Valorization of seaweed residue and agricultural side streams for molded biocomposite production: sequential RSM optimization of formulation and hot pressing

Abstract

Seaweed residue after agar extraction is an underutilized secondary biomass with limited self-bonding ability, which restricts its direct use in molded biocomposites. In this study, seaweed residue was combined with rejected corn kernel flour, rejected cassava root flour, polyvinyl alcohol as binder, and glycerol as plasticizer to produce a molded biocomposite by hot pressing. Response surface methodology (central composite design) was applied in two sequential stages to optimize material formulation and hot-pressing conditions, using thickness, density, hardness (defined as peak bending load), and flexibility (defined as deflection at peak load) as responses. The optimum formulation was identified at 0.59 g rejected corn kernel flour and 0.38 g rejected cassava root flour per g seaweed residue, and 6% polyvinyl alcohol, while the optimum hot-pressing condition was 95 °C, 2.7 min, and 187 N. Polyvinyl alcohol, although present at only 3–6%, was the only formulation variable with a significant effect on all four responses, and a control mixture without polyvinyl alcohol, tested at one fixed formulation, could not be consolidated. Verification experiments agreed with predictions within 5% for thickness, hardness, and flexibility (about 8.7% for density). A glycerol level of 0.2 mL g⁻¹ maintained the highest peak bending load, whereas 0.4 mL g⁻¹ caused pronounced softening. The optimized composite had a moisture content of 9.8%, a 5 min surface water uptake of 4.7%, and 65% weight loss after 28 days in a preliminary soil-burial test.

Sequential RSM optimization enabled seaweed residue and rejected corn and cassava flours to be formulated and hot-pressed into a molded biocomposite.

1. Introduction

The extensive use of petroleum-based plastics in food packaging is mainly driven by their low cost, ease of processing, and favorable mechanical performance. However, their poor degradability has intensified interest in biodegradable alternatives, particularly for short-life products such as trays, disposable containers, and molded packaging.^1,2 In this context, biomass-derived composites are attractive not only because they reduce dependence on fossil resources, but also because they provide a route to convert low-value industrial and agricultural side streams into functional materials.^3–5 Compared with cast films or coatings, compression-molded or hot-pressed trays are particularly relevant because they allow low-value particulate biomass to be consolidated directly into shape-defined packaging, without the casting or film-forming steps required by many bio-based alternatives.⁶ Within this landscape of biomass resources, residues from marine and agricultural processing represent particularly attractive feedstocks because they remain underutilized despite retaining functional polymeric components. In the marine sector, the hydrocolloid (agar) industry is a notable source of such residues at scale. Seaweed residue (SR), the solid by-product remaining after agar extraction from red agarophyte biomass, represents a distinct secondary feedstock from this industry.⁷ At the global scale, FAO reported algae production at 37.8 million wet tonnes in 2022, with 97 to 99.5% of cultivated production concentrated in Asia.⁸ In Vietnam, Gracilaria cultivation and processing are particularly relevant in the northern coastal provinces, where the cultivation area has been estimated at about 6000 ha, yielding approximately 6350 tonnes of dried raw seaweed per year (equivalent to about 40 000 tonnes fresh weight), with around 10 agar-processing enterprises in Hai Phong producing 1370–1760 tonnes of agar annually.⁹ These figures indicate that agar extraction generates a residue stream of practical significance rather than a marginal by-product. After agar extraction, SR still retains substantial organic matter, including protein, fiber, and carbohydrate-related components, with carbohydrate typically reported as the dominant fraction in agar-processing residues.^7,10 In industrial practice, this composition may be further complicated by process-derived inorganic matter, as filter aids are commonly added during hot-extract clarification and fine filtration.³

Although SR has been explored in some valorization routes, its use has remained comparatively limited and has mainly been directed toward lower-value or non-structural applications.^3,7 Seaweed-industry by-products have been incorporated into lightweight foam concrete as partial replacements for natural sand,¹¹ organic fertilizers or soil amendments, and more recently as sources of bioactive extracts.⁷ These studies confirm that SR still contains recoverable value after agar removal, but they also suggest that direct conversion into molded biocomposites remains much less developed. From a materials perspective, this limitation is understandable: once much of the original gel-forming agar fraction has been removed, the remaining residue has weak self-bonding ability and cannot easily be consolidated into a coherent molded structure on its own.^2,12

Beyond SR, rejected corn kernels and rejected cassava roots represent additional low-value agricultural streams that remain widely available but underutilized for higher-value applications; in this study, these raw materials were milled into corn flour (CoF) and cassava flour (CaF), respectively. Maize is a starch-dominant cereal material, with starch typically accounting for about 68.0% to 74.0% of kernel weight, together with 8.0% to 11.5% protein, 4.5% oil, and a smaller fiber fraction.^13,14 Cassava flour is likewise starch-rich, but unlike purified cassava starch, it still retains non-starch components; reported cassava flour compositions include 81% starch, 1% protein, 1.2% fiber and 2% ash of dried weight.^15,16 Starch- and flour-based biomass materials of this type have been widely used in the development of biocomposites and films. Unlike purified starches, rejected corn and cassava flours retain protein, fiber, lipids, and other non-starch constituents, making them compositionally heterogeneous starch-rich feedstocks for biocomposite applications.

Even with these three solid biomass streams available, a separate binder component is commonly required to consolidate heterogeneous particulate biomass mixtures into a coherent molded network. Hydroxyl-rich polyvinyl alcohol (PVA) has been widely used as such a binder, as it is known to interact with starch- and cellulose-containing systems through hydrogen bonding, which has been reported to improve interfacial compatibility and cohesion in various biocomposite formulations.^17,18 Glycerol is commonly incorporated alongside such binders as a plasticizer, regulating chain mobility and reducing brittleness in starch- or cellulose-based composite systems.^19,20 In parallel, hot pressing (compression molding) has been established as an effective technique for consolidating particulate biomass–polymer mixtures into shape-defined molded products, combining thermal and mechanical action to promote particle bonding and densification within a defined mold geometry.²¹ Together, these components – binder, plasticizer, and consolidation technique – represent the processing-related counterparts to the solid biomass streams described above, each with an established role reported in the literature.

Despite these established roles, most studies on biodegradable biocomposites used purified starches or a single biomass source. Few studies combined heterogeneous marine and agricultural residues in one formulation. Much of the work also focused on cast films or coatings. Compression-molded trays made from such waste streams are less explored. The properties of these products depend on both the formulation and the processing conditions.^22,23 In a system with several biomass solid phases and a small amount of binder, these variables probably act nonlinearly and interact with each other. One-factor-at-a-time experiments cannot resolve this well.²⁴ Response surface methodology (RSM) can evaluate main effects, interactions, and curvature with a limited number of experiments.^25–27 Most RSM studies on biocomposites optimized either the formulation or the processing parameters, not both in sequence.

