Scholarlink

<p>Accurate crop yield prediction is essential for ensuring food security and serves as a critical reference for managing the supply and demand of agricultural products, stabilizing market prices, and safeguarding farm incomes. To enable early yield forecasting, we developed a rice yield prediction system by integrating remote sensing time-series data, grain number observations, simulated yields from a crop growth model, and statistical yield data. We employed a Gated Recurrent Unit (GRU) model, a deep learning architecture optimized for sequential data, to predict rice yields. The dataset was split into training and validation sets at a 9:1 ratio. The GRU model was configured with a structure including five GRU hidden layers followed by two fully connected layers and was trained using the Adamax optimizer for enhanced accuracy. Model training was conducted with 1000 epochs, a learning rate of 0.005, and a batch size of 20. A total of 28 input variables were used, including simulated yields, grain number, and the Normalized Difference Vegetation Index (NDVI) time-series data from day of year (DOY) 105–233, derived from MYD13Q1 and MYD09Q1 products. Yield predictions were performed at the provincial level and aggregated to derive national-scale estimates. Evaluation results for the 2024 growing season demonstrated high predictive performance, with the Root Mean Square Error (RMSE) ranging from 1.5 to 3.4&#xa0;kg 10a⁻¹ and the Normalized Root Mean Square Error (NRMSE) ranging from 0.29 to 0.67%. While further validation is warranted, the proposed GRU-based yield prediction system demonstrates strong potential as a reliable tool for early yield forecasting and improved food security.</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

Development of a prototype system for a rice yield prediction using deep learning

  • 期刊论文

  • 发表日期：

    2026年4月30日

查看全文
  • Ho-Young Ban,
  • Seo-Young Yang,
  • Ju-Hee Kim,
  • Jae-Kyeong Baek,
  • Yeong-Seo Song,
  • Yu-Na Kim,
  • Kyung-Do Lee,
  • Sera Jo,
  • Kyo-Moon Shim,
  • Suk-won Roh,
  • Miok Eom,
  • Jiyoung Shon,
  • Woon-Ha Hwang,
  • Jung-Il Cho,
  • Suk-Young Hong,
  • Jong-Min Ko,
  • Hyeon-Seok Lee

摘要

Accurate crop yield prediction is essential for ensuring food security and serves as a critical reference for managing the supply and demand of agricultural products, stabilizing market prices, and safeguarding farm incomes. To enable early yield forecasting, we developed a rice yield prediction system by integrating remote sensing time-series data, grain number observations, simulated yields from a crop growth model, and statistical yield data. We employed a Gated Recurrent Unit (GRU) model, a deep learning architecture optimized for sequential data, to predict rice yields. The dataset was split into training and validation sets at a 9:1 ratio. The GRU model was configured with a structure including five GRU hidden layers followed by two fully connected layers and was trained using the Adamax optimizer for enhanced accuracy. Model training was conducted with 1000 epochs, a learning rate of 0.005, and a batch size of 20. A total of 28 input variables were used, including simulated yields, grain number, and the Normalized Difference Vegetation Index (NDVI) time-series data from day of year (DOY) 105–233, derived from MYD13Q1 and MYD09Q1 products. Yield predictions were performed at the provincial level and aggregated to derive national-scale estimates. Evaluation results for the 2024 growing season demonstrated high predictive performance, with the Root Mean Square Error (RMSE) ranging from 1.5 to 3.4 kg 10a⁻¹ and the Normalized Root Mean Square Error (NRMSE) ranging from 0.29 to 0.67%. While further validation is warranted, the proposed GRU-based yield prediction system demonstrates strong potential as a reliable tool for early yield forecasting and improved food security.