The study was conducted at five strategically selected locations representing major rice-growing agro-ecological zones across India: Khudwani, Jammu & Kashmir (33.7237° N, 75.0916° E; northern zone), Karaikal, Puducherry (10.9504° N, 79.7797° E; southern zone), Moncompu, Kerala (9.4377° N, 76.4284° E; southern zone), Pant Nagar, Uttarakhand (25.9845° N, 85.6742° E; northern zone), and Pusa, Bihar (25.9845° N, 85.6742° E; eastern zone). This selection encompassed key rice cultivation regions in the northern, eastern, and southern parts of the country. This spatial diversity across these sites provided a robust framework for evaluating treatment effects under diverse climatic and edaphic conditions. Prior to trial initiation, composite surface soil samples (0–15 cm depth) were collected from each experimental site and analyzed for key initial physicochemical properties (Table 1). Samples were collected at two key stages: before and after the experiment. For each composite sample, five subsamples were collected from a given plot. A five-point zigzag pattern was used to minimize the impact of spatial variability.

The climate of the study sites ranges from temperate (Khudwani), sub-tropical (Moncompu, Karaikal, and Pant Nagar), and tropical (Pusa). Mean maximum temperature, minimum temperature, and cumulative rainfall during the crop growth period are depicted in Figure 1.

Field experiments were conducted during two consecutive wet seasons (2023 and 2024) across the five study locations. The experimental design was a randomized block design with six treatments. Treatments comprised varying fertiliser levels (50%, 75%, and 100% RDF) combined with crop residue incorporation (equivalent to 25% and 50% of the recommended dose of fertilizers) and/or green manure (GM)/green leaf manure (GLM). The specific treatments were T1 - 100% RDF via fertilizers, T2 - 50% RDF via fertilisers + 50% RDF via residue, T3 - 50% RDF + 50% RDF via residue + PD, T4 - 50% RDF via residue + 50% RDF via GM/GLM, T5- 75% RDF via fertilisers + 25% RDF via residue + PD, and T6 - absolute control. Detailed descriptions of all treatments and the initial characteristics of each location are provided in Tables 1 and 2.

Rice seedlings (25–30 days old) were transplanted at a spacing of 20 × 15 cm, with 2–3 seedlings per hill, across all study locations. The most common rice cultivar in each study location is utilized for experimental purposes. For details about the rice cultivar used in the study, refer to Table 1. The RDF varied by study location, adhering to state-specific recommendations. Moncompu received the lowest RDF, while Karaikal received the highest. Detailed RDF values for each location, along with their application intervals, are presented in Table 1. N, P, and K were supplied as urea, single super phosphate (SSP), and muriate of potash (MOP), respectively. Crop residues were incorporated into the soil at N-equivalent rates specific to each treatment.

Wider bunds and a water channel as a buffer zone measuring 1 m in length were used to separate the individual plots. This prevented the treatments from mixing and avoided cross-contamination or nutrient leaching between plots. Aerobic conditions were maintained in the field during the mid-tillering and panicle initiation stages to optimize N fertilizer application. All other agronomic management practices, including weed management, irrigation, and insect pest and disease control, followed the standard state recommendations for each respective location. Irrigation was applied manually to each experimental plot, with great care taken to prevent over-irrigation, nutrient leaching, and water movement between plots. The bunds separating the individual treatments were periodically plastered with mud to ensure an effective barrier against water flow between treatments.

The measured quantity of rice straw for each treatment was uniformly spread across the designated plots. Subsequently, the straw was incorporated into the soil via human foot treading after the final land preparation and three days before the transplanting of rice seedlings (Figure 2). The same procedure was followed for GM and GLM. Glyricidia sepium is used as GLM in Karaikal, Moncompu, and Pantnagar, while Trifolium is used in Khudwani and Sesbania aculeata (Daincha) in Pusa. The total quantity of crop residue incorporated in each treatment, along with its N content, was provided in Tables 1 and 2. To mitigate the risk of N immobilization associated with microbial decomposition of high-carbon rice residue, 50% of the recommended N dose was applied basally just before transplanting. It is assumed that this strategy will ensure adequate early-season N availability for both crop establishment and microbial activity.