Accordingly, this study investigated the co-utilization of seaweed residue (SR), rejected corn kernel flour (CoF), and rejected cassava flour (CaF) for the production of a molded biocomposite. A sequential response surface methodology was used. The formulation was optimized first, then the hot-pressing conditions, with thickness, density, hardness, and flexibility as responses. The effect of glycerol content on the optimized composite was further examined to find a balance between rigidity and flexibility. Its moisture content, short-term water absorption, and soil-burial degradation were evaluated as well. Through this approach, the study aims to establish a practical formulation–processing framework for converting heterogeneous marine and agricultural side streams into a molded biocomposite product with potential for tray-type packaging.

2. Materials and methods

2.1. Materials

Seaweed residue (SR) obtained after agar extraction was collected from a jelly powder production facility in Ha Nam Province, Vietnam. The residue was sun-dried to a moisture content of approximately 10–12% and transported to the laboratory for further processing. The inorganic fraction of the collected SR was 10–12% of its dry mass (ash at 550 °C), and the remaining 88–90% was organic matter. This inorganic fraction is attributed mainly to Celite. Celite (diatomaceous earth) was added as a filter aid during agar extraction and remains in the residue and was counted as part of SR in all formulations. The SR was ground and sieved through a 0.25 mm mesh to obtain a uniform powder fraction. Rejected corn kernels and cassava roots were milled to obtain CoF and CaF with a particle size of <0.25 mm in order to retain their native cellulose and starch fractions. Polyvinyl alcohol (PVA-124, GHTECH, China) and glycerol (99%, Xilong, China) were used as a binder and plasticizer, respectively.

2.2. Process of making biocomposite

All solid components were formulated on a dry-mass basis. For each formulation, SR was used as the reference material, while CoF and CaF were added according to the designed CoF/SR and CaF/SR ratios (w/w). PVA was used as the binder and expressed as wt% based on the total dry mass of SR, CoF, and CaF. The PVA solution was prepared by dissolving PVA in distilled water at 85–90 °C under continuous stirring for 60 min to obtain a homogeneous solution with a concentration of 10% (w/v). The required amount of PVA was added as this stock solution, diluted with distilled water so that the total liquid volume was about 0.75 mL g⁻¹ of the total dry blend (SR + CoF + CaF). The PVA solution was gradually added to the dry SR/CoF/CaF blend and mixed mechanically. Glycerol was subsequently incorporated as a plasticizer and mixed until a uniform wet blend was obtained. The glycerol dosage was expressed as mL per g of total dry biomass blend (SR, CoF, and CaF). The mixture was transferred into a custom-designed dumbbell-shaped mold (Fig. 1), with an overall length of 170 mm and a central width of 10 mm, and evenly distributed to minimize thickness variation. The mold was then hot-pressed under the designed processing conditions, including pressing temperature (PTe), pressing time (PTi), and pressing force (PF). After hot pressing, the molded specimens were removed from the mold and conditioned at room temperature for 24 h before characterization (Fig. 2).

Fig. 1. Custom-designed dumbbell-shaped mold and dimensions of the fabricated composite specimen.

Fig. 2. Summary of the biocomposite process.

The samples were characterized in terms of thickness, density, hardness, and flexibility. Thickness was measured at three random positions using a digital caliper (Mitutoyo, Japan), and the average value was recorded. Density was calculated from the mass-to-volume ratio. For hardness and flexibility, specimens were tested on the central (waist) section of the dumbbell-shaped specimen (approximately 40 mm in gauge length and 10 mm in width, with thickness varying according to the formulation), using a Brookfield (AMETEK) CT3 texture analyzer (TexturePro CT software) fitted with a three-point bending fixture (TA-TPB) and a TA3/100 probe, with a support span of 30 mm and a 4500 g load cell. Specimens were tested in rupture mode at a pretest speed of 2.0 mm s⁻¹ and a test speed of 1.0 mm s⁻¹, with a trigger load of 4.5 mN; the test was terminated at the target stop-load condition, which coincided with the point of first fracture. For the purposes of this study, hardness was operationally defined as the peak load (mN) reached during the three-point bending test, while flexibility was defined as the corresponding deflection (mm) at peak load. Both responses were obtained from the same bending measurement.

2.3. Evaluation of influential factors using response surface methodology (RSM)

Response surface methodology (RSM) was employed to optimize biocomposite fabrication in two sequential stages using Design-Expert 13 software (Stat-Ease Inc., Minneapolis, USA). In each stage, a central composite design (CCD) was applied to systematically assess the linear, interaction, and quadratic effects of the selected variables on the product properties, while regression analysis was used to establish predictive models and identify the optimum processing conditions.^28,29 The axial distance was set at α = 1.682, corresponding to a rotatable CCD for three independent variables. In the first stage, RSM was used to optimize the mixing ratio of the main raw materials. In the second stage, RSM was applied to optimize the hot-pressing operating conditions using the optimum formulation obtained from the first stage. All experiments in both stages were carried out in triplicate, and the mean values were used for model development and statistical analysis. The experimental data were fitted to a second-order polynomial equation (eqn (1))

Y = β₀ + β₁A + β₂B + β₃C + β₁₂AB + β₁₃AC + β₂₃BC + β₁₁A² + β₂₂B² + β₃₃C² 1

According to the above equation, Y is the predicted response (thickness, density, hardness, and flexibility), β₁, β₂ and β₃ represent the linear; β₁₂, β₁₃ and β₂₃ represent the interaction; and β₁₁, β₂₂, and β₃₃ represent the quadratic coefficients, respectively. Also, β₀ is the intercept. A, B and C are factors.²⁶

2.3.1. Evaluation of raw material formulation

The first RSM experiment was conducted to optimize the formulation of the biocomposite. Three independent variables were selected: CoF content (A), CaF content (B), and PVA content (C). The investigated ranges of these variables were determined based on preliminary experiments (SI, Section S1): CoF at 0.25–0.75 g g⁻¹ SR, CaF at 0.13–0.38 g g⁻¹ SR, and PVA at 3–6% of the composite mass (defined as the total mass of SR, CoF, and CaF) (Table 1). In these trials, a mixture of SR, CoF, and CaF without PVA (CoF and CaF each at 50% of SR mass, with the same liquid volume as the PVA-containing mixtures) remained loose and could not be molded into an intact specimen, whereas the addition of PVA solution enabled consolidation. At 7% PVA, the specimen could be formed but adhered to the mold and was difficult to demold (Table S1a). Glycerol content was maintained at 0.1 mL g⁻¹ of total dry biomass blend. During this stage, the processing conditions were fixed at a PTe of 100 °C, a PF of 160 N, and a PTi of 3 min. Analysis of variance (ANOVA) was applied to determine the significance of the model terms and to identify the optimum material ratio for the next stage.

