Scholarlink

<p>Improved crop yield prediction has significant implications for food security, resource optimization, and climate change adaptation. In this direction, Deep learning (DL) models and environmental data enable more precise and reliable crop yield predictions, which can inform policy decisions and help optimize resource allocation for agricultural productivity. However, the effectiveness of these models for wheat yield estimation based on stage-wise environmental factors (rainfall, min. and max. temperature) has not been thoroughly investigated. This study fills literature gaps by determining minimal monthly data needs for precise wheat yield estimation and identifying the best DL model for district-wise estimation among the five state-of-the-art models (TCN, LSTM, GRU, Transformer and FFNN) using Full-factorial Design-based statistical assessment. In addition, the study systematically analyses district-level wheat yield data spanning a decade (2011-2021) and the corresponding countrywide weather data for the same period. The study results indicate that DL models based on Temporal Convolutional Networks exhibit higher precision in estimating wheat yield using environmental data, with a reported RMSE of 0.67 t/ha and MAE of 0.44 t/ha. By determining minimal data requirements and showcasing comparisons of the models, the study informs policymakers and agricultural practitioners, facilitating more informed decisions for resource allocation and improved productivity. Moreover, the accurate crop yield prediction methods described in the study can help provide early warning of potential crop failures, which can help ensure food security and improve agricultural productivity.</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

Nationwide wheat yield forecasting using deep learning models: A full-factorial analysis of early-stage prediction

  • 期刊论文

  • 发表日期：

    2025年9月12日

查看全文
  • Vanaja V,
  • G. Avinash,
  • Samarth Godara,
  • Pratap S. Birthal,
  • Rajni Jain,
  • Deepak Singh

摘要

Improved crop yield prediction has