, the lamp was switched on at least 15 min prior to each experiment. Temperature was maintained at 25 °C using a thermostated water bath, and homogeneous mixing was achieved with a magnetic stirrer. For NOM degradation experiments, 100 mL of SRNOM synthetic water (prepared at the desired DOC concentration) was mixed with 5 mL of H₂O₂ solution at a predefined concentration. Samples were withdrawn at selected time intervals to monitor absorbance at 254 nm and determine DOC concentration. Each experimental run lasted 3 h.

2.3.2. Experimental Design

To optimize the removal of NOM by the UV/H₂O₂ process, several preliminary experiments were first conducted, in which the H₂O₂ dose varied while maintaining a fixed DOC concentration. The selection of H₂O₂ doses was guided by prior experience with waters of similar composition. The greatest reduction in DOC concentration was achieved with an H₂O₂ dose of 250 mg L⁻¹ in preliminary tests where the initial DOC concentration was 5 mg L⁻¹. This dose was subsequently used as a constant in further experiments that investigated the effect of varying initial DOC concentrations on NOM degradation. Initial DOC concentrations of 2, 5, and 10 mg L⁻¹ were tested, with 5 mg L⁻¹ showing the most pronounced effect on NOM removal. All preliminary experiments were performed in at least duplicate.

To systematically assess the influence of two independent variables—H₂O₂ dose and initial DOC concentration—on the percentage removal of DOC (∆DOC), a central composite design (CCD) was employed. This design included four factorial runs (2², n_F = 4), four axial runs at distance α from the center point (n_A = 4), and five center runs (n_C = 5), yielding a total of 13 experimental runs. The total number of experiments was determined according to Equation (1):

$$N=2^k+2k+n_{\mathrm{c}}$$

(1)

where k is the number of factors and n_c is the number of center runs (k = 2, n_c = 5). Axial runs were included to account for quadratic terms in the model. To ensure rotatability of the design, a value of α = (n_F)^1/4 = 1.414 was used, providing equal model precision across the design space. Replication of the center point enabled independent estimation of pure error without biasing the effect estimates or introducing imbalance into the design [23]. In this study, the center point corresponded to an H₂O₂ dose of 250 mg L⁻¹ and a DOC concentration of 6 mg L⁻¹. The H₂O₂ dose (X₁) ranged from 108.6 to 391.4 mg L⁻¹, while the initial DOC concentration (X₂) ranged from 0.4 to 11.7 mg L⁻¹ (

Table 1. CCD for two independent variables with the observed and predicted response values.

#	Run	Natural Variables		Coded Variables		∆DOC Observed (%)	∆DOC Predicted (%)
		H2O2 (mg L−1)	DOC (mg L−1)	x1	x2
7	1	250	0.4	0	−1.414	13.98	14.64
4	2	350	10	1	1	28.87	28.31
12	3	250	6	0	0	35.59	34.42
3	4	150	10	−1	1	34.60	33.86
11	5	250	6	0	0	37.68	34.42
8	6	250	11.7	0	1.414	34.27	34.92
1	7	150	2	−1	−1	24.66	23.92
6	8	391.4	6	1.414	0	35.22	35.75
9	9	250	6	0	0	29.76	34.42
5	10	108.6	6	−1.414	0	32.80	33.57
2	11	350	2	1	−1	33.13	32.57
13	12	250	6	0	0	33.99	34.42
10	13	250	6	0	0	35.10	34.42

‘#’ denotes the design point number as defined by the experimental design, while ‘Run’ refers to the randomized execution order. 2.3.3. Determination of Energy Consumption.

). Actual values of natural variables (X_i) were coded to unitless variables (x_i) as:

$$x_i={\displaystyle \frac{X_i-X_0}{\Delta X_i}}$$

(2)

where X_i is the natural variable, X₀ the center-point value, and ∆X_i the step size. Factor levels were chosen to span a range relevant to practical applications.

The percentage removal of DOC by UV/H₂O₂ process was calculated as:

$$\Delta \mathrm{DOC} (\%)={\displaystyle \frac{{(\mathrm{DOC}}_0-{\mathrm{DOC}}_t)}{{\mathrm{DOC}}_0}}\times 100\%$$

(3)

where DOC₀ is the initial DOC concentration (mg L⁻¹) and DOC_t is the DOC concentration (mg L⁻¹) at time t.

