Related Experiment Videos
Explaining the spatial heterogeneity in soil-rice yield interactions via a hybrid interpretable machine learning and
Meiling Sheng1, Xufeng Fei2, Zhaohan Lou3
1Zhejiang Academy of Agricultural Sciences, Hangzhou, China; Key Laboratory of Information Traceability of Agriculture Products, Ministry of Agriculture and Rural Affairs, China; State Key Laboratory for Quality and Safety of Agro-Products, China.
Journal of Environmental Management
|July 31, 2026
Summary
Understanding soil properties is key to boosting rice yield for global food security. This study uses a hybrid framework to identify specific soil factors influencing rice production across different regions, aiding precision agriculture.
Area of Science:
- Agricultural Science
- Environmental Science
- Data Science
Background:
- Rice (Oryza sativa L.) is a staple food for over half the world's population, making its production critical for food security and sustainable development.
- Understanding the complex interplay between environmental variables, particularly soil properties, and rice yield is essential for optimizing agricultural practices.
Purpose of the Study:
- To elucidate the spatially heterogeneous relationships between environmental variables and rice yield in Jiaxing city.
- To identify specific soil properties that significantly influence rice production and to map these relationships spatially.
Main Methods:
- A hybrid framework combining random forest (RF), Shapley additive explanations (SHAP), and bivariate local spatial autocorrelation (BI-LISA) was employed.
- High-resolution soil and remote sensing datasets were utilized to analyze environmental factors affecting rice yield.
Main Results:
- Soil physical and chemical factors collectively contributed approximately 66.8% to rice yields.
- BI-LISA analysis identified distinct soil-related constraints on rice yield in different regions: southwest (high bulkiness, sandy/acidic soil, low nutrients), north (low bulkiness, clay/acidic soil, imbalanced nutrients), and east coast (low bulkiness, sandy/alkaline soil, low nutrients).
- Mean soil conditions were generally suitable, with a mean rice yield of 8437 kg/ha.
Conclusions:
- The hybrid RF-SHAP-BI-LISA framework effectively reveals spatially heterogeneous relationships between soil properties and rice yield.
- The findings provide valuable insights for precision agriculture, enabling targeted nutrient management to address soil-specific limitations and enhance rice production.