The PD developed by the Division of Microbiology, ICAR - Indian Agricultural Research Institute (ICAR-IARI), New Delhi, is a fungal consortium that accelerates straw decomposition through the enzymatic breakdown of cellulose, lignin, and pectin (26). The PD solution was prepared as follows: 150 g of jaggery was dissolved in 5 L of boiled water and filtered through a sieve to remove impurities. After cooling, 50 g of chickpea flour was added to the solution. Four PD capsules were inoculated into the mixture. The solution was covered with a muslin cloth and incubated at ambient temperature (25–30 °C) for 10 days (Figure 3). The resulting culture was diluted with 100 L of water to prepare the final inoculum. The diluted inoculum was uniformly sprayed onto straw residue at a rate of 80 L per tonne using a manual knapsack sprayer with a spray nozzle (delivery point) maintained approximately 50 cm above the ground to prevent spray drift (Figure 2). Residues were incorporated after three days of application.

Yield attributes (tiller/m² and panicle number/m²) were recorded from the specified area (1 m²) of each plot (30 m²). Representative fully developed plants were selected from each treatment for recording filled grain per panicle manually. Grain and straw yields were estimated in net plot areas, and plants were cut at 15 cm above the ground level. The crop was harvested at physiological maturity, determined by a grain moisture content reaching approximately 14–16%. Harvested plants were subsequently sun-dried for 3–4 days before threshing. After threshing, grain yield was weighed separately (~ 14% moisture content) and expressed in kg ha^-1. The straw samples were dried in a hot-air oven at 70 °C for 72 hrs, and subsequently, straw yield was measured and expressed in kg ha^-1. Also, the sample for thousand-grain weight is collected from each plot, weighed using a grain counter (A-Square, Ambala Associates, Haryana, India), and expressed in grams.

Composite soil samples were collected immediately after harvest from the experimental sites at a depth of 0–15 cm for the analysis of the soil physicochemical properties. Collected samples were air dried at room temperature (25–28 °C), gently crushed, and sieved (< 2 mm) to remove gravel and plant debris. The processed samples were analyzed for key physicochemical parameters like pH and EC (1:2.5 ratio) using pH and EC meter; SOC using 1 N K₂Cr₂O₇ (27), available N using Kjeldhal method (28), available P using Bray-1 method (29) at Momcompu and blue colour method (30) at other locations, and available K using ammonium acetate method (25).

Harvested grain and straw samples were dried and ground using a high-speed blender. N content of straw and grain was determined through the macro-Kjeldahl method. Precisely, 0.5 g of finely ground grain and straw samples were digested separately with 10 mL of concentrated H₂SO_4. and a catalyst mixture in Kjeldahl digestion tubes until clear. The digest was then subjected to steam distillation using the Kelplus instrument, where liberated ammonia was captured in boric acid. Finally, the collected ammonia was quantified by titration with a standard acid, with a reagent blank included for accuracy. P and K in the grain and straw samples were analysed after wet digestion using a di-acid mixture (HNO₃: HClO₄ in the ratio of 4:1). For P, the digested aliquot was reacted with a vanadomolybdate solution. The absorbance of the resulting yellow phospho-vanadomolybdate complex was then measured at 470 nm using a visible spectrophotometer. K content in the digested extracts was determined using a flame photometer.

The partial factor productivity of N (PFP_N) was calculated using the formula below and expressed as kg grain/kg N applied. While calculating the N input, both fertilizer N and organic N via crop residue and GM/GLM were summed together in each treatment.

The contents of N, P, and K in straw and grain remained generally consistent across treatments and locations, except Karaikal (Tables 4, 5). Treatment T1 (100% RDF via fertilizers) and T5 (75% RDF via fertilizers + 25% RDF via residue) resulted in statistically similar straw N content across sites, excluding Karaikal, where the variation was non-significant in both 2023 and 2024. At Khudwani and Pantnagar, substituting up to 50% of N with crop residues (T2 and T3) yielded comparable N assimilation (0.54 ± 0.02 and 0.55 ± 0.01, respectively). However, in Pusa, Moncompu, and Karaikal, N supplied solely through residue and GLM failed to significantly enhance N content compared to the absolute control.