The CCD with two factors at two levels was used to fit a response surface model of the form:

$$y={\beta }_0+{\displaystyle \sum _{i=1}^k{\beta }_i{x}_i}+{\displaystyle \sum _{i<j}^k{\beta }_{ij}{x}_i}{x}_j+{\displaystyle \sum _{i=1}^k{\beta }_{ii}}{x}_i^2+{\displaystyle \sum _{i=1}^k{\displaystyle \sum _{j\ne i}{\beta }_{iij}{x}_i^2{x}_j+{\displaystyle \sum _{i<j<k}{\beta }_{ijk}{x}_i{x}_j{x}_k+\epsilon }}}$$

(4)

where y is the response variable, x_i and x_j are coded independent variables, β coefficients represent regression terms, k is the number of factors, and ε accounts for experimental error.

Because third-order terms are aliased, the highest-order polynomial model estimable with two independent variables in this design is the reduced cubic model given by Equation (5):

$$y={\beta }_0+{\beta }_1{x}_1+{\beta }_2{x}_2+{\beta }_{12}{x}_1{x}_2+{\beta }_{11}{x}_1^2+{\beta }_{22}{x}_2^2+{\beta }_{112}{x}_1^2{x}_2+{\beta }_{122}{x}_1{x}_2^2+\epsilon .$$

(5)

Accordingly, the reduced cubic model was applied in this study. Design-Expert 14.0.7.0 (State-Ease Inc., Minneapolis, MN, USA) was used to estimate regression coefficients, perform analysis of variance (ANOVA), and generate three-dimensional response surface plots.

The electric energy per order (E_EO) was introduced by Bolton et al. [29] as a parameter for evaluating AOPs. The E_EO represents the amount of electric energy required to reduce the contaminant concentration by one order of magnitude in a unit volume. For batch operation, the E_EO (kWh m⁻³ order⁻¹) can be calculated using the following equation [29]:

$$E_{\mathrm{EO}}={\displaystyle \frac{P\cdot t\cdot {10}^3}{V\cdot \log \left(\frac{c_{\mathrm{i}}}{c_{\mathrm{f}}}\right)}}$$

(6)

where P is the electric power of the radiation source (kW), V is the volume of the solution containing the substance to be degraded (L), t is the treatment time (h), c_i is the initial concentration of the substance (mol L⁻¹), c_f is the final concentration of the substance (mol L⁻¹), and the factor 10³ converts liters to cubic meters. Assuming pseudo-first-order kinetics, Equation (6) can be rearranged as:

$$E_{\mathrm{EO}}=38.4{\displaystyle \frac{P}{V\cdot k_1^{\prime }}}$$

(7)

where k′₁ is the pseudo-first-order rate constant (min⁻¹).

3. Results and Discussion

3.1. Optimization with RSM

Thirteen experiments were conducted in random order, and the observed and predicted DOC removals, together with both coded and natural variable values, are presented in Table 1.

A reduced cubic model (Equation (5)) was fitted to the experimental results, including only statistically significant terms as well as those necessary to preserve model hierarchy. The final regression equation of the reduced cubic model, expressed in terms of coded variables, is given by:

$$\Delta \mathrm{DOC}=+34.42+0.7722x_1+7.30x_2-3.55x_1{x}_2+0.1194x_1^2-4.88x_2^2-5.88x_1^2{x}_2$$

(8)

The adequacy of the model was evaluated using the coefficient of determination (R²), the adjusted R², the F-test, and the lack-of-fit test. The ANOVA results for the reduced cubic response surface model are presented in

Table 2. ANOVA results of the response surface reduced cubic model.

Source	Sum of Squares	DF	Mean Square	F-Value	Prob > F	Remark
Model	435.19	6	72.53	11.50	0.0045	Significant
A-H2O2	4.77	1	4.77	0.76	0.4180
B-DOC	212.43	1	212.43	33.67	0.0011
AB	50.38	1	50.38	7.99	0.0301
A2	0.099	1	0.999	0.016	0.9044
B2	164.92	1	164.92	26.14	0.0022
A2B	69.05	1	69.05	10.95	0.0162
Residual	37.85	6	6.31
Lack of fit	3.47	2	1.73	0.20	0.8252	Not significant
Pure error	34.38	4	8.50
Corr. total	473.04	12

Nonsignificant terms A (p = 0.4180) and A² (p = 0.9044) are retained to preserve model hierarchy due to significant interactions AB (p = 0.0301) and A²B (p = 0.0162).

