The purpose of this study was to evaluate the effects of different nitrogen (N), phosphorus (P), and potassium (K) fertilization ratios on the carbon (C), N, and P contents and their ecological stoichiometric characteristics in the leaf–soil–microbial system of Sapindus saponaria and elucidate their relationship with yield. A “3414” experimental design was employed in a 6-year-old Sapindus saponaria woodland located in Fujian Province of China. Fourteen N–P–K fertilization treatments with three replicates were established. Leaf, soil, and microbial samples were collected and analyzed for C, N, and P contents. Redundancy Analysis (RDA), Partial Least Squares Path Modeling (PLS–PM), and the entropy-weighted technique of ranking preferences by similarity to optimal solutions (TOPSIS) were utilized to assess the relationships among variables and determine optimal fertilization strategies. It was found through research that different fertilization treatment methods have a significant impact on both the soil nutrient content and the C, N, and P contents of soil microorganisms. Compared with the control group, soil organic C, total N, and total P, and microbial C, N, and P contents increased by 14.25% to 52.61%, 3.90% to 39.84%, 9.52% to 150%, 6.65% to 47.45%, 11.84% to 46.50%, and 14.91% to 201.98%, respectively. Results from Redundancy Analysis (RDA) indicated that soil organic C, total N, and total P exerted a significant influence on the leaf nutrients. PLS-PM demonstrated that fertilization indirectly affected leaf nutrient accumulation and yield by altering soil properties, with soil total phosphorus and leaf phosphorus being key determinants of yield. Additionally, soil microbial entropy impacted yield by regulating microbial biomass stoichiometric ratios. The entropy-weighted TOPSIS model identified the N₂P₂K₂ treatment (600 kg/ha N, 500 kg/ha P, and 400 kg/ha K) as the most effective fertilization strategy. Optimizing N–P–K fertilization ratios significantly enhances leaf nutrient content and soil microbial biomass C, N, and P, thereby increasing Sapindus saponaria yield. This research clarifies the underlying mechanisms through which fertilization exerts an impact on the C–N–P stoichiometry within the leaf–soil–microbial system. Moreover, it furnishes a scientific foundation for the optimization of fertilization management strategies in Sapindus saponaria plantations.

Background: Understanding how agricultural management practices regulate plant–soil–microbe interactions through carbon (C)–nitrogen (N)–phosphorus (P) stoichiometry is essential for optimizing nutrient cycling in orchard ecosystems. The integration of cover crops with fertilization represents a promising strategy to enhance ecosystem functionality, yet the stoichiometric pathways through which these practices jointly influence citrus productivity remain poorly quantified. Method: A field experiment was conducted in a citrus orchard to evaluate the combined effects of fertilization and smooth vetch ( Vicia villosa Roth var. glabrescens Koch) cover on soil–plant–microbial nutrient dynamics. Four treatments were established in a factorial design: VF (fertilization with smooth vetch cover), VnoF (fertilization without smooth vetch cover), VFno (no fertilization with smooth vetch cover), and VnoFno (no fertilization and no smooth vetch cover). Soil properties, leaf nutrient concentrations, microbial characteristics, and fruit yield were systematically analyzed. Results: Smooth vetch and fertilization exhibited distinct regulatory effects on soil–plant nutrient dynamics. Smooth vetch alone exerted minimal influence on soil properties, whereas fertilization consistently reduced soil C:N ratio regardless of vetch presence, likely attributable to accelerated residue decomposition and enhanced nutrient cycling. The VF treatment significantly decreased soil C:N ratio while showing no significant effects on C:P and N:P ratios, indicating selective impacts on N rather than P dynamics. Notably, these soil responses did not directly translate to plant nutrient patterns. Leaf P concentrations varied significantly among treatments, peaking in VFno and reaching a minimum in VnoF, which drove substantial shifts in leaf C:P and N:P ratios—demonstrating a decoupling between soil and plant stoichiometric responses. Based on Random Forest analysis, soil microbial properties were the strongest predictors of leaf nutrient variations. PLS-PM analysis revealed that soil clay content had the largest total effect on yield, likely through its association with aggregate stability and nutrient retention. The VF treatment produced a mean fruit yield of 3.58 ± 0.76 kg tree ⁻¹ , which was 13-fold higher than VnoFno (0.28 ± 0.04 kg tree ⁻¹ ; p < 0.01). The entropy-weighted TOPSIS comprehensive evaluation integrating yield, soil quality, and environmental indicators confirmed VF as the optimal treatment (relative closeness coefficient C _i = 0.535 vs. 0.457 for VnoFno). Conclusions: Integrating balanced fertilization with smooth vetch cover substantially enhances citrus orchard productivity, with microbial biomass stoichiometry identified as the strongest predictor linking soil nutrient status to leaf nutrition and yield. This combined management practice is recommended for subtropical orchards to improve nutrient cycling efficiency and ecosystem functionality, though causal mechanistic pathways require further experimental validation.