This study followed a sequential experimental workflow. First, soil samples were collected and prepared, and four compost types were selected as organic amendments. Second, the soil and compost were mixed and incubated for 90 d. Third, CO₂ emissions were monitored periodically throughout the incubation period. Fourth, upon completion of the incubation, a comprehensive suite of physicochemical properties of the incubated soils was analyzed. Fifth, the relationships between CO₂ emissions and soil physicochemical properties were examined. Finally, multiple regression analysis was used to quantify the independent effects of each property and calculate their relative contribution weights to CO₂ emissions. Details of the materials, CO₂ measurement, analytical methods, and statistical procedures are provided in Sect. 2.2–2.5, respectively.

Using a five-point random sampling strategy, a composite soil sample was collected exclusively from the 10–20 cm subsurface layer (excluding the 0–10 cm topsoil) at the Agricultural Research Center of the Universiti Malaya and named ARC-soil. The 0–10 cm topsoil was intentionally omitted to minimize spatial heterogeneity caused by surface litter, rhizosphere activity, and frequent anthropogenic disturbances (e.g., tillage, weeding), which could otherwise mask the specific effects of compost incorporation. The 10–20 cm layer represents a more homogenized baseline for evaluating the soil physicochemical changes and CO₂ emission patterns investigated in this study. The soil is a highly weathered, acidic ultisol typical of the humid tropics, with low natural fertility, a clayey texture, and a mineralogy dominated by kaolinite and oxides. The sample was air-dried, disaggregated, sieved (<2 mm), and stored airtight at 3 °C until use. The methods for testing soil physico-chemical properties are described in Sect 2.3.

Four types of compost were obtained from the composting facility at the Faculty of Science, Universiti Malaya. They were produced from distinct feedstocks and designated as: FWC, primarily derived from food waste; GLC, produced from a mixture of goat manure and shredded leaves; CDC, originating from chicken dung and amended with sawdust and rice bran as bulking agents; and SL, generated from dewatered food and beverage wastewater sludge. The four composts (FWC, GLC, CDC, and SL) used in this study were purchased from the Faculty of Science, Universiti Malaya. Detailed information on their production processes is not disclosed by the supplier. However, it is known that CDC and SL are commercially available products, and additional phosphorus and potassium were supplemented during their production. The physico-chemical properties of the composts, characterizing their quality and maturity, are summarized in Table 1. These parameters indicate that all four composts were well-stabilized and suitable for the subsequent experiments.

The experiment involved preparing soil-compost mixtures at two amendment ratios (2.5 % and 10 %, w/w), both adjusted to a total weight of 200 g per incubation unit. The two amendment ratios (2.5 % and 10 %, w/w) were selected based on the following: (1) the 2.5 % ratio corresponds to a typical field application rate of approximately 50 t ha⁻¹, which is commonly used in local agricultural practice; (2) the 10 % ratio was chosen to amplify the treatment effects, allowing for clearer observation of compost-induced changes in soil properties and CO₂ emissions under controlled incubation conditions. Each of the four composts was tested at both ratios, creating eight distinct amended mixtures. These, along with a pure soil control (200 g), constituted the nine core treatments, each replicated three times for a total of 27 units. The soils were incubated at a constant temperature of 28±2 °C. The CO₂ efflux was measured on 12 occasions over the 90 d incubation period, specifically on days 1, 5, 7, 14, 28, 40, 47, 52, 60, 70, 80, and 90.

The CO₂ emitted from the soil was trapped in 20 mL of 1 mol L⁻¹ NaOH. After a 24 h accumulation period, BaCl₂ was added to precipitate the trapped carbonate as BaCO₃. The residual NaOH was then titrated with standardized HCl to quantify the CO₂ evolved. These procedures were performed as described by Cheng et al. (2025).