Table 1. Central composite design matrix for formulation optimization using coded and actual levels of CoF, CaF, and PVA.

Run	Coded variables			Uncoded variables
	A	B	C	CoF (g g−1 SR)	CaF (g g−1 SR)	PVAa (wt% based on composite mass)
1	1	−1	1	0.75	0.13	6.0
2	−1	−1	−1	0.25	0.13	3.0
3	−1	1	1	0.25	0.38	6.0
4	1	−1	−1	0.75	0.13	3.0
5	1.682	0	0	0.92	0.25	4.5
6	0	0	−1.682	0.50	0.25	2.0
7	0	0	1.682	0.50	0.25	7.0
8	−1.682	0	0	0.08	0.25	4.5
9	0	0	0	0.50	0.25	4.5
10	1	1	1	0.75	0.38	6.0
11	1	1	−1	0.75	0.38	3.0
12	0	0	0	0.50	0.25	4.5
13	0	0	0	0.50	0.25	4.5
14	0	0	0	0.50	0.25	4.5
15	0	0	0	0.50	0.25	4.5
16	0	1.682	0	0.50	0.46	4.5
17	−1	1	−1	0.25	0.38	3.0
18	−1	−1	1	0.25	0.13	6.0
19	0	−1.682	0	0.50	0.04	4.5
20	0	0	0	0.50	0.25	4.5

^aPVA content was calculated based on the total dry mass of SR, CoF, and CaF.

2.3.2. Evaluation of process operating conditions

After determining the optimum formulation, a second RSM experiment was conducted to optimize the hot-pressing conditions. At this stage, the material composition was kept constant, while pressing temperature (PTe, D), pressing time (PTi, E), and pressing force (PF, F) were selected as the independent variables. The investigated ranges of these operating parameters were established based on handling trials to ensure appropriate molding and sample integrity: PTe at 90–110 °C, PTi at 2–4 min, and PF at 120–200 N (Table 2). Glycerol content was maintained at 0.1 mL g⁻¹. The optimum operating conditions were selected based on the combined desirability of the measured responses.

Table 2. Central composite design matrix for hot-pressing optimization using coded and actual levels of pressing temperature, pressing time, and pressing force.

Run	Coded variables			Uncoded variables
	D	E	F	PTe, °C	PTi, min	PF, N
1	1	−1	1	110	2	200
2	−1	−1	−1	90	2	120
3	−1	1	1	90	4	200
4	1	−1	−1	110	2	120
5	1.682	0	0	117	3	160
6	0	0	−1.682	100	3	93
7	0	0	1.682	100	3	227
8	−1.682	0	0	83	3	160
9	0	0	0	100	3	160
10	1	1	1	110	4	200
11	1	1	−1	110	4	120
12	0	0	0	100	3	160
13	0	0	0	100	3	160
14	0	0	0	100	3	160
15	0	0	0	100	3	160
16	0	1.682	0	100	4.7	160
17	−1	1	−1	90	4	120
18	−1	−1	1	90	2	200
19	0	−1.682	0	100	1.3	160
20	0	0	0	100	3	160

2.4. Evaluation of glycerol content

After optimization of the formulation and operating conditions, the effect of glycerol content on product properties was further investigated. Glycerol was added at four levels, namely 0.1, 0.2, 0.3, and 0.4 mL g⁻¹ of total dry biomass blend, while the optimized material ratio and hot-pressing conditions were kept constant. The produced samples were evaluated in terms of thickness, density, hardness, and flexibility in order to determine the most appropriate glycerol level.

2.5. Moisture content, water absorption, and preliminary soil-burial degradation

The moisture content, water absorption, and soil-burial degradation of the optimized composite were evaluated as additional performance indicators. The moisture content was determined from the mass difference before and after oven drying. The sample was weighed before drying as m_i, dried at 105 °C until constant weight, and then weighed again as m_d. Moisture content was calculated on a wet basis using eqn (2):

2

Water absorption was determined by placing 0.5 mL of distilled water onto the surface of the dried specimens at room temperature for 5 min. This protocol reflects short-term surface contact conditions relevant to food-tray applications, in contrast to the prolonged immersion specified in standardized methods such as ASTM D570, and results should be interpreted as an indicator of surface water resistance rather than bulk water uptake. The specimens were gently wiped with tissue paper to eliminate surface water, and weighed as m_t. m₀ was the initial weight of the specimens. Water absorption was calculated using eqn (3):

3

A preliminary soil-burial test was conducted to evaluate the degradation behavior of the optimized composite. Specimens were buried at a depth of 10 cm in sandy loam garden soil, under natural garden-soil moisture conditions associated with active plant cultivation, with ambient soil temperatures of approximately 28–32 °C. Soil moisture was not artificially controlled during the burial period. The specimens were dried to constant weight and weighed before burial as m₀. They were then buried in soil under ambient conditions for 28 days. At weekly intervals, the specimens were removed, gently cleaned to remove adhering soil particles, dried to constant weight (105 °C), and weighed as m_s. Weight loss was calculated using eqn (4):

4

All measurements were performed in three replicates, and the results were reported as mean ± standard deviation.

3. Results and discussion

3.1. Optimization of material formulation by response surface model

To evaluate how formulation affected the structural and mechanical properties of the biocomposite, a first RSM experiment was performed using SR, CoF, CaF, and PVA as the main formulation variables. SR was used as the reference solid phase, while CoF and CaF were incorporated as starch-rich flour fractions to improve the formulation of the molded composite. PVA was included as the binder to promote cohesion among the particulate components.^30,31 Preliminary one-factor-at-a-time screening at fixed levels of the other components (SI, Section S1) showed that omission of CoF, CaF, or PVA produced clearly different behavior – specimens without CoF were soft and prone to breakage, those without CaF had a rougher surface, and those without PVA did not consolidate – supporting their inclusion as formulation variables in the subsequent RSM analysis. Thickness, density, hardness, and flexibility were selected as the target responses because they collectively reflected structural consolidation and the mechanical balance of the molded material. The significance levels of the fitted response surface model terms for these responses are summarized in Table 3.

