Wei Zhou

Introduction: Accurate and transferable rice yield prediction is essential for precision agriculture and food security, yet existing remote sensing–based models often suffer from limited generalization across regions, cultivars, and field scales. Methods: This study developed an interpretable automated machine learning framework for rice yield prediction using UAV-based multispectral imagery collected at the maturity stage. A total of 143 rice samples, including 79 experimental plots and 64 production fields across 15 counties in Sichuan Province, China, were investigated. 20 vegetation indices and 36 gray-level co-occurrence matrix texture features were extracted from multispectral orthomosaics, and three feature selection strategies: Pearson correlation coefficient (PCC), Random Forest feature importance (RF-I), and AutoGluon feature importance(AutoGluon-I), were systematically compared. Four regression approaches, including CatBoost, ExtraTrees, Random Forest, and an AutoGluon stacked ensemble, were evaluated using R ² , RMSE, and MAE. Results: The results showed that the AutoGluon ensemble consistently outperformed individual machine learning models, improving testset R ² from 0.403–0.670 to 0.528–0.736. The best performance was achieved by combining Pearson correlation-based feature selection with AutoGluon, yielding a training R ² of 0.821 and a test R ² of 0.736, with RMSE and MAE values of 0.749 and 0.568 t ha ⁻¹ , respectively. Shapley Additive Explanations (SHAP) analysis further revealed that texture features, particularly red-band contrast and angular second moment features, contributed substantially to yield prediction, indicating the importance of canopy structural heterogeneity at maturity. Discussion: Overall, the proposed PCC–AutoGluon–SHAP framework provides a lightweight, accurate, and interpretable approach for UAV-based rice yield estimation across heterogeneous field conditions, offering practical potential for scalable precision agriculture and regional yield monitoring.

Understanding the vertical distribution pattern of nitrogen in the canopy and selecting an ideal indicator leaf can provide a reference to guide precision nitrogen management. However, most existing studies have used SPAD values as indirect proxies for plant nitrogen status, and systematic evaluations of leaf position representativeness based on directly measured leaf nitrogen nutrition levels are lacking. A two-factor randomized block design with two rice varieties and four nitrogen application rates was conducted in 2022 and 2023. Canopy nitrogen distribution and its response to nitrogen application rate were investigated. The results showed that nitrogen content and accumulation in leaves at different positions were in the order L1, L2 > L3 > L4 (1st, 2nd, 3rd, and 4th fully expanded leaves from the top), and these values for L1-L4 were significantly and positively correlated with those of the entire canopy and plant (R² > 0.83). L4 was the most sensitive to nitrogen, whereas L2 was the most stable. Considering their representativeness, stability, and sensitivity, L2 and L4 can serve as target leaves for rapid field nitrogen diagnosis, and using L2 to predict nitrogen accumulation can streamline the screening of nitrogen-efficient varieties and reduce analytical costs. This study provides a theoretical basis and a new methodological reference for rice nitrogen nutrition diagnosis.

Crop rotation plays a critical role in enhancing cropping intensity and ensuring food security. To evaluate its long-term effects on soil quality, a fixed-site field experiment established in 2014 including four cropping systems—winter fallow–rice (Oryza sativa L.) (FR), potato (Solanum tuberosum L.) –maize (Zea mays L.) (PM), potato–rice (PR), and potato–rice → rapeseed (Brassica napus L.) –rice (RRPR)—was conducted. A minimum data set (MDS) was screened from 21 soil indicators via principal component analysis (PCA), and the soil quality index (SQI) was calculated by integrating membership functions and indicator weights to comprehensively evaluate the impact of different patterns on soil quality. Results showed that paddy–upland rotations (PR and RRPR) significantly improved soil physical properties, increasing soil moisture content, porosity, and macro-aggregate proportion by 2.27–10.17%, while reducing bulk density by 10.32–13.38%, compared to FR and PM. PR and RRPR rotations also increased total nitrogen (TN), available phosphorus (AP), and available potassium contents (AK) by 5.19–114.00% (p < 0.01). PM rotation notably enhanced available nutrients, with NH₄⁺-N, AP, and AK rising by 3.65–243.50% (p < 0.05), compared to FR. The MDS-based SQI, comprising NH₄⁺-N, AP, mean weight diameter, and soil porosity, showed a highly significant positive correlation with the total data set-based SQI (p < 0.0001). PM exhibited the highest and most stable SQI, exceeding other systems by 8.15–19.30%, while PR and RRPR increased SQI by 9.04–10.30%, compared to FR. In conclusion, potato-based cropping systems enhance soil quality by improving soil structure and increasing nutrient content and availability. The results of this study provide a theoretical basis for nutrient management and sustainable production in cropping systems.