Soil pH and electrical conductivity (EC) were measured in a 1:2.5 soil-water suspension (Allison, 1954). Total carbon (TC) was determined by dry combustion using an elemental analyzer (TruSpec Micro, LECO). TOC was determined by wet oxidation (Walkley and Black, 1934). Organic matter (OM) contents were calculated based on TOC values (Estefan et al., 2013) (OM (%) = 1.742 × TOC (%)). AP was extracted with 0.5 M NaHCO₃ and measured spectrophotometrically (Olsen et al., 1954). Ammonium (NH4+–N) and nitrate (NO3-–N) were extracted with 2 mol⋅L⁻¹ KCl and analyzed by a continuous flow analyzer (Mulvaney, 1996). Available potassium (AK) was extracted with 1 mol L⁻¹ NH₄OAc and quantified via flame atomic absorption spectroscopy (AAS) (Helmke and Sparks, 1996). Cation exchange capacity (CEC) was determined by the ammonium acetate (NH₄OAc) saturation method at pH 7.0 (Chapman, 1965). Quantitative measurements of bulk density were not performed because the visual difference was so pronounced that sample processing for other analyses was prioritized. Nevertheless, the visual observation suggests a qualitative improvement in soil structure following compost addition.

To determine whether compost type and application rate significantly affected soil physicochemical properties, one-way ANOVA was performed on all measured parameters (pH, moisture, EC, CEC, TOC, TC, OM, NH4+–N, NO3-–N, AP, and AK) across the nine treatments (unamended ARC-soil control plus FWCS, GLCS, CDCS, and SLS each at 2.5 % and 10 %, w/w). Before running ANOVA, the Shapiro–Wilk test confirmed normality of residuals (p>0.05 for all variables) and Levene's test confirmed homogeneity of variances (p>0.05 for all variables), thus satisfying the prerequisites for parametric testing. Where assumptions held, Tukey's HSD post-hoc test was applied for pairwise comparisons among treatment means, with statistical significance defined at p<0.05. All analyses were conducted using SPSS version 26.0 (IBM Corp., Armonk, NY, USA).

A standard multiple linear regression model (Montgomery et al., 2021) was employed to evaluate the effects of soil properties on the response variable, following the general form: 1Y=ξ0+∑ξiKi+ϵ. In this formulation, Y is the response variable (cumulative CO₂ emission). Each Ki corresponds to an explanatory or predictor variable included in the model. The parameters ξi quantify the partial effect of their respective Ki and Y, holding other variables constant. Finally, ϵ captures all unexplained variation, commonly known as the model residual.

Prior to model interpretation, the following assumptions were verified: (1) significant correlations between predictors and the response; (2) independent, normally distributed, and homoscedastic residuals. Normality was supported by the histogram and normal P–P plot of standardized residuals (Fig. S1a and b in the Supplement), residual independence was confirmed by the Durbin-Watson statistic (value of 2.597, close to 2; Table S1 in the Supplement), and homoscedasticity was examined using scatter plots of standardized residuals vs. predicted values; and (3) absence of severe multicollinearity among predictors, diagnosed using tolerance and variance inflation factor (VIF), with tolerance < 0.2 or VIF > 10 indicating potential multicollinearity (Table S2). To address multicollinearity, ridge regression was performed, and the optimal penalty parameter k was selected based on the ridge trace and the trade-off curve of k vs. R2 (Fig. S1c and d). Only after satisfying these criteria were the regression outputs deemed reliable.

The ARC-soil appeared visually compacted, whereas the compost-amended soils exhibited a more aggregated and porous structure (Fig. 1).

The detailed ANOVA results and Tukey HSD grouping letters are provided in Table 2. Prior to ANOVA, normality and homogeneity of variances were verified using the Shapiro–Wilk and Levene's tests, respectively (Table S4). Most parameters satisfied both assumptions (p>0.05); for the few cases with minor violations. To further ensure reliability, this study additionally performed the non-parametric Kruskal–Wallis H test for AK, EC, and AP, which confirmed the ANOVA results (Table S5). One-way ANOVA followed by Tukey's HSD post hoc test (p<0.05) was used to compare treatments. Compared to the unamended control, compost application significantly elevated soil TN content by 1.5 to 16 fold, and all compostamended soils showed significantly higher TN than the control (p<0.05). Both NH4+-N and NO3--N levels were also markedly higher in amended soils, with GLCS 10 % exhibiting the highest values among all treatments (p<0.05). Furthermore, increasing the compost addition rate from 2.5 % to 10 % consistently enhanced the concentrations of all three nitrogen forms (p<0.05). Notably, while the levels of NH4+-N and NO3--N in the original soil were similar (Fig. 2a), compost-treated soils exhibited consistently lower NH4+-N than NO3--N across all application rates, indicating that compost addition promoted the nitrification process in the ARC-soil system.