Table 3. p-Values of the fitted response surface model terms for thickness, density, hardness, and flexibility in the optimization of raw material mixing ratio.

Independent variable	Thickness	Density	Hardness	Flexibility
Linear
A – CoF	0.5208	0.3792	<0.0001	<0.0001
B – CaF	0.0524	0.4359	0.0006	0.0029
C – PVA	<0.0001	<0.0001	<0.0001	0.0004
Interaction
AB	0.1917	0.6724	0.0717	0.0818
AC	0.0637	0.2807	0.0042	0.0011
BC	0.0445	0.6269	0.6078	0.0108
Quadratic
A 2	0.0077	0.1642	0.0026	0.0084
B 2	0.1227	0.4643	0.0648	0.8258
C 2	<0.0001	0.0024	0.0010	0.0001
Model
p-Value	<0.0001	0.0017	<0.0001	<0.0001
F-Value	17.19	7.86	24.99	20.19
Lack of fit	0.3945	0.7626	0.5535	0.7398
R 2	0.9393	0.8762	0.9574	0.9478
Adjusted R2	0.8846	0.7648	0.9191	0.9009
Predicted R2	0.7033	0.5658	0.8454	0.8084
CV%	3.83	5.48	8.6	4.55

The ANOVA results showed that all four quadratic models were statistically significant, with F-values of 17.19, 7.86, 24.99, and 20.19 for thickness, density, hardness, and flexibility, respectively, and all corresponding p-values below 0.01. In addition, the lack of fit was not significant, indicating that the fitted models adequately described the experimental data within the investigated formulation domain. The coefficients of determination were 0.9393 for thickness, 0.8762 for density, 0.9574 for hardness, and 0.9478 for flexibility, while the adjusted R² values remained reasonably close to the corresponding R² values. The predicted R² values further supported the adequacy of the models, although the density model showed lower predictive performance (0.5658) than thickness (0.7033), hardness (0.8454), and flexibility (0.8084). Accordingly, density trends were interpreted cautiously and considered together with thickness when evaluating structural changes in the composite. The predicted-versus-actual plots (Fig. 3) also showed satisfactory agreement between the experimental and fitted values for all four responses, supporting the use of the developed models for response interpretation and optimization.

Fig. 3. Predicted vs. actual for thickness (mm) (a), density (g cm⁻³) (b), hardness (mN) (c) and flexibility (mm) (d).

3.1.1. Structural responses: thickness and density

Thickness was used as the primary structural response because it most directly reflected the extent of specimen consolidation during molding. In the thickness model, the linear term of PVA (C) and the quadratic term C² were both highly significant (p < 0.0001), while BC (p = 0.0445) and A² (p = 0.0077) also contributed significantly to the fitted response. The fitted regression equation in terms of coded factors was:

Thickness = 1.6 + 0.0101A + 0.0335B − 0.126C – 0.0279AB – 0.0415AC − 0.0457BC − 0.0493A² − 0.025B² − 0.117C². 5

The negative linear coefficient of PVA (−0.126C) was the largest in magnitude among the retained terms, indicating that increasing PVA content markedly reduced composite thickness. This effect became even more pronounced at higher PVA levels, as shown by the negative quadratic coefficient of C² (−0.117). The contour plot (Fig. 4a) supports this interpretation, since the response gradient was steeper along the PVA axis than along the CaF axis, confirming that thickness was governed primarily by the binder phase rather than by variation in the solid flour content. From a materials perspective, this behavior is consistent with PVA acting as the primary consolidating agent, promoting cohesion and compressibility of the particulate mixture – as also indicated by the PVA-free control (Table S1a), which failed to consolidate at the formulation tested, despite comparable moisture content. As binder content increased, this cohesion likely allowed more effective particle rearrangement and void collapse during molding.^19,28

Fig. 4. Contour plots showing the interactive effects of CaF (B) and PVA concentration (C) on the thickness (a) and density (b) of the biocomposite. Factor levels are shown in coded units (−1 to +1), where −1 and +1 correspond to the actual minimum and maximum values of each variable within the studied range: CoF, 0.25–0.75 g g⁻¹ SR; CaF, 0.13–0.38 g g⁻¹ SR; PVA, 3–6% of composite mass.

Density followed the same general response pattern as thickness, but with weaker predictive strength. In the density model, only C (p < 0.0001) and C² (p = 0.0024) were significant, and the final regression equation was:

Density = 0.9167 + 0.0118A + 0.0104B − 0.0905C − 0.0073AB − 0.0191AC + 0.0084BC − 0.0187A² − 0.0095B² − 0.0503C². 6

As in the thickness response, PVA was the only formulation variable exerting a statistically significant effect on density within the studied range, and the negative coefficients of C and C² indicate that density decreased as PVA increased. The contour plot (Fig. 4b) likewise showed that density varied mainly along the PVA direction and was only weakly affected by the flour variables. However, the lower predicted R² of the density model suggests that this response was less directly captured by the regression than thickness. This difference is reasonable because density was calculated from both sample mass and geometrical volume. Unlike thickness, which mainly reflected structural consolidation, density was also influenced by the mass contribution and intrinsic packing behavior of the formulation components. In the present system, SR, CoF, CaF, and PVA differed in composition and likely differed in local packing, void generation, and mass distribution within the molded body.¹ Density was calculated from both specimen mass and volume; variations in material distribution and packing may therefore have introduced additional variability and partly contributed to the lower predictive performance of the density model.³² Nevertheless, density and thickness showed a similar response to PVA, although this trend was more clearly captured by the thickness model. The decrease in density with increasing PVA may reflect concurrent changes in specimen mass and volume during molding; however, their individual contributions were not quantified.

