Scholarlink

Purpose <p>This study focuses on modeling selenium (Se) accumulation in soil‒rice systems within karst areas, a task crucial for understanding and managing micronutrient bioavailability in such challenging agricultural environments. The primary aim is not only to predict selenium concentrations in rice grains but to delve into the key factors influencing Se transfer from soil to rice, identifying gaps in current knowledge and introducing novel insights by comparing multiple machine learning models, especially the advanced XGBoost algorithm.</p> Methods <p>In this research, four machine learning (ML) models, namely, multiple linear regression (MLR), K-nearest neighbor (KNN), support vector regression (SVR), and extreme gradient boosting (XGBoost), were constructed on the basis of 10 soil factors to predict Se uptake in rice grains collected from Liujiang District.</p> Results <p>The results revealed that the average concentrations of Se in the soil and rice grains were 0.38 mg/kg and 0.063 mg/kg, respectively. The XGBoost model exhibited the best performance for the rice Se content (R<sup>2</sup> = 0.805), followed by the SVR model (R<sup>2</sup> = 0.691) and the KNN model (R<sup>2</sup> = 0.662), which exhibited adequate performance for prediction. The MLR model demonstrated the least predictive accuracy, with an R<sup>2</sup> value of 0.358. The feature importance, SHAP, and correlation analyses revealed that the soil total organic carbon, S, and extractable S contents had significant impacts, whereas the soil Se, CaO, and Cd contents and pH values were also identified as contributing factors to Se uptake in rice grains.</p> Conclusion <p>The study underscores a complex and nonlinear interaction among Se transport factors in the karst soil‒rice system, with the XGBoost model providing a new dimension of predictive accuracy and insight into selenium dynamics. These findings provide critical insights for enhancing selenium bioavailability in agricultural systems globally, particularly in karst regions.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Application of machine learning methods for predicting selenium accumulation in the soil‒rice system of a typical karst area

  • 期刊论文

  • 发表日期：

    2025年4月3日

查看全文
  • Molan Tang,
  • Bolun Fan,
  • Guanghui Lu,
  • Xiaodong Yang,
  • Shuyun Xie,
  • Huan Tian,
  • Zhengyu Bao,
  • Jianhui Jiang,
  • Biquan Chen,
  • Tinglei Huang

摘要

Purpose

This study focuses on modeling selenium (Se) accumulation in soil‒rice systems within karst areas, a task crucial for understanding and managing micronutrient bioavailability in such challenging agricultural environments. The primary aim is not only to predict selenium concentrations in rice grains but to delve into the key factors influencing Se transfer from soil to rice, identifying gaps in current knowledge and introducing novel insights by comparing multiple machine learning models, especially the advanced XGBoost algorithm.

Methods

In this research, four machine learning (ML) models, namely, multiple linear regression (MLR), K-nearest neighbor (KNN), support vector regression (SVR), and extreme gradient boosting (XGBoost), were constructed on the basis of 10 soil factors to predict Se uptake in rice grains collected from Liujiang District.

Results

The results revealed that the average concentrations of Se in the soil and rice grains were 0.38 mg/kg and 0.063 mg/kg, respectively. The XGBoost model exhibited the best performance for the rice Se content (R2 = 0.805), followed by the SVR model (R2 = 0.691) and the KNN model (R2 = 0.662), which exhibited adequate performance for prediction. The MLR model demonstrated the least predictive accuracy, with an R2 value of 0.358. The feature importance, SHAP, and correlation analyses revealed that the soil total organic carbon, S, and extractable S contents had significant impacts, whereas the soil Se, CaO, and Cd contents and pH values were also identified as contributing factors to Se uptake in rice grains.

Conclusion

The study underscores a complex and nonlinear interaction among Se transport factors in the karst soil‒rice system, with the XGBoost model providing a new dimension of predictive accuracy and insight into selenium dynamics. These findings provide critical insights for enhancing selenium bioavailability in agricultural systems globally, particularly in karst regions.