As shown in Fig. 2b, the ARC-soil had a negligible AP content, while all compost-amended soils exhibited significantly elevated AP levels. However, as the compost addition increased from 2.5 % to 10 %, both AP and AK contents increased significantly. According to Tukey's HSD test, SLS 10 % and SLS 2.5 % showed the highest AP levels with no significant difference between them, while SLS 10 % had the highest AK among all treatments (p<0.05). Notably, the high AP and AK contents in CDCS and SLS were due to additional external additions during composting, whereas FWCS and GLCS did not receive external additions of AP and AK. Figure 2c demonstrates that incubation with compost generally increased soil TOC relative to the nonincubated control. Tukey HSD comparisons revealed that FWCS 10 % and GLCS 10 % gave significantly higher TOC than the control and all 2.5 % treatments (p<0.05). Among the amendments, soils receiving FWC and GLC exhibited higher TOC levels than those treated with CDC or SL. It should be noted that lower TOC in a given amendment does not directly translate to reduced CO₂ emissions, as the stability of the carbon present also plays a determining role.

As shown in Fig. 2d, after adding 10 % compost, the pH values of FWCS and GLCS were slightly lower than that of the original ARC-soil, with values of 6.17 and 5.86 (the ARC-soil pH was 6.23), while the pH values of CDCS and SLS significantly increased. EC increased substantially with compost addition (Fig. 2e). At the 10 % application rate, FWCS, GLCS, and CDCS raised EC values by approximately 2 to 8 fold relative to the original soil. SLS produced the most pronounced effect: EC reached 18.02 and 7.07 dS m⁻¹ at 10 % and 2.5 % application rates, respectively, which were 13 to 34 times higher than the unamended control. This exceptionally high EC reflects the substantial soluble salt content of the sewage sludge compost, which originated from food and beverage wastewater treatment.

As shown in Fig. 2f, after adding 10 % compost, the moisture contents of FWCS, GLCS, CDCS, and SLS after 90 d were 1.12 %, 1.57 %, 3.03 %, and 2.63 %, respectively, all of which were higher than that of the original ARC-soil (0.5 %). Moreover, as the compost addition increased from 2.5 % to 10 %, the moisture content of FWCS, GLCS, CDCS, and SLS all increased to varying degrees, with CDCS 10 % being significantly higher than other treatments (p<0.05). During the incubation experiment, the irrigation time and water volume were kept consistent for all samples, indicating that CDCS and SLS exhibited higher water retention compared to FWCS and GLCS.

As shown in Fig. 2g, after adding 10 % compost, the CEC of the soil increased 3–5 times overall, with SLS 10 % and CDCS 10 % showing significantly higher CEC than FWCS 10 % and GLCS 10 % (p<0.05). A higher CEC indicates a greater capacity to retain nutrient cations, improved soil fertility, and a stronger buffering effect against fluctuations in soil pH and EC. Thus, CDCS and SLS conferred higher soil fertility than FWCS and GLCS, and more effectively regulated pH and EC fluctuations.

Throughout the 90 d incubation period, CO₂ emissions from FWCS, GLCS, CDCS, and SLS remained consistently higher than ARC-soil, although temporal fluctuations were observed (Fig. 3a). Over time, the CO₂ emissions from FWCS 10 %/FWCS 2.5 % and GLCS 10 %/GLCS 2.5 % significantly increased, peaking around day 52 (0.264/0.245 and 0.266/0.242 g kg⁻¹ d⁻¹, respectively). Thereafter, CO₂ emissions declined briefly from day 52 to 60 before stabilizing. At the 10 % application rate, CDCS and SLS also exhibited considerable fluctuation but remained higher than the unamended control (Fig. 3b). CDCS emissions gradually increased after two weeks, peaked around day 30, then declined over the following 20 d before stabilizing. In contrast, SLS emissions showed a declining trend after day 10 and stabilized after day 50. The CO₂ emission patterns for both CDCS and SLS were broadly consistent between the 2.5 % and 10 % application rates.