.

The adjusted R² value of 0.84 indicates that the regression model explains approximately 84% of the total variation in the data, confirming the adequacy of the model. Furthermore, the model F-value of 11.50 (p < 0.01) demonstrates that the regression is statistically significant. The signal-to-noise ratio (S/N) exceeds 11, confirming the adequacy of the signal. The lack-of-fit test yielded an F-value of 0.20 with p > 0.05, indicating that the lack of fit is not statistically significant. This suggests that the proposed model accurately describes the relationship between the independent variables and the response, i.e., the test did not reveal any systematic deviation. The non-significance of the lack of fit thus confirms the suitability of the model.

In addition, model adequacy is often evaluated by analyzing residuals, typically through the construction of a normal probability plot. Figure 1 presents the normal probability plot of internally studentized residuals for the reduced cubic model.

It can be seen from Figure 1 that most of the residual data points lie close to a straight line, indicating that the error distribution is approximately normal. In addition, no significant outliers are observed. Therefore, the proposed model equation is considered reliable and can be applied to predict DOC removal by the UV/H₂O₂ process.

To further verify the accuracy of the model, additional experiments were carried out. The results of three verification points are presented in

Table 3. Verification of the proposed model.

Verification Point	DOC (mg L−1)	H2O2 (mg L−1)	Mean ΔDOC (%)	−95% CI ΔDOC (%)	+95% CI ΔDOC (%)
1	5.132	193.0	30.13	29.49	34.88
2	7.766	314.7	32.82	32.33	37.96
3	3.444	211.0	26.32	24.24	30.10

.

The results show that all single-point values fall within the 95% confidence interval of the mean, which also implies that they satisfy the 95% prediction interval of the single value. Hence, the model is both useful and sufficiently precise for estimating the response across the design space. Its predictive accuracy, validated by these additional experiments, highlights its practical relevance for process optimization in water treatment applications.

Figure 2 presents the three-dimensional response surface plot of the proposed model, illustrating the combined effects of the two studied factors on DOC removal efficiency.

According to the ANOVA results (Table 2), the significant terms included in the model are B (p = 0.0011), AB (p = 0.0301), B² (p = 0.0022) and A²B (p = 0.0162) while the linear (p = 0.4180) and quadratic (p = 0.9044) terms of factor A (H₂O₂ concentration) are not individually significant. These nonsignificant terms were retained to maintain model hierarchy, a standard RSM practice when higher-order interactions (e.g., AB and A²B) are significant, as indicated by the footnote in Table 2. This retention reflects a context-dependent effect of H₂O₂ concentration on DOC removal, modulated by DOC levels.

For instance, when DOC concentration is 10 mg L⁻¹, higher H₂O₂ concentrations lead to a slight decrease in DOC removal. Conversely, at relatively low DOC concentrations (2 mg L⁻¹), DOC removal efficiency generally increases with increasing H₂O₂ concentration. In general, increasing DOC concentration leads to higher light absorbance of the water matrix, thereby lowering the average irradiance in the photoreactor. The reduced irradiance decreases the steady-state concentration of hydroxyl radicals, ultimately reducing DOC removal efficiency [30]. On the other hand, the modest reduction in DOC removal efficiency observed at higher H₂O₂ concentrations can be attributed to hydroxyl radical scavenging by the excess of H₂O₂. Under such conditions, less reactive hydroperoxyl radicals (HOO^•) are formed (Equation (9)), which reduces the overall efficiency of the process. This scavenging effect is consistent with the known second-order rate constant for the reaction between HO^• and H₂O₂ (2.7 × 10⁷ M⁻¹ s⁻¹ [13]), which can substantially lower the steady-state HO^• concentration at elevated H₂O₂ levels.

$${\mathrm{HO}}^{•}+{\mathrm{H}}_2{\mathrm{O}}_2\to {\mathrm{HOO}}^{•}+{\mathrm{H}}_2\mathrm{O}$$