Grain P content showed inconsistent responses across years and locations, especially at Pusa, Pantnagar, and Karaikal. In Karaikal, no significant improvement in grain P was observed in 2023, and an unusual significant trend appeared in 2024, where the no-fertilizer (T6: 0.65%) and organic-only (T4: 0.59%) treatments recorded surprisingly higher values. At Pusa and Pantnagar, treatments with 100% RDF (T1) and PD combined with 25% rice residue and 75% urea N (T5) consistently enhanced grain P in both years. Pooled data highlighted the superiority of T1 (0.22% and 0.44%) and T5 (0.21% and 0.46%) at Khudwani and Moncompu, respectively. While residue-based substitutions (with or without decomposer) performed comparably to T1 and T5 at Khudwani, they led to poor P accumulation at Moncompu. In the case of straw P, treatments T1 and T5 recorded the significantly highest straw P content at Karaikal during 2023. Conversely, organic residue treatments (T2, T3, T4), aided by PD, maintained higher straw P content over both years by promoting labile soil P. Overall, pooled results indicated T1 to be significantly superior at Pusa (0.11%) and Pantnagar (0.13%), whereas T5 excelled at Moncompu (0.42%) and Khudwani (0.13%). Higher substitution levels (T2, T3, T4) with residues and GM/GLM reduced P availability, resulting in suboptimal P content.

For K, both straw K and grain K concentrations were significantly higher in T1. In Pusa and Kudwani, the concentrations in T1 were on par with T5, and in Karaikal, they were on par with T4. These treatments were followed by other residue-substituted options, which showed notable improvement in K content except in the silty clay soils of Pantnagar. Remarkably, the integration of rice straw residue with GLM (Glyricidia) achieved grain K levels comparable to those under RDF, particularly in the sandy loam soils of Karaikal, pressing the potential of organic amendments in enhancing K uptake under such soil conditions.

Our bivariate analysis clearly demonstrates a direct relationship between nutrient concentrations in the plant (both grain and straw) and grain yield across various locations.

The PFP of N varied significantly across treatments and locations (Figure 8). Treatment T1 (100% RDF) consistently yielded the highest PFP of N at all locations. However, in Khudwani, Moncompu, and Pusa, T1 was statistically comparable to T5. In contrast, at Pant Nagar, T5 recorded a significantly lower PFP of N than T1. No clear trend was observed in Karaikal, where differences were non-significant. The higher PFP of N under T1 and T5 suggests the immediate availability of N from the urea fertilizer source. Pooled analysis across all locations reflected a similar trend: T1 achieved the highest PFP of N (47 kg grain/kg N), followed closely by T5 (44 kg grain/kg N). Notably, treatment T3 (42.5 kg grain/kg N) was statistically equivalent to T5. Conversely, T4 recorded the lowest PFP of N (39 kg grain/kg N), underscoring the necessity for at least some mineral N input. Overall, the hierarchy of PFP of N was T1 > T5 ≈ T3 > T2 ≈ T4. This indicates potential for productivity gains by increasing N fertilization in T2 and T4, as substituting 50% RDF with crop residue resulted in slightly lower PFP compared to full fertilizer N supply (T1).

The effect size analysis of the different locations was explained using Eta² (a measure of effect size indicating the proportion of total variation in a dependent variable explained by a treatment or factor, η²), Omega² (a less biased estimate of the true treatment effect size compared to Eta², ω²), and Minimum Effect Size of Detectable Difference (MESD), which is the magnitude and consistency of treatment responses across five experimental locations over two years. The effect analysis identified Pantnagar as the most responsive site, as it exhibited the highest Eta² values for key parameters. The high Eta² values for grain yield (1.00, 0.99), straw yield (1.00, 0.99), panicle number (0.90, 0.86), and 1000-grain weight (0.94, 0.84) in 2023 and 2024, respectively, suggest that treatment effects were strong and residual variation was minimal (Table 8). In contrast, lower MESD values were observed, except for grain (51.31, 85.97) and straw yield (75.0, 112.31), which indicates that smaller treatment differences were detected at this location.

At the Pusa center, a moderate to high treatment responsiveness was observed for most parameters. Grain and straw yield exhibited notable η² values of 0.78 and 0.76 in 2023, which increased to 0.79 and 0.79 in 2024, respectively (Table 8). These values indicate that a significant proportion of the parameter variation was attributable to the treatment effects. Conversely, consistently poor η² and ω² values for panicle number, tiller count, and grain P suggest a more stable effect from the treatments. Moderately high MESD values for grain yield (427.49, 473.79) and straw yield (533.53, 623.47) indicate a high level of effect detection, whereas lower MESD values for other nutrient traits point to greater detection sensitivity.