Figure 4 presents the cumulative CO₂ emissions from soils amended with four distinct compost types (FWC, GLC, CDC, SL) at two application rates for a total of nine treatments. The emission patterns varied considerably depending on the compost type. As illustrated in Fig. 4a, soils treated with GLCS consistently exhibited slightly higher cumulative emissions than those receiving FWCS. Figure 4b further demonstrates that CDC produced the highest cumulative emissions throughout the experiment, significantly surpassing the levels observed in soils treated with SL, GLC, and FWC. Overall, the compost types can be ranked in descending order of their effect on cumulative soil CO₂ emissions as follows: CDCS > SLS > GLCS > FWCS.

Figure 5a and b presents the correlation and significance matrices for twelve soil physico-chemical properties and CO₂ emissions. The analysis reveals that CO₂ efflux is closely linked to several factors, showing strong positive correlations with available phosphorus (0.79), pH (0.76), moisture content (0.74), available potassium (0.73), and electrical conductivity (0.62). Based on the correlation coefficients, the relative influence of these key drivers on CO₂ emissions decreased in the following order: AP > pH > moisture content > AK > EC.

Validating the absence of residual autocorrelation is critical for confirming that regression model assumptions hold. The Durbin-Watson statistic, which ranges from 0 to 4, reflects residual independence when near 2. Values deviating toward 0 or 4 imply possible autocorrelation. In this analysis, the obtained statistic of 2.597 (Table S1) closely approaches 2, indicating no substantial autocorrelation and thus satisfying the independence assumption.

The histogram of standardized residuals (Fig. S1a) displays a shape consistent with normality, centered near zero with a standard deviation of about 0.5. Correspondingly, the close alignment of points along the reference line in the first quadrant of the normal probability plot (Fig. S1b) also supports the normality of the residual distribution.

Collinearity diagnostics based on tolerance and variance inflation factor (VIF) are summarized in Table S2. Conventionally, tolerance < 0.2 or VIF > 10 indicates potential multicollinearity. In this dataset, total carbon (TC), moisture content, available phosphorus (AP), and available potassium (AK) all exhibited tolerance values below 0.2 and VIF values exceeding 10, confirming multicollinearity among these predictors.

To address this problem, the study used ridge regression analysis. As depicted in Fig. S1c, the trajectories of standardized coefficients across increasing values of the ridge parameter (k) show that each coefficient curve descends rapidly before stabilizing, and all curves converge to a common range. Based on the stabilization patterns in Fig. S1c and the corresponding MSE profile in Fig. S1d, a ridge parameter of k=0.05 was selected. The resulting ridge-regression coefficients are reported in Table 3. The established regression equation is: 2Y=5.815×AP-1.583×AK+0.726×pH-0.661×Moisture+0.123×EC+0.062×OM. In the regression equation, Y represents cumulative CO₂ emissions. The regression model's overall validity was confirmed by analysis of variance (ANOVA), which yielded a statistically significant result (p<0.1). The model demonstrated strong explanatory power, as evidenced by an adjusted R2 value of 0.902, indicating that approximately 90.2 % of the variance in the response variable is accounted for by the selected predictors. Within the fitted equation, the estimated coefficient for each independent variable quantifies its individual effect size and direction on the dependent variable.