(9)

As illustrated in Figure 2, the maximum DOC removal efficiency of approximately 34% occurs at DOC concentrations around 8 mg L⁻¹ and slightly lower H₂O₂ concentration of about 250 mg L⁻¹. This optimal condition reflects a balance between hydroxyl radical generation and scavenging, highlighting the critical role of H₂O₂ dosing in maximizing NOM degradation. Compared to prior studies reporting DOC removal efficiencies ranging from 11% to 80% under varying UV/H₂O₂ conditions [31], the 34% removal achieved here is moderate but was obtained under controlled and neutral pH conditions, which are representative of typical natural waters. The relatively low removal efficiency can be attributed to the low photon flux of the UV source used in this study. Although the reaction time was long (3 h), the low photon flux limited the rate of hydroxyl radical generation and thereby the extent of NOM oxidation. This is consistent with the findings of Ahn et al. [28], who reported lower mineralization (20.5%) under a similar photon flux but a much shorter reaction time (174 s). These findings underline that identifying the optimal H₂O₂ concentration in relation to DOC content is crucial for achieving effective and consistent process performance, even under low-intensity conditions, highlighting the importance of tailored oxidant dosing in UV/H₂O₂ treatment. It should be noted that the photon flux applied in this study was considerably lower than in typical full-scale UV/H₂O₂ systems. This limitation restricts the direct extrapolation of the present results to real-world conditions. However, the low-intensity bench-scale setup was intentionally chosen to allow controlled optimization and detailed characterization of NOM degradation. These optimized conditions may serve as a basis for future scale-up studies using higher-intensity UV sources such as UV-LED arrays.

Overall, these findings demonstrate that, even under the applied low-intensity irradiation conditions, the use of RSM enabled the identification of an operational region where DOC removal was maximized within the studied factor ranges. This highlights the value of statistical design for disentangling the complex interactions between process variables and for providing quantitative guidance on process behavior, complementing empirical observations.

3.2. Energy Efficiency of NOM Degradation

The pseudo-first-order rate constant k′₁ was estimated based on the data obtained by monitoring the decrease in DOC concentration over time. The data were fitted to a pseudo-first-order kinetic model, yielding a reaction rate constant of 0.0024 min⁻¹ (Figure 3a).

The estimated E_EO value under the applied conditions was 2.44 × 10³ kWh m⁻³ order⁻¹. This relatively high value can be attributed to slow mineralization kinetics, as E_EO is strongly influenced by the nature of the target compound [30]. Aquatic NOM mainly consists of humic and fulvic substances [32], which undergo substantial degradation during UV/H₂O₂ treatment, although complete mineralization is typically not achieved [28,33]. Gowland et al. [32] provided an insightful discussion on energy efficiency, emphasizing the impact of the power consumption of the radiation source, which directly enters the numerator of the E_EO expression. The factor of 10³ used for unit conversion from liters to cubic meters further amplifies E_EO values, leading to large disparities. They also highlighted that short treatment times are important to maintain favorable E_EO values. These considerations underline the importance of recent advances in LED technology, which have considerably improved efficiency and reduced the energy consumption of UV radiation sources compared to conventional mercury lamps. To the best of our knowledge, no peer-reviewed bench-scale study on the optimization of NOM degradation by UV/H₂O₂ AOP reports an E_EO estimate suitable for direct comparison. Although the E_EO was calculated following the standard definition, this very high value should not be interpreted as a realistic estimate of energy demand for full-scale UV/H₂O₂ treatment. Instead, it reflects the combination of low photon flux and modest DOC removal under the applied bench-scale conditions and is presented here as an indicative benchmark for comparing system efficiency rather than as a measure of practical feasibility.

No significant DOC removal was observed with H₂O₂ alone (250 mg L⁻¹ initial dose, 5 mg L⁻¹ DOC, pH 7.0), with less than 1% variation in UV absorbance at 254 nm over 3 h (Supplementary Figure S1). In the present study, direct UV photolysis (t = 180 min) resulted in a 10.7% decrease in absorbance at 254 nm (A₂₅₄), while in the presence of H₂O₂ the decrease reached approximately 67% (Figure 3b). This strong contrast confirms the synergistic nature of the UV/H₂O₂ process: UV alone induces only limited photolysis of NOM, and H₂O₂ alone is largely ineffective, whereas their combination generates hydroxyl radicals through photolytic cleavage of H₂O₂, which drive the extensive oxidation observed. The greater reduction in A₂₅₄ indicates fragmentation of NOM macromolecules into lower-molecular-weight products that do not absorb UV radiation and are more resistant to hydroxyl radical attack [34].