Although significant differences in soil nitrogen levels exist under different paddy-upland rotations, the main reason for this is unclear. The nitrogen retention capacity and loss of ammonia volatilisation, leaching, etc. of paddy soil with large differences in nitrogen levels from two long-term rotations, garlic-rice and wheat-rice, were measured using the soil column simulation method. The results showed that the loss rate of leaching was only 5.4%, whereas that of ammonia volatilisation was up to 22.8%, which was the main nitrogen loss way of paddy soil under the two rotations. The average ammonia volatilisation rates under wheat-rice rotation with high and low nitrogen application rates were 12.1% and 40.2% higher than that under garlic-rice rotation, leading to a decrease in the total nitrogen loss amount and rate through ammonia volatilisation by 29.8% and 8.8%, respectively. As a result, nitrogen retention in the soil under garlic-rice rotation increased by 12.7%. In conclusion, the long-term garlic-rice rotation could significantly inhibit ammonia volatilisation, thus improving the soil nitrogen retention capacity. The straw return may increase soil organic matter content, reduce ammonia volatilisation loss, and enhance soil nitrogen retention capacity and productivity.

Sowing quality is directly influenced by the seed-filling characteristics of a centralized seed-metering device, and vigorous seed-motion status deteriorates seeding performance resulting from field surface roughness, seeder vibration, etc. However, the relationship between vibration and seed-filling properties remains unclear. This study measured the vibration characteristics of a seed-metering device, examined unmeasurable seed-motion characteristics based on DEM simulation, and used bench tests to determine the seed-filling characteristics under vibration conditions. The frequency distribution of the rapeseed seeder was 0–180 Hz. The simulation results revealed that the vibration mainly changed the level of seed in the seed-filling area; moreover, it significantly increased the seed-filling angle, ranging from 56.00°–59.00° to 69.78°–91.40°, which affected the seed-filling performance. A vibration frequency of 10–40 Hz resulted in a target seed rate within 88.67–99.44%, but when there is no vibration, the target seed rate reached 100%. The bench test results demonstrated that the seed-filling angle under vibration will increase with an increase in longitudinal distance, and will change between 15.49°–102°. The seed-filling performance affected by longitudinal distance was larger than that affected by lateral distance. By adjusting the longitudinal distance to range from 0–20 mm, the target seed rate could be maintained above 91%. The findings help to improve our knowledge of seed-filling characteristics under vibration conditions, and should be conducive to improving seeding performance by adjusting the seed layer height to reduce the impact of vibration on field operations.

Rapeseed is mainly cultivated in hilly areas in China and the inclination angle is an important parameter for influencing the seeding quality of a seed-metering device. However, the relationship between seeding performance and inclination angle remains unclear. This study analyzes the influence of inclination angle (θ) on the characteristics of seed-filling and seeding performance in a centralized seed-metering system of rapeseed using DEM simulation and the bench test. The results revealed that the inclination angle (θ) had a significant effect on seed-filling performance. The missing seeds ratio and the deviation of seed layer height increased with an increase in a lateral inclination angle. The missing seeds ratio in the forward direction (−Y direction) and seed-filling angle, double seed-filling ratio, and seed layer height in the back direction (Y direction) increased with the increasing inclination angle. Under seed-cleaning treatment, the missing seeds ratio was significantly reduced in the −Y direction at a seed layer height of 25 mm, and the qualified ratio was more than 91.0% at a seed layer height of 20 mm in X direction. The phenomenon of seeds passing through the seed-metering wheel disappeared with a qualified ratio larger than 90.0% in Y direction by installing a seed-cleaning brush. The results contributed to improving the knowledge of seed-filling characteristics of the centralized seed-metering device at inclination conditions and should help to improve seeding performance by adjusting the seed layer height and installing a seed-cleaning brush.