The application of compost significantly enhances key soil physico-chemical properties. Firstly, the visual observations of soil structure (Fig. 1) indicate that compost amendment alleviated soil compaction in a dose-dependent manner. This effect is likely attributable to the humic substances in the compost, which promote the formation of soil aggregates through adhesion, thereby improving soil porosity and structural stability. Secondly, compost application significantly altered soil nitrogen dynamics, as reflected by the decreased NH4+–N and increased NO3-–N levels in all compost-treated soils (Fig. 2a), suggesting enhanced nitrification. This is consistent with the findings of Zhang et al. (2015), who reported that the addition of compost to lateritic soils increased the abundance of ammonia-oxidizing bacteria, thereby promoting nitrification. In this study, the CDCS treatment exhibited the strongest nitrification activity, likely due to its highest pH among the treatments. This aligns with the finding that low pH is detrimental to NH₃ retention in soil, as it promotes the ionization of NH₃ to NH4+, potentially creating a substrate shortage for nitrification and thus inhibiting the process (Ying et al., 2010). Furthermore, compost amendment markedly improves the bioavailability of phosphorus and potassium. Phosphorus in compost primarily exists in the form of mineral phosphates, such as calcium phosphate (Ca₃(PO₄)₂) and iron phosphate (FePO₄). Microbial activity during and after incorporation can solubilize these compounds, converting them into plant-available forms such as H₂PO4- and HPO42- (Lanno et al., 2021). Potassium in compost remains relatively stable and bioavailable as it exists mainly in ionic form (K⁺). During the composting process and subsequent soil incubation, these potassium ions are released into the soil solution, directly increasing the soil's readily AK content. The final levels of AP and AK contributed by compost depend on the initial nutrient content of the raw materials or any losses during the composting process.

The cumulative CO₂ emissions from the four compost-incubated soils followed a distinct descending order: CDCS > SLS > GLCS > FWCS. This pattern demonstrates that the specific type of compost significantly influences the magnitude of soil heterotrophic respiration. Notably, the CDCS treatment, characterized by the highest pH and substantial nutrient availability, also exhibited the greatest CO₂ efflux. This correlation suggests a strong link between compost-induced soil conditions and carbon mineralization rates.

Statistical analysis of influencing factors revealed that soil AP and pH were the primary drivers of CO₂ emissions in this compost-amended system, followed by moisture content and OM. The high adjusted R2 (0.902) demonstrates that the selected physico-chemical properties are strong predictors of cumulative CO₂ emission, collectively accounting for the vast majority of the observed variance. This confirms that compost-induced changes in these soil properties are closely linked to enhanced microbial activity. This hierarchy underscores the complex interplay between compost-derived nutrients, soil abiotic conditions, and microbial activity in controlling carbon dynamics.

The dominant role of pH stems from its comprehensive control over soil biogeochemistry. It is a master variable regulating microbial community composition and metabolic activity. For instance, bacterial growth rates are typically optimized in neutral to slightly alkaline conditions, and specific microbial taxa dominate at pH extremes (e.g., Actinobacteria in alkaline soils) (Rousk et al., 2009; Fierer and Jackson, 2006). Most extracellular enzymes involved in decomposition also exhibit peak activity between pH 6 and 8. Crucially, pH influences the stability of SOM. Lower pH increases the positive surface charge on iron and aluminium oxides, enhancing their capacity to adsorb negatively charged organic molecules and thereby promoting SOM stabilization (Wang et al., 2024b; Oliver et al., 2019). Conversely, a rise in pH, as seen in CDCS, can accelerate the decomposition of particulate organic matter (Hayashi et al., 2024), explaining the higher CO₂ emissions from this treatment.

Phosphorus availability emerged as another critical factor. In many ecosystems, especially tropical ones, phosphorus is a key limiting nutrient for biological activity (Yang et al., 2014). The addition of compost, which increased soil AP, likely alleviated this limitation. As suggested by Sanchez-Rodríguez et al. (2024), phosphorus addition can indirectly stimulate CO₂ emissions by promoting the desorption of organic carbon compounds from mineral surfaces, thereby increasing substrate availability for microbes. This mechanism may explain the positive correlation between AP and CO₂ flux observed here. Soil moisture content, a secondary yet significant driver, primarily regulates the diffusion of oxygen within soil pores. This, in turn, governs the dominant mode of microbial respiration (aerobic vs. anaerobic) and overall microbial activity, making it a fundamental regulator for the soil biota responsible for decomposition (Lacroix et al., 2021).

In summary, our results supported all three hypotheses proposed in the Introduction. Hypothesis 1 was confirmed by the significant differences in cumulative CO₂ emissions among compost types, with CDCS > SLS > GLCS > FWCS. Hypothesis 2 was validated by the multiple regression analysis, which identified soil pH and available phosphorus as the primary interactive drivers of emissions. Hypothesis 3 was substantiated by the fact that CDCS induced the highest pH elevation and AP increase, leading to the greatest CO₂ flux, whereas FWCS showed the opposite effect.