At the Khudwani location, the effect size analysis revealed moderate and trait-specific treatment responsiveness over the two years (Table 8). In 2023, strong treatment-induced variation was observed for grain N (η² = 0.79; ω² = 0.72) and grain yield (η² = 0.65; ω² = 0.55). In contrast, lower η² and ω² values for grain P and panicle number suggested weaker treatment effects. The effect size for grain yield increased in 2024 (η² = 0.75; ω² = 0.67), indicating a moderate treatment influence. High MESD values for yield and 1000-grain weight, however, suggest that greater error variability was associated with these traits. The low effect size at Karaikal indicates a weak and variable treatment response, classifying it as a low-responsive site. For instance, the low η² and ω² values for grain yield (0.33, 0.14 in 2023 and 0.21, 0.00 in 2024) point to minimal treatment variance and high residual error (Table 8). Poor treatment effects were similarly observed for grain N, straw N, and organic carbon, which also showed low ω² values. The high MESD values for straw yield (1162.18, 997.73) and grain yield (370.8, 891.4) in 2023 and 2024, respectively, suggest that only large treatment differences could be detected. An exception was the modest effect of P treatment on grain P in 2024 (η² = 0.48; ω² = 0.33; MESD = 0.15).

The Moncompu site exhibited moderate to high treatment responsiveness, as evidenced by the high η² values for grain yield, straw yield, panicle number, and tiller count, though these values slightly decreased from 2023 to 2024 (Table 8). Panicle number (η² = 0.91; ω² = 0.87 in 2023) and tiller count (η² = 0.86) showed particularly strong responsiveness, which suggests a clear and reproducible treatment effect. While MESD values for grain and straw yield were moderately high (e.g., grain yield = 422.39, 727.59; straw yield = 585.13, 865.08 in 2023 and 2024, respectively), indicating lower measurement precision, lower MESD values for other parameters point to a greater sensitivity for detecting treatment effects. Overall, this effect size analysis highlights that the strength, consistency, and performance of treatment effects are highly location and trait-specific. Among these experimental sites, Pantnagar expressed high experimental sensitivity, can be considered an optimal site for evaluating agronomic interventions. On the other hand, Karaikal require refined evaluation practices such as increased replication to enhance treatment effect.

This study indicated that a reduction in the recommended dose of fertiliser application, along with straw incorporation, resulted in statistically comparable improvement in tiller number and panicle number with the full recommended dose of fertiliser. This superior response in residue replacement treatment is due to better nutrient availability by residue decomposition, better synchronization of nutrient release and rice plants demand, and reduced tiller mortality. Moreover, this higher tiller and panicle number could be attributed to several factors, including improved soil structure due to residue retention, enhanced nutrient cycling, and increased microbial activity, and the balanced application of synthetic fertilisers, organic materials and decomposers (40, 41). Also, it could be attributed to the high N supply of the native soil of the respective locations. Reduced yield attributes observed in the control (T6) and less RDF plots involving 50% RDF and residues may be attributed to the limited nutrient supply capacity of these treatments. The 50% residue + 50% GM/GLM treatment led to a significant decline across all measured parameters compared to 100% RDF and 50% residue + 50% RDF treatments, except in Pant Nagar and Karaikal during 2023. These results highlight the need for balanced fertilisation to ensure robust early crop development. This finding is aligned with Lamba and Gill (21), who reported that the combined use of recommended fertilizer, residue, and microbial decomposer improves the nutrient availability and yield attributes under irrigated rice system.

The 100% RDF treatment demonstrated superior grain and straw yield, reflecting an optimal balance of nutrition and synchronization with plant N demand, which significantly enhanced plant growth and dry matter accumulation. In contrast, partial residue incorporation (25%) combined with PD (T5) produced yields comparable to 100% RDF, attributed to improved nutrient availability and accelerated residue decomposition. The fungal consortia present in PD expedite the enzymatic breakdown of crop residues, leading to rapid nutrient release and enhanced uptake, thereby fostering plant development and productivity. This finding aligns with Singh et al. (42), who reported that the integration of PD with RDF accelerates rice straw decomposition and enhances both yield and yield parameters. Additionally, the observed increase in total dry matter accumulation could be attributed to multiple contributing factors, including improved soil physio-chemical properties, microbial activity, organic matter content, root development, enhanced K recycling, and faster N assimilation (mineralization).