3.1.2. Mechanical response: hardness

Hardness, defined here as the peak load during three-point bending, was used as an operational indicator of the load-bearing resistance of the molded specimen. In the hardness model, the linear terms of CoF (A), CaF (B), and PVA (C) were all significant (p < 0.0001, p = 0.0006, and p < 0.0001, respectively), together with the interaction term AC (p = 0.0042) and the quadratic terms A² (p = 0.0026) and C² (p = 0.0010). The fitted quadratic equation in coded factors was:

Hardness = 9547.34 + 1274.46A + 932.32B − 1937.46C − 501.97AB − 918.32AC − 132.03BC − 739.90A² − 385.24B² − 850.78C². 7

The positive coefficients of CoF (+1274A) and CaF (+932B) indicate that increasing the two flour phases increased hardness, whereas the negative coefficient of PVA (−1937C) indicates that increasing binder content reduced the specimen's resistance to fracture under bending load. This response is consistent with a greater contribution of CoF and CaF to load-bearing resistance within the molded composite. Both were used as unrefined flour rather than purified starches, so they retained insoluble structural matter (such as cellulose and hemicellulose), which may have contributed to the higher peak bending load observed at increasing flour content.³² By contrast, the negative coefficient of PVA suggests that, at higher levels, the binder-rich phase increased compliance, thereby lowering the peak load required to fracture the specimen under bending.³³

The contour plot (Fig. 5) showed that the highest hardness values were located in the high-CoF/low-PVA region, indicating that rigidity was favored when the flour-derived framework dominated over the binder phase. The interaction term AC further showed that the hardening effect of CoF became less effective as PVA increased. This trend was also evident in the interaction plot (Fig. S1a, SI), where hardness increased more steeply with CoF at low PVA than at high PVA. The perturbation plot confirmed the same hierarchy of influence around the center point: PVA exerted the strongest negative effect, whereas both CoF and CaF contributed positively, with CoF showing the larger effect across the studied range (which was twice as wide as that of CaF). Overall, the highest peak bending loads were obtained at relatively high flour content and low PVA levels, indicating that this formulation range favored greater resistance to bending-induced fracture.

Fig. 5. Contour plot (a) showing the effect of CoF (A) and PVA (C) on hardness at the center level of CaF (B), together with the perturbation plot (b) for hardness. Factor levels are shown in coded units (−1 to +1), where −1 and +1 correspond to the actual minimum and maximum values of each variable within the studied range: CoF, 0.25–0.75 g g⁻¹ SR; CaF, 0.13–0.38 g g⁻¹ SR; PVA, 3–6% of composite mass.

3.1.3. Mechanical response: flexibility

Flexibility, defined as the deformation at peak load, reflected the extent of specimen deformation before reaching the maximum bending load. The flexibility model included significant linear effects of CoF (A), CaF (B), and PVA (C) (p < 0.0001, p = 0.0029, and p = 0.0004, respectively), together with the interaction terms AC (p = 0.0011) and BC (p = 0.0108), and the quadratic terms A² (p = 0.0084) and C² (p = 0.0001). The fitted quadratic equation in coded factors was:

Flexibility = 0.8705 − 0.0801A − 0.0430B + 0.0577C + 0.0278AB + 0.0646AC + 0.0448BC − 0.0350A² + 0.0024B² + 0.0647C² 8

The negative coefficients of CoF (−0.0801A) and CaF (−0.0430B) show that more flour reduced flexibility. The positive coefficient of PVA (+0.0577C) shows that more PVA increased deformability. The positive AC term means PVA partly offset the stiffening effect of CoF. Similarly, the positive BC interaction indicates that the effect of PVA on flexibility depended on CaF content. The contour plots (Fig. 6a and b) show the highest flexibility at high PVA and low flour content. The perturbation plot (Fig. 6c) shows that CoF caused the largest decrease across the studied range, followed by CaF.

Fig. 6. Contour plots showing the interactive effects of CoF (A) and PVA concentration (C) (a), CaF (B) and PVA concentration (C) (b) on the flexibility (mm) of the biocomposite and the perturbation plot for flexibility (c). Factor levels are shown in coded units (−1 to +1), where −1 and +1 correspond to the actual minimum and maximum values of each variable within the studied range: CoF, 0.25–0.75 g g⁻¹ SR; CaF, 0.13–0.38 g g⁻¹ SR; PVA, 3–6% of composite mass.

The different experimental ranges of CoF and CaF were established from the preliminary screening tests (Section S1). CaF was restricted to a narrower range because higher levels produced rigid specimens with poor bendability, whereas CoF could be investigated over a wider range before a comparable loss of bendability was observed. Accordingly, direct comparison of the coded coefficients should be made with caution. When expressed per g g⁻¹ SR at the design center, the linear effects on flexibility were similar, at approximately −0.32 mm for CoF and −0.34 mm for CaF (eqn (8)). The two flours nevertheless showed different response patterns: CoF exhibited significant quadratic effects and interactions with PVA, whereas CaF behaved more linearly and interacted with PVA mainly in the thickness and flexibility responses.

These differences may partly reflect their distinct compositions and structures. Maize kernels contain 8.0–11.5% protein, 4.5% oil, and fibrous pericarp and germ fractions, whereas cassava flour is predominantly starch with much lower protein and fiber contents.^16,30 In maize endosperm, starch granules are embedded in a protein matrix,³⁴ while cassava parenchyma contains comparatively little surrounding protein. The protein and lipids may limit granule swelling and water uptake. Oil and protein on particle surfaces may also reduce wetting by the PVA solution. Pericarp fragments may stiffen the composite. These are possible explanations only. They were not checked by structural characterization in this study.³⁵

3.1.4. Numerical optimization of formulation

Given the different optimization trends of the four responses, a desirability-based numerical optimization was applied to determine the most suitable formulation. Thickness and density were set to be minimized, whereas hardness and flexibility were set to be maximized. Lower thickness and density were targeted to favor a lightweight and material-efficient molded product, while mechanical performance was given higher priority in the desirability function. All factors were kept within their studied ranges. Importance was set to 3 for thickness and density, 5 for hardness, and 4 for flexibility (on a 1–5 scale, with 5 being the most important), reflecting the priority given to mechanical performance over structural compactness. Hardness was assigned the highest importance because the intended molded tray application requires sufficient mechanical resistance during handling. A total of 47 candidate solutions were generated within the studied formulation domain.

The selected optimum corresponded to CoF = 0.59 g g⁻¹ SR, CaF = 0.38 g g⁻¹ SR, and PVA = 6%, with predicted response values of 1.289 mm for thickness, 0.777 g cm⁻³ for density, 7018 mN for hardness, and 0.997 mm for flexibility. The overall desirability of this solution was 0.555. Although this desirability value was only moderate, it represented the most suitable compromise among the four target responses within the studied formulation range. Given the lower predictive performance of the density model, a sensitivity analysis was performed to evaluate its influence on the selected formulation. Reducing the importance of density from 3 to 1, or excluding density from the desirability function, resulted in only minor changes in the optimum (Table S4). This result indicates that the lower predictive accuracy of density did not materially affect selection of the optimum formulation. The optimization reflected a trade-off between mechanical and structural responses: increasing flour content generally increased hardness but reduced flexibility, whereas increasing PVA improved flexibility and consolidation while lowering hardness. Notably, PVA, although present at only 3–6%, significantly affected all four responses, indicating a strong influence of the binder fraction on the overall behavior of the composite. Similar effects of low-dose polymeric additives have been reported in RSM-based composite formulations.³⁶ As a result, no single formulation could simultaneously maximize all desired attributes to the same extent. The selected condition was therefore considered the most appropriate formulation for the subsequent optimization of hot-pressing parameters in stage 2.