Tak and Vellanki [5] recently suggested that the E_EO parameter can be expressed using either DOC or UV absorbance at 254 nm. When calculating E_EO based on the 67% reduction in A₂₅₄, the obtained value is approximately 2.5 times lower than that calculated from the DOC decrease, although it still cannot be considered economically feasible. This further emphasizes the importance of selecting appropriate UV radiation sources. Continued advances in UV light efficiency are required to enhance the applicability of this oxidation process for DOC removal in complex, high-NOM water matrices [31]. In full-scale systems, residual hydrogen peroxide is typically removed after treatment by catalytic decomposition or, where applicable, by further UV irradiation, which prevents its carryover into downstream processes.

3.3. Identification of Degradation Products

NOM in synthetic water was partially oxidized by hydroxyl radicals generated during UV/H₂O₂ AOP as evidenced by a substantial decrease in UV absorbance. Hydroxyl radicals react with SRNOM’s humic and fulvic acid constituents via addition to double bonds in aromatic and unsaturated aliphatic compounds, hydrogen abstraction in saturated and unsaturated molecules, and electron transfer reactions [35]. These reactions likely initiate the cleavage of aromatic rings and aliphatic side chains, yielding smaller, oxygenated compounds, such as the long-chain fatty acids identified here (

Table 4. Summary of non-target compounds identified.

#	Name	Mass (DB)	Formula (DB)	Base Peak	Mass	m/z	RT
1	Myristoleic acid	226.1933	C14H26O2	225.1892	226.1965	225.1893	18.741
2	13,14-dihydro Prostaglandin F1α	358.2719	C20H38O5	357.2646	358.2720	357.2647	19.380
3	12-Methyltridecanoic acid	228.2089	C14H28O2	227.2049	228.2121	227.2048	19.508
4	Δ2-trans-Hexadecenoic Acid	254.2246	C16H30O2	253.2204	254.2278	253.2206	19.538
5	Pentadecanoic acid	242.2246	C15H30O2	241.2205	242.2277	241.2204	19.82
6	Stearolic acid	280.2402	C18H32O2	279.2364	280.2435	279.2362	19.821
7	Elaidic acid	282.2559	C18H34O2	281.2519	282.2593	281.2521	20.177
8	14(Z)-Eicosenoic acid	310.2872	C20H38O2	309.284	310.2911	309.2838	20.659
9	Stearic acid	284.2715	C18H36O2	283.2678	284.275	283.2677	20.848
10	Docosanoic acid	340.3341	C22H44O2	339.3304	340.3377	339.3304	21.438
11	Tricosanoic acid	354.3498	C23H46O2	353.3461	354.3535	353.3462	21.596
12	Lignoceric acid	368.3654	C24H48O2	367.3619	368.3692	367.3619	21.878
13	Isoundecylic acid	214.1933	C13H26O2	213.1892	214.1965	213.1893	18.783
14	Myristic acid	228.2089	C14H28O2	227.2047	228.2121	227.2048	19.263
15	15-Methyl palmitic acid	270.2559	C17H34O2	269.2521	270.2592	269.2519	20.238
16	Arachidic acid	312.3028	C20H40O2	311.2994	312.3066	311.2993	20.978

). While SRNOM’s complex, heterogeneous structure precludes a detailed mechanistic pathway, these transformations suggest selective degradation of aliphatic or lipid-like components. As a result, the fraction of NOM not fully mineralized underwent structural changes such as decreased aromaticity, a shift toward lower-molecular-weight compounds, and the formation of more biodegradable products [36].