Specifically, co-treatments incorporating 75% RDF, rice straw, GM, and decomposer synergistically improved soil N, K, and OC availability, further driving rice productivity. The mechanisms by which straw incorporation, in conjunction with N fertilization, enhances rice yield could be attributed to following factors, i.e. straw and N fertilization altered the soil’s physical structure, facilitated the leaching of salts from the soil, and less loss of NH4⁺-N and NO₃⁻-N (43). Moreover, the combined application of organic and inorganic inputs effectively enhances soil fertility, boosts grain yield, and reduces dependency on synthetic fertilizers (44, 45). However, 50% N substitution through residue and GLM (with or without decomposer) failed to yield consistent results across locations due to spatial variability in nutrient release and soil nutrient reserves. This reduction is primarily associated with the high carbon-to-nitrogen (C: N) ratio of rice straw, which slows decomposition and induces N immobilization, limiting its availability for rice uptake (46, 47).

T1 treatment achieved the highest average grain yield (5395 kg ha^-1) among all treatments. However, it exhibited low stability across the study locations, as indicated by its relatively high ASV (30). This suggests that while it can produce excellent yields under optimal conditions, its performance might fluctuate significantly in varying environments. It is best suited for areas where the goal is to maximize yield, potentially with controlled conditions or in specific environments where its performance is consistently high (Table 9, Figure 10). T2 treatment yielded a moderate average grain yield (4767 kg ha^-1) among all treatments. Crucially, it demonstrated the best stability across the study locations (lowest ASV = 10). This indicates that T2 is very reliable and consistent in its performance across different study locations, even under variable environmental conditions. While its yield isn’t the highest, its reliability makes it a strong candidate for farming systems prioritizing consistent returns.

T3 achieved a good average grain yield (4927 kg ha^-1) among all treatments and was found to be very stable, with an ASV (11) close to the best-performing T2. This excellent combination of good yield and high stability makes this treatment particularly suitable for broad adaptation. It suggests that T3 can perform well and consistently across a wide range of environments, offering a robust solution for diverse farming conditions. T4 treatment resulted in a comparatively low average grain yield (4615 kg ha^-1) among the improved treatments. It showed moderate stability, falling somewhere in the middle of the stability range (ASV = 21). While it incorporates organic components, its overall performance in terms of both yield and stability is not as competitive as T1, T3, or T5.

T5 produced a high average grain yield (5360 kg ha^-1), very close to the highest yield achieved by T1. It showed slightly better stability than T1 (lower ASV), suggesting a marginal improvement in consistency. However, similar to T1, its high yield may still be environment-specific, meaning its superior performance might not be consistent across all possible environmental variations. As expected, T6 (the absolute control with no external inputs) resulted in the poorest average grain yield (3462 kg ha^-1) and was the least stable (highest ASV = 38) among all treatments. This confirms its unsuitability for optimal crop production and highlights the necessity of proper nutrient management and soil improvement strategies. It is not recommended for practical application.

Diagnostic checks were conducted to support the validity of the AMMI model. Residual normality was assessed using the Shapiro-Wilk test (W = 0.9577, p = 0.2705), which indicated no significant deviation from a normal distribution. This was further confirmed by a Q-Q plot (Supplementary Figure S1). Homogeneity of variance was tested using Levene’s test (F = 1.2897, p = 0.3008), suggesting that the assumption of equal variances across environments holds. The residual vs. fitted values plot showed no discernible pattern (Supplementary Figure S2), which supports the assumption of homoscedasticity. A histogram of the residuals also demonstrated a symmetric distribution centered around zero (Supplementary Figure S3), further supporting the normality assumption. Collectively, these diagnostics validate the core assumptions of the AMMI model for our multi-environment trial data. The multiplicative structure of the G×E interactions was supported by the first two significant interaction principal components (IPC1 and IPC2), which together explained over 95% of the interaction variance. IPC1, accounting for over 60% of the variation, was highly significant (p = 0.0071), while IPC2 also contributed meaningfully (p = 0.0635). These results confirmed that the AMMI model was appropriate and interpretable for capturing G × E patterns (Supplementary Figure S4).