3.2. Optimization of processing conditions by RSM

After the optimum formulation had been identified in stage 1, a second RSM experiment was conducted to optimize the hot-pressing conditions of the SR-based biocomposite. At this stage, the formulation was kept constant, and the processing variables, namely PTe (D), PTi (E), and PF (F), were evaluated in order to determine how the optimized formulation responded to changes in hot-pressing temperature, time, and force during final molding. The significance levels of the fitted response surface model terms are summarized in Table 4. All four response models were statistically significant, with F-values of 28.16, 9.06, 12.20, and 15.33 for thickness, density, hardness, and flexibility, respectively, and the lack of fit was not significant for any response. The coefficients of determination were 0.9620 for thickness, 0.6295 for density, 0.9165 for hardness, and 0.9324 for flexibility, while the corresponding predicted R² values were 0.7625, 0.3654, 0.6847, and 0.6723. The predicted R² values were higher for thickness, hardness, and flexibility than for density, indicating lower predictive performance of the density model. The predicted-versus-actual plots (Fig. 7) also showed acceptable agreement between the experimental and fitted values for all four responses.

Table 4. Significance level (P) calculated for investigating processing-condition optimization of composite.

Independent variable	Thickness	Density	Hardness	Flexibility
Linear
D – pressing temperature, PTe	<0.0001	0.0253	0.4307	<0.0001
E − pressing time, PTi	0.3219	0.5799	0.7830	0.0040
F – pressing force, PF	<0.0001	0.0003	0.8982	0.1319
Interaction
DE	0.0019	—	0.5326	0.0355
DF	0.1447	—	0.0335	0.0403
EF	0.4914	—	0.5304	0.0218
Quadratic
D 2	0.6669	—	<0.0001	0.0404
E 2	0.0013	—	<0.0001	0.0096
F 2	0.0016	—	0.0212	0.0113
Model
p-Value	<0.0001	0.0010	0.0003	<0.0001
F-Value	28.16	9.06	12.2	15.33
Lack of fit	0.1178	0.1840	0.7347	0.4144
R 2	0.9620	0.6295	0.9165	0.9324
Adjusted R2	0.9279	0.5600	0.8414	0.8716
Predicted R2	0.7625	0.3654	0.6847	0.6723
CV%	2.07	5.30	4.95	5.47

Fig. 7. Predicted vs. actual for thickness (mm) (a), density (g cm⁻³) (b), hardness (mN) (c) and flexibility (mm) (d).

3.2.1. Structural responses: thickness and density

Thickness reflected the extent of dimensional consolidation achieved during hot pressing. In the thickness model, the linear terms of PTe (D) and PF (F) were highly significant (p < 0.0001), while the interaction term DE (p = 0.0019) and the quadratic terms E² (p = 0.0013) and F² (p = 0.0016) also contributed significantly to the fitted response. The fitted quadratic equation in coded factors was:

Thickness = 1.37 + 0.0570D + 0.0083E − 0.0957F − 0.0434DE + 0.0164DF + 0.0074EF + 0.0034D² + 0.0343E² + 0.0332F². 9

The positive coefficient of PTe (+0.057D) indicates that increasing pressing temperature tended to increase thickness, whereas the negative coefficient of PF (−0.0957F) indicates that increasing pressing force reduced thickness. Within the investigated range, the larger absolute coefficient of PF suggests that thickness was more strongly affected by mechanical compression than by temperature. Higher PF likely promoted particle rearrangement and reduced void space within the molded structure, resulting in a thinner specimen.²⁸ The aqueous PVA solution may also have promoted starch swelling or partial gelatinization during hot pressing. However, the PVA-free control prepared at the same liquid-to-solid ratio did not form a cohesive specimen after pressing at 100 °C (Table S1a, single formulation tested), indicating that moisture-assisted starch transformation alone was insufficient for consolidation in that mixture. The significant DE interaction further shows that the effect of PTe on thickness depended on pressing time (Fig. 8a). Changes in binder softening and moisture redistribution may have contributed to this interaction, although post-pressing moisture was not measured and the underlying mechanism remains unverified.

Fig. 8. Contour plots showing the interactive effects of PTe and PTi on thickness (a), and PTe and PF on density (b) of the biocomposite. Factor levels are shown in coded units (−1 to +1); actual ranges were PTe, 90–110 °C; PTi, 2–4 min; and PF, 120–200 N.

Density showed a similar general trend, but remained a secondary structural response. In the density model, only PTe (p = 0.0253) and PF (p = 0.0003) were significant, and the final regression equation was:

Density = 0.9221 + 0.0326D − 0.0075E − 0.0603F. 10

With the formulation fixed, density was significantly affected by PTe and PF, whereas PTi had no significant effect within the investigated range. The positive coefficient of PTe (+0.0326D) indicates a slight increase in density with temperature, while the negative coefficient of PF (−0.0603F) indicates a decrease in density as pressing force increased (Fig. 8b). The latter trend, despite the concurrent reduction in thickness, suggests that changes in specimen mass and material redistribution may have accompanied mechanical compression rather than density being controlled by volume reduction alone. Similarly, the effect of PTe may reflect changes in packing and moisture redistribution during heating. However, these mechanisms were not measured directly, and the relatively low predicted R² of the density model limits further mechanistic interpretation.