Degradation products were identified using UHPLC–QTOF–MS. Non-target screening revealed the generation of sixteen degradation products during UV/H₂O₂ treatment of SRNOM, most of which were long-chain saturated fatty acids (Table 4). This screening was intended to provide a qualitative indication of the types of degradation products formed, to complement the observed DOC removal results; however, detailed mechanistic analysis, quantitative determination, and toxicity assessment were not part of the objectives of this optimization-focused study. Previous literature, as summarized by Schmitt-Kopplin et al. [37], indicates that oxidative degradation of humic substances yields fatty acids along with aliphatic and aromatic carboxylic acids, reflecting a broad range of degradation products. Notably, the predominance of long-chain saturated fatty acids (e.g., stearic, arachidic, lignoceric acid) in this study, compared to the mixed products reported, may reflect the specific UV/H₂O₂ conditions employed (e.g., high H₂O₂ dose of ~250 mg L⁻¹, pH 7.0 with 0.02 M phosphate buffer). The predominance of long-chain saturated fatty acids under low UV intensity (photon flux: 1.37 × 10⁻⁷ einstein s⁻¹) suggests a selective cleavage of aliphatic components, differing from the broader product range reported under varied oxidative conditions [37]. An unexpected feature was initially annotated as 13,14-dihydro Prostaglandin F1α (m/z 357.2647, [M–H]⁻), detected via UHPLC-QTOF-MS non-target screening in negative mode (–ESI) with a retention time of ~19.38 min and matching the Agilent PCDL library (Δm < 5 ppm, C₂₀H₃₈O₅). MS/MS fragmentation of m/z 357.2647 [z = 1] at 10–40 eV (Supplementary Figure S2) produced a prominent fragment at m/z 337 (consistent with loss of H₂O) at higher collision energy, while the precursor ion remained dominant at lower energies. This library match was treated with caution and is considered a tentative annotation rather than a confirmed identification. Importantly, the spectrum lacked hallmark prostaglandin-specific diagnostic ions (e.g., m/z 193, 209, and 219 [38]), which reduces confidence in the annotation. Moreover, formation of prostaglandins under UV/H₂O₂ AOP conditions is mechanistically improbable, given the non-specific reactivity of hydroxyl radicals and the absence of polyunsaturated fatty acid precursors in SRNOM. Overall, the evidence suggests that the observed feature is more plausibly attributable to an oxidized polyhydroxylated fatty acid or a structurally related compound, rather than a true prostaglandin. Accordingly, the annotation remains tentative and cannot be confirmed without authentic standards.

The presence of such compounds may affect the quality of AOP-treated water upon chlorination, since hydroxylated NOM moieties exhibit higher reactivity toward chlorine, leading to increased total organic chlorine formation [35]. Although total organic chlorine increases after UV/H₂O₂ pretreatment, the ratio of unknown to total organic chlorine decreases, likely due to enhanced formation of identifiable by-products such as trihalomethanes (THMs) and haloacetic acids (HAAs). To mitigate these effects, Toor and Mohseni [39] suggested coupling UV/H₂O₂ AOP with biological activated carbon (BAC) treatment, since advanced oxidation improves water biodegradability, which is crucial for effectively reducing disinfection by-products (DBPs) in AOP–BAC systems.

A conceptual reaction pathway was proposed to illustrate the plausible transformation of aromatic and aliphatic moieties within SRNOM into the observed oxygenated products (Supplementary Figure S3). Hydroxyl radical attack can lead to aromatic ring opening and side-chain cleavage, producing smaller aliphatic fragments that undergo further oxidation and accumulate as long-chain saturated fatty acids. While the heterogeneous composition of SRNOM precludes establishing a definitive mechanism, this diagram summarizes the most likely degradation routes under UV/H₂O₂ treatment.

4. Conclusions

This study showed that UV/H₂O₂ advanced oxidation can fragment natural organic matter (NOM) in water, although complete mineralization remained limited under the applied low-intensity conditions. The process performance was optimized using a central composite design and response surface methodology, providing useful guidance for selecting operational settings under these bench-scale conditions. While the achieved DOC removal was moderate, the treatment induced structural changes in NOM, yielding smaller and more oxygenated molecules that are typically considered more biodegradable. The study also provided one of the first indicative estimates of energy efficiency (E_EO) for UV/H₂O₂ treatment of Suwannee River NOM, which may serve as a reference point for future optimization efforts. Overall, these findings underline the importance of carefully tailored oxidant dosing and suggest that employing more energy-efficient UV sources (e.g., UV-LEDs) could enhance the feasibility of this process in future large-scale applications.