This study demonstrates that straw nutrient (NPK) concentrations were generally consistent across treatments and locations, with the exception of Karaikal, where high spatial variability in soil characteristics influenced nutrient dynamics. Treatments involving 100% RDF and 75% RDF with partial residue incorporation maintained comparable straw N levels across most sites, underscoring the efficacy of integrated nutrient management. Notably, T2 (50% substitution with residue and PD) resulted in substantial N assimilation at Khudwani (0.54 ± 0.02%) and Pantnagar (0.55 ± 0.01%), illustrating the potential of organic N recycling when optimized with microbial inputs. This outcome aligns with findings by Wang et al. (48), who reported that the co-application of straw and decomposers enhances N availability in soils with moderate fertility. However, at locations such as Pusa and Karaikal, N assimilation was comparatively lower, likely due to site-specific constraints such as suboptimal mineralization, poor microbial activity, or lower soil N reserves.

Rice straw, with its modest N content (~0.7%) and inherently high C:N ratio, undergoes slow decomposition, leading to N immobilization and delayed release (49). Sole reliance on residue or GM/GLM proved insufficient nutrient supply and synchronize to match the performance of inorganic or integrated treatments, reinforcing that organic sources alone often lack the rapid nutrient availability needed for high-yielding rice systems. The application of microbial decomposers in conjunction with inorganic N has been shown to significantly accelerate straw degradation, enhance microbial-mediated nutrient turnover, and improve N uptake efficiency (50). This synergy is vital for meeting crop N demands, especially during critical growth stages.

P dynamics revealed unique trends in the studied sites. At Karaikal, organic-only treatments outperformed conventional fertilizer treatments, likely due to localized variations in soil P and microbial activity. Research suggests that organic amendments, including microbial decomposers, can increase the bioavailability of labile P through mineralization and acidification processes (51). Despite this, the superiority of T1 and T5 across most locations highlights the critical need to maintain adequate inorganic P inputs under residue-based nutrient management. Organic inputs, though beneficial for building soil P pools, often fall short in delivering sufficient P to meet immediate plant demands, especially under high-residue scenarios with elevated P sorption or low mineralization rates. Interestingly, higher P concentrations in the control (T6) and organic-only treatments (T4) at Karaikal and Moncompu may be linked to P from past fertilization or enhanced microbial solubilization under low-input systems.

K concentrations in both straw and grain were significantly influenced by treatments T1 and T5, affirming the benefit of combining full or partial inorganic K with organic amendments. Residue-based treatments also elevated K levels across sites, except in the clay-heavy soils of Pantnagar, where K fixation likely limited its plant availability. This aligns with findings by Yan et al. (52), who reported that soil texture critically affects K dynamics and exchangeability. In particular, Karaikal’s sandy loam soils responded well to straw plus GM combinations, achieving grain K levels comparable to full RDF. This can be attributed to the high intrinsic K content of rice straw and the capacity of organic residues to recycle up to 80% of the crop’s K demand when adequately decomposed with microbial inoculants (43). Moreover, in coarse-textured soils, organic inputs enhance K retention by improving cation exchange capacity and microbial activity, which are essential for sustaining crop productivity in coastal and salt-affected environments.

The combined application of synthetic N fertiliser and rice straw significantly influenced soil N and OC dynamics across treatments and locations. As primary indicators of soil fertility, both soil N and OC play pivotal roles in sustaining rice productivity. Treatments integrating synthetic fertilizer with rice straw (T5) demonstrated improved soil N availability, likely due to synergistic effects that promote a reduced C:N ratio and enhanced nutrient mineralization. The inclusion of PD further accelerated straw breakdown, facilitating quicker nutrient release and promoting microbial activity. In contrast, treatments relying solely on organic inputs such as crop residues and GM/GLM (T3 and T4) maintained a relatively higher C:N ratio, which slowed decomposition and resulted in reduced N mineralization.

Nonetheless, these treatments contributed to an increase in soil OC across locations, reflecting the long-term benefits of organic matter accumulation despite lower immediate nutrient availability. Significant differences in soil OC and N content suggest that continuous application of combined synthetic and organic inputs fosters gradual enrichment of soil organic matter and N pools. This aligns with findings by Chen et al. (53), who highlighted that residue retention enhances carbon transformation pathways, forming stable carbon intermediates that contribute to OC buildup. Similarly, Lal (54) emphasized that integrated nutrient strategies improve nutrient-use efficiency and support carbon sequestration, critical for maintaining soil health under intensive rice systems. Furthermore, studies by Bhattacharyya et al. (55) and Kumar et al. (56) corroborate that balanced N input with residue incorporation minimizes nutrient losses, stimulates microbial turnover, and enhances soil structure and fertility.

