After years of cultivation, tea garden soils gradually become acidified and compacted. Liming to ameliorate soil acidification alters the soil microbial activity and community structure, ultimately affecting the soil nitrogen cycling and nitrogen use efficiency of tea plants. In this study, ammonium-preferring tea plants (Camellia sinensis) were cultivated in lime-amended soils across a pH gradient (pH 4.5, pH 5.5, pH 6.5, and pH 7.5) to assess the impacts of soil liming on the structure and composition of the rhizosphere microbial community. The results demonstrated that, as the soil pH increased, the diversity and richness of both bacterial and fungal communities exhibited a gradual decline. The biomasses of AOA and AOB were dominated in acid soils and alkaline soils, respectively. The abundance of Proteobacteria, Actinobacteria, and Bacteroidetes in pH 6.5 and pH 7.5 were 16.56–22.56%, 16.29–18.09%, and 1.65–4.52% times higher than those in pH 4.5 and pH 5.5 soils. Ascomycota and Basidiomycota were accounting for 89.93% of all the species. Across the rising pH gradient, the relative abundance of Ascomycota significantly increased by 9.64–20.49%, whereas Basidiomycota decreased by 1.11–15.01%. The RDA analysis results showed the soil pH was the main effect factor for the differences in the structure and composition of bacterial and fungal communities. This conclusion provides theoretical support for the optimization of acidic soil improvement techniques after long-term tea cultivation.

Accurate crop yield estimation under differentiated management practices is a core requirement for the development of smart agriculture. However, current yield estimation models face two major challenges: limited adaptability to different management practices, thus exhibiting poor generalizability, and ineffective integration of multi-source remote sensing features, limiting further improvements in estimation accuracy. To address these issues, this study integrated UAV-based multispectral and thermal infrared remote sensing data to propose a yield estimation framework based on multi-source feature fusion. First, three machine learning algorithms—Partial Least Squares Regression (PLSR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost)—were employed to retrieve key biochemical parameters of winter wheat. The RF model demonstrated superior performance, with retrieval accuracies for chlorophyll, nitrogen, and phosphorus contents of R² = 0.8347, 0.5914, and 0.9364 and RMSE = 0.2622, 0.4127, and 0.0236, respectively. Subsequently, yield estimation models were constructed by integrating the retrieved biochemical parameters with phenotypic traits such as plant height and biomass. The RF model again exhibited superior performance (R² = 0.66, RMSE = 867.28 kg/ha). SHapley Additive exPlanations (SHAP) analysis identified May chlorophyll content (Chl-5) and March chlorophyll content (Chl-3) as the most critical variables for yield prediction, with stable positive contributions to yield when their values exceeded 2.80 mg/g and 2.50 mg/g, respectively. The quantitative assessment of management practices revealed that the straw return + 50% inorganic fertilizer + 50% organic fertilizer (RIO50) treatment under the combined organic–inorganic fertilization regime achieved the highest measured grain yield (11,469 kg/ha). Consequently, this treatment can be regarded as an optimized practice for attaining high yield. This study confirms that focusing on chlorophyll dynamics during key physiological stages is an effective approach for enhancing yield estimation accuracy under varied management practices, providing a technical basis for precise field management.