3.2.2. Mechanical response: hardness

The peak bending load depended mainly on the combined effects of hot-pressing temperature, time, and force. In the hardness model, none of the linear terms D, E, and F were significant (p > 0.05), whereas the interaction term DF (p = 0.0335) and the quadratic terms D², E², and F² were all significant (p < 0.0001, p < 0.0001, and p = 0.0212, respectively). The fitted quadratic equation in coded factors was:

Hardness = 12380 − 117.85D − 40.60E − 18.83F − 121.18DE + 461.67DF − 121.85EF − 1140.80D² − 917.20E² − 381.41F². 11

The absence of significant linear coefficients indicates that hardness did not increase or decrease monotonically with any individual processing variable. The significant positive DF interaction (+461.67DF) indicates that the effect of pressing temperature depended on pressing force. However, the strong negative quadratic terms for PTe (−1140D²), PTi (−917E²), and PF (−381F²) produced a clear intermediate optimum rather than a continuous increase in hardness with processing intensity. The contour plot (Fig. 9) showed a central high-hardness region rather than a simple increasing or decreasing trend, and the interaction plot (Fig. S1b, SI) confirmed that the effect of PTe depended on the level of PF. From a materials perspective, this behavior suggests that the formulated material required sufficient thermal and mechanical input to achieve effective consolidation, whereas excessive treatment may reduce the specimen's resistance to bending-induced fracture. Similar behavior has been reported in thermoformed fiber-based materials, where intermediate processing intensity improved inter-fiber adhesion and densification, while overly severe conditions led to loss of extensibility or lower mechanical performance.^28,37,38 Thus, hardness was maximized only when consolidation was sufficient but not excessive.

Fig. 9. Contour plot showing the effects of pressing temperature (D) and pressing force (F) on hardness at the center level of pressing time (E) (a) and the perturbation plot for hardness (b). Factor levels are shown in coded units (−1 to +1); actual ranges were PTe, 90–110 °C; PTi, 2–4 min; and PF, 120–200 N.

3.2.3. Mechanical response: flexibility

For the mechanical response, flexibility was influenced more strongly by thermal history than by mechanical compression alone. In the flexibility model, the linear terms of PTe (D) and PTi (E) were significant (p < 0.0001 and p = 0.0040, respectively), whereas the linear term of PF (F) was not significant (p = 0.1319). However, all interaction terms, namely DE, DF, and EF, were significant (p = 0.0355, p = 0.0403, and p = 0.0218, respectively), together with the quadratic terms D², E², and F² (p = 0.0404, p = 0.0096, and p = 0.0113, respectively). The fitted quadratic equation in coded factors was:

Flexibility = 0.9426 − 0.1160D − 0.0477E + 0.0211F + 0.0408DE − 0.0395DF − 0.0455EF − 0.0294D² − 0.0400E² − 0.0387F². 12

The negative linear coefficients of PTe (−0.116D) and PTi (−0.0477E) indicate that increasing these two variables tended to reduce flexibility, with PTe showing the stronger adverse effect. Higher PTe and longer PTi may have promoted greater moisture loss during pressing, which could reduce the plasticizing contribution of residual water and consequently decrease specimen deformability.³⁹ By contrast, PF did not exert a significant independent linear effect, indicating that its influence on flexibility emerged mainly through interaction with the thermal variables rather than as a single-factor response. The positive interaction coefficient of DE (+0.0408DE) suggests that the combined effect of PTe and PTi was not purely additive, whereas the negative coefficients of DF (−0.0395DF) and EF (−0.0455EF) indicate that increasing PF together with higher PTe or longer PTi further reduced flexibility. The combined effects of higher thermal and mechanical input were associated with lower deformation at peak load, consistent with the significant DF and EF interactions.⁴⁰ In addition, the negative quadratic coefficients of D² (−0.0294D²), E² (−0.0400E²), and F² (−0.0387F²) indicate that flexibility declined as the processing variables moved away from the intermediate region, particularly toward more severe operating conditions.

This trend was also evident in the contour plot (Fig. 10), where the region of higher flexibility was located at lower PTe and shorter PTi, while flexibility gradually decreased toward higher PTe and longer pressing duration. The interaction plots (Fig. S1c–e, SI) further showed that the effect of PTe became more detrimental when combined with either longer PTi or higher PF, confirming that flexibility was governed by the combined thermal and mechanical history imposed during molding. The perturbation plot supported this interpretation by showing that flexibility was most sensitive to PTe, followed by PTi, whereas PF produced a weaker but still noticeable curvature effect around the center point. From a processing perspective, these results suggest that excessive thermal exposure promoted a stiffer, less deformable structure. Greater moisture loss during prolonged heating may partly explain this response; however, post-pressing moisture content was not measured, and this interpretation remains tentative.¹² Under the investigated conditions, flexibility was therefore better preserved under relatively mild thermal conditions, while the role of PF was mainly expressed through its interaction with PTe and PTi rather than as an independent effect.

Fig. 10. Contour plots showing the interactive effects of PTi and PTe on the flexibility of the biocomposite (a) and the perturbation plot for flexibility (b). Factor levels are shown in coded units (−1 to +1); actual ranges were PTe, 90–110 °C; PTi, 2–4 min; and PF, 120–200 N.

3.2.4. Numerical optimization of processing conditions

The same desirability criteria and response importance values used in Stage 1 were applied to the processing optimization. The optimum condition was identified at a PTe of 95.25 °C, a PTi of 2.72 min, and a PF of 187.48 N, with predicted responses of 1.280 mm for thickness, 0.876 g cm⁻³ for density, 11 781.537 mN for hardness, and 1.024 mm for flexibility. The overall desirability of the selected solution was 0.739, indicating a satisfactory compromise among the target responses within the investigated processing domain. Sensitivity analysis showed only minor shifts in the optimum when the contribution of density was reduced or excluded: PTe remained close to 95 °C, PTi near 2.7 min, and PF ranged from approximately 178 to 187 N (Table S4). The selected condition was therefore used for subsequent experimental validation. Desirability-based optimization has also been widely applied in multi-response RSM studies to balance competing product responses.⁴¹

3.3. Verification experiments and validation of the model equations

To validate the adequacy of the model equations obtained from the two-stage RSM optimization, verification experiments were carried out under the overall optimum conditions identified in this study. In stage 1, the optimum formulation consisted of 0.59 g CoF per g SR, 0.38 g CaF per g SR, and 6% PVA, while in stage 2, the optimum hot-pressing conditions were 95.25 °C, 2.72 min, and 187.48 N. Verification experiments were therefore conducted using this optimized formulation and processing combination, and the experimental results were compared with the corresponding values predicted by the fitted models.

Thickness, hardness, and flexibility showed good agreement between predicted and experimental values, with absolute percentage errors of 3.01%, 3.85%, and 4.5%, respectively (Table 5). Density showed a larger prediction error of 8.7%, consistent with the lower predictive performance of its RSM model. Since each response was modeled independently, the greater uncertainty associated with density did not affect the prediction of the other responses. Moreover, reducing or excluding density from the desirability function caused only minor changes in the optimum formulation and processing conditions (Table S4), confirming that the density prediction error did not materially affect the overall optimization outcome.