Our two-year post-experiment assessment revealed complex, site-specific changes in soil OC, N, P, and K, driven by a combination of input types, climate, microbial activity, and inherent soil characteristics.

The posterior probabilities for nutrient replacement through crop residue, GM and PD showed improvements in soil nutrient parameters (K, P, N, and OC) compared to the 100% RDF (T1) across five locations (Figure 11). N exhibited moderate improvement across locations, with enhanced response under integrated treatments. T2 (50% substitution through residue) showed lower efficacy at Karaikal and Khudwani, while Moncompu reported better N dynamics under T2 and T3. Pantnagar had largely negative N responses, except moderate improvement under T5 (partial residue + decomposer), which consistently performed better across locations.

P showed high spatial variability and was the most limiting nutrient across the locations due to poor nutrient mineralization from residues and GLM. Positive responses were seen at Karaikal, especially under T2, while Khudwani and Moncompu showed minimal improvements. Only T5 showed any P enhancement at Pantnagar, supporting the importance of integrated inputs and decomposer for efficient mineralization and P availability. K was the most responsive nutrient, especially at Moncompu under T2 and T3 due to efficient K release in coastal soils. Moderate K gains were noted at Karaikal and Khudwani under organic treatments, but Pantnagar showed poor response, likely due to leaching or fixation. Overall, residue-based treatments (T2 and T5) contributed positively to K improvement in Moncompu and Karaikal due to richer K content of straw.

As expected, OC showed the most consistent improvement across sites due to organic inputs. Karaikal and Moncompu recorded high OC probabilities, particularly under T4 and T5. Khudwani showed strong OC buildup under residue and GM (T2 and T4), while Pantnagar had limited gains, with T5 showing marginal improvement. Treatments involving residue and decomposer (T3, T4, and T5) demonstrated a strong role in enhancing soil OC, reinforcing the value of organic amendments in sustainable nutrient management.

Irrespective of treatment, yield attributes (tillers, panicles, grain weight, 1000-grain weight) showed strong dependence on soil N and K availability. Similarly, yield attributes showed a strong positive correlation with grain yield across all treatments (Figure 12). On the contrary, grain K concentration and soil EC showed a negative correlation with grain yield across all treatments. In T1 (100% RDF), grain yield exhibited strong positive correlations with key yield parameters: tiller number (r = 0.83), panicle number (r = 0.76), thousand-grain weight (r = 0.56), and straw yield (r = 0.68). Soil N (r = 0.60), K (r = 0.50), and OC (r = 0.43) also correlated positively with grain yield. These results demonstrate that the supply of 100% nutrient requirement of the crop through synthetic fertilization ensures optimal nutrient supply, driving yield component development and final productivity.

In T2 (50% RDF + 50% Residue), grain yield maintained robust correlations with tiller number (r = 0.83), panicle number (r = 0.79), soil N (r = 0.73), and soil K (r = 0.64), though linkage to thousand-grain weight weakened (r = 0.44). This confirms that substituting 50% synthetic N with residue sustains yield-critical relationships. Residue decomposition effectively releases nutrients, supporting growth dynamics and grain yield, indicating synergistic integration of chemical and organic N sources. In T3 (T2 + PD), the microbial inoculant amplified residue mineralization, strengthening grain yield correlations with tiller number (r = 0.87) and panicle number (r = 0.84). Soil N exhibited enhanced efficiency, showing exceptionally high correlations with straw yield (r = 0.88), thousand-grain weight (r = 0.87), and OC (r = 0.58). Soil K remained strongly linked to grain yield (r = 0.74) and N availability, confirming that accelerated residue breakdown improves K mobilization alongside N.

T4 (50% Residue + 50% GM/GLM) drove the strongest positive correlation between grain yield and yield-attribute: tiller number (r = 0.84), panicle number (r = 0.84), and straw yield (r = 0.81). Soil N remained pivotal for grain yield (r = 0.73), while K demonstrated integrated cycling with N (soil K-N: r = 0.58). This highlights GM/GLM as a functionally equivalent alternative to rice residue for N supplementation, enabling flexible organic sourcing strategies. Overall, the 50% N substitution through residue (T2, T3, and T4) outperformed 100% synthetic fertilization (T1) in nutrient-yield coupling, notably enhancing N-K synergies. This aligns with the synchrony hypothesis: Crop residue mineralization better matches rice growth stages than split-applied synthetic fertilizers. T3’s PD further amplified this effect, boosting N mineralization, confirming accelerated enzymatic breakdown of residue. T4’s GM/GLM-residue blend achieved the highest yield-attribute correlations, demonstrating functional complementarity such as slow-release N, K, reduced leaching losses, and physical soil structuring.

