Scholarlink

<p>Accurate prediction of paddy yield is essential for ensuring food security, optimizing agricultural management, and supporting policy decisions in regions characterized by intensive paddy cultivation. This study develops a comprehensive machine learning-based framework to model and explain paddy yield variability using a combination of meteorological parameters, MODIS-derived vegetation indices, and ancillary crop yield datasets for the lower part of the Ganga Basin. A series of preprocessing steps including spatial and temporal aggregation, standardization, and correlation analysis were applied to ensure data integrity and robustness. Four machine learning models, namely Random Forest (RF), Gradient Boosting Machine (GBM), Support Vector Machine (SVM), and Multilayer Perceptrons (MLP), were trained using an 80:20 split and validated through 9-fold cross-validation. Model evaluation metrics such as R<sup>2</sup> (0.818) RMSE (0.206) and MAE (0.143), demonstrated that RF achieved the highest predictive accuracy, outperforming other approaches due to its capability to capture nonlinear relationships and variable interactions. To enhance interpretability, SHAP (SHapley Additive exPlanations) analysis was employed, providing a transparent assessment of feature contributions across models. The results revealed that solar radiation, soil moisture, NDVI, and ET exerted the strongest influence on yield prediction, aligning with physiological drivers of paddy growth. Model robustness and reliability were further supported by consistent feature importance trends across multiple algorithms. By integrating remote sensing, meteorology, and machine learning, this study generates reliable yield estimates to support early warning systems and precision agriculture.</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

Predictive modeling of paddy yields with MODIS variables and meteorological data using machine learning

  • 期刊论文

  • 发表日期：

    2026年5月30日

查看全文
  • Piyasa Gayen,
  • Dipanwita Dutta

摘要

Accurate prediction of paddy yield is essential for ensuring food security, optimizing agricultural management, and supporting policy decisions in regions characterized by intensive paddy cultivation. This study develops a comprehensive machine learning-based framework to model and explain paddy yield variability using a combination of meteorological parameters, MODIS-derived vegetation indices, and ancillary crop yield datasets for the lower part of the Ganga Basin. A series of preprocessing steps including spatial and temporal aggregation, standardization, and correlation analysis were applied to ensure data integrity and robustness. Four machine learning models, namely Random Forest (RF), Gradient Boosting Machine (GBM), Support Vector Machine (SVM), and Multilayer Perceptrons (MLP), were trained using an 80:20 split and validated through 9-fold cross-validation. Model evaluation metrics such as R2 (0.818) RMSE (0.206) and MAE (0.143), demonstrated that RF achieved the highest predictive accuracy, outperforming other approaches due to its capability to capture nonlinear relationships and variable interactions. To enhance interpretability, SHAP (SHapley Additive exPlanations) analysis was employed, providing a transparent assessment of feature contributions across models. The results revealed that solar radiation, soil moisture, NDVI, and ET exerted the strongest influence on yield prediction, aligning with physiological drivers of paddy growth. Model robustness and reliability were further supported by consistent feature importance trends across multiple algorithms. By integrating remote sensing, meteorology, and machine learning, this study generates reliable yield estimates to support early warning systems and precision agriculture.