Table 5. Predicted and experimental responses obtained under the optimized formulation and hot-pressing conditions.

Responses	Unit	Predicted value	Experimental value	Percentage errora
Thickness	mm	1.280	1.32 ± 0.04	+3.01%
Density	g cm−3	0.876	0.959 ± 0.02	+8.7%
Hardness	mN	11 782	11 345 ± 796	−3.85%
Flexibility	mm	1.024	0.98 ± 0.08	−4.5%

^aPercentage error = [(experimental – predicted)/experimental] × 100.

3.4. Effect of glycerol content on the properties of the SR-based biocomposite

The effect of glycerol content on the physical and mechanical properties of the optimized composite is shown in Fig. 11. Within the investigated range, glycerol had little influence on the structural responses. Thickness remained nearly constant, varying only from approximately 1.30 to 1.34 mm as glycerol content increased from 0.1 to 0.4 mL g⁻¹, while density changed only slightly, from about 0.98 to 1.10 g cm⁻³. These results indicate that glycerol did not substantially alter the overall compactness or external dimensions of the molded composite under the fixed formulation and hot-pressing conditions. In contrast, glycerol had a much stronger effect on the mechanical responses. Hardness increased slightly from 0.1 to 0.2 mL g⁻¹ and reached a maximum of approximately 11 500 mN at 0.2 mL g⁻¹, representing the highest peak bending load among the tested glycerol levels without causing excessive softening. However, further increases in glycerol reduced hardness markedly, to about 10 400 mN at 0.3 mL g⁻¹ and approximately 6200 mN at 0.4 mL g⁻¹. Flexibility showed the opposite trend. It remained low at 0.1–0.2 mL g⁻¹, increased slightly at 0.3 mL g⁻¹, and then rose sharply to around 7.2 mm at 0.4 mL g⁻¹, indicating a pronounced increase in deformability. This behavior is characteristic of progressive plasticization. At higher glycerol levels, the decrease in peak bending load and increase in deformation were consistent with progressive plasticization of the composite. The sharp decrease in hardness together with the abrupt increase in flexibility at 0.4 mL g⁻¹ is characteristic of over-plasticization. Among the tested levels, 0.2 mL g⁻¹ was selected as the most balanced dosage because it maintained the highest hardness while avoiding the excessive softening observed at higher glycerol contents.

Fig. 11. Effect of glycerol content on structural (a) and mechanical properties of the SR-based biocomposite (b).

3.5. Moisture content, water absorption, and preliminary soil-burial degradation of the optimized composite

The optimized composite (Fig. 12) was further evaluated for moisture content, water absorption, and soil-burial degradation. The moisture content of the optimized sample, determined from the mass difference before and after oven drying at 105 °C, was 9.8 ± 0.09%. This result indicates that a certain amount of residual moisture remained in the molded structure after conditioning, which is expected for biomass-based composites containing hydrophilic components. The water absorption of the optimized composite was 4.7 ± 0.12%, indicating limited short-term surface water uptake under the tested conditions despite the presence of hydrophilic components such as SR, CoF, CaF, and PVA.

Fig. 12. Photograph of the optimized molded composite specimen (a) and tray (b).

The preliminary soil-burial test showed a progressive decrease in residual dry weight over 28 days (Fig. 13), indicating gradual loss of structural integrity under soil-burial conditions. This degradation may be attributed to water penetration, microbial activity, and progressive weakening of the SR-, CoF-, and CaF-based molded composite. Correspondingly, weight loss increased with burial time and reached 65.0 ± 0.8% after 28 days. The remaining 35% of the dry weight after 28 days may partly reflect the short duration of this preliminary test. It may also include inert inorganic matter inherited from SR, mainly Celite. Celite is a non-biodegradable diatomaceous silicate added as a filter aid during agar extraction.⁴² The inorganic fraction of SR was 10–12% (Section 2.1), which is roughly 5% of the dry mass of the optimized composite (estimated from the formulation). This explains only a small part of the remaining mass. The rest may come from slowly degrading components, which were not investigated. Celite was not measured separately, and its effect on the mechanical properties and biodegradability of the composite was not isolated. This is a limitation of the study. Overall, the observed weight loss indicates substantial degradation under the tested soil-burial conditions, although the 28-days test does not constitute a standardized assessment of biodegradability or compostability.

Fig. 13. Changes in residual dry weight and weight loss of the optimized composite during 28 days of soil burial.

4. Conclusion

A heterogeneous, poorly self-bonding waste stream – seaweed residue combined with rejected corn and cassava flour – was consolidated into a molded biocomposite using only a small amount of PVA binder, through a sequential formulation-then-processing RSM strategy. The optimum formulation (0.59 g CoF per g SR, 0.38 g CaF per g SR, 6% PVA) and hot-pressing condition (95 °C, 2.7 min, 187 N) were experimentally verified, with prediction errors within 5% for thickness, hardness, and flexibility and 8.7% for density. PVA, although present at only 3–6%, was the only formulation variable with a significant effect on all four responses, and its omission, unlike that of CoF or CaF, prevented consolidation in the one-factor screening trials. CoF and CaF also contributed differently to stiffness and flexibility despite their similar starch content. In the hot-pressing stage, hardness and flexibility were governed mainly by interaction and quadratic effects rather than by any single variable, indicating an intermediate optimum rather than a monotonic response to temperature, time, or force. Glycerol dosage and the moisture, water-uptake, and soil-burial behavior of the composite further indicate a hydrophilic but degradable material suited to short-life, low-moisture applications.

These conclusions rest on statistical trends and simple control trials rather than direct structural evidence; the proposed mechanisms of PVA-mediated cohesion and the differing roles of CoF and CaF should be further evaluated by SEM or FTIR, and the water-absorption and soil-burial tests were preliminary rather than standardized. Beyond this specific formulation, the two-stage RSM strategy itself is a transferable approach: it offers a practical route for turning other heterogeneous, poorly self-bonding marine and agricultural residues into molded, biodegradable packaging materials.

Author contributions

Nhung Thi-Tuyet Hoang: conceptualization, methodology, formal analysis, writing – editing. Anh Thi-Kim Tran: data curation, validation. Duy-Dat Nguyen: validation and editing.

Conflicts of interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability

The data supporting this article have been included as part of the supplementary information (SI). Supplementary information: preliminary screening results, RSM analyses, sensitivity analysis, and detailed experimental data for the formulation and hot-pressing optimization, glycerol evaluation, validation experiments, short-term surface water uptake, and soil-burial test. See DOI: https://doi.org/10.1039/d6ra03696a.