This study utilized PLSR to identify key factors influencing rice grain yield (GY) across six treatments (T1-T6) (Figure 13). The PLSR models consistently explained a substantial proportion of the variability in both predictor variables and GY, with the first two components (PLS1 and PLS2) cumulatively accounting for 63.8% to 70.0% of predictor variance and 82.9% to 90.8% of GY variability across all treatments, demonstrating their efficacy in capturing complex relationships. T1 (100% RDF) captured 64.9% of predictor variability and 82.9% of GY; T2 (50% RDF via fertilizer + 50% RDF via residue) explained 69.4% of predictor variability and 85.0% of GY; T3 (50% RDF via urea + 50% RDF via residue + PD) accounted for 70.0% of predictor variability and 85.8% of GY; T4 (50% RDF via Residue + 50% RDF via GM/GLM) showed 67.5% of predictor variability and 86.2% of GY; T5 (75% RDF via fertilizer + 25% RDF via Residue + PD) captured 63.8% of predictor variability and 88.8% of GY; and T6 (Absolute Control) explained 64.7% of predictor variability and 90.8% of GY.

Across these treatments, several consistent patterns emerged regarding the drivers of GY. Grain nitrogen (Grain_N) and straw yield (SY) consistently appeared as dominant positive drivers, highlighting their crucial role in productivity across varying input levels. Similarly, key yield attributes such as tillers per m², panicles per m², and thousand-grain weight consistently exhibited strong positive associations with GY, emphasizing their fundamental importance in yield formation. Soil N and soil K were also frequently identified as significant positive contributors. Conversely, grain phosphorus (Grain_P), soil OC, soil P, and soil EC generally showed minimal direct influence or even negative associations with GY, while grain K often exhibited an inverse relationship. These findings collectively underscore the primary role of nitrogen availability (both in grain and soil), biomass production (SY), and core plant yield components in determining rice grain yield, with other soil factors playing a less direct or consistent role.

Implementing large-scale rice straw residue recycling requires a combination of strong institutional support and effective extension services. Governments must lead the way by creating favorable policies and offering financial incentives, such as subsidies for essential machinery like balers. They also need to establish a stable market with fair prices for the baled straw. A robust regulatory framework is essential to enforce anti-burning laws and encourage public-private partnerships that build efficient supply chains. While some states in India have already implemented strict regulations, these policies need to be uniformly applied across the entire country. Extension services are critical for closing the gap between policy and on-the-ground practice. This includes extensive farmer training and awareness campaigns, which can be delivered through demonstration plots and on-farm workshops. These platforms allow farmers to see the environmental and economic benefits of recycling paddy straw. It’s also vital to provide technical guidance through well-trained extension agents and to promote farmer producer organizations and custom hiring centers. These initiatives empower farmers with the skills and resources to adopt sustainable practices, helping them transition away from burning residue and toward a more circular agricultural economy. To further accelerate this shift, governments should prioritize the year-round availability of low-cost microbial solutions, such as PD, across the country. This requires investing in infrastructure for the mass production of these microbial consortia while ensuring quality and price control. Finally, to promote wider adoption, on-farm and on-station trial videos demonstrating the entire process from sowing to harvest should be recorded and broadcast in all vernacular languages through television and other mass media.

While this study provides valuable insights into the efficacy of rice straw recycling as a nutrient supplement, it is important to acknowledge certain limitations that may affect the generalizability of our findings. The experiment was conducted under specific agro-climatic conditions and soil type, which may not be fully representative of the diverse rice-growing regions across India. The effectiveness of the microbial inoculant and the decomposition rate of the rice straw can be highly sensitive to environmental factors such as temperature, moisture, and soil microbial communities, which vary significantly across different locations. Additionally, the composition and quality of rice residue itself can differ based on the rice variety, harvesting practices, and local conditions, which may influence the release of nutrients. Further investigation into the economic feasibility of this practice at the farm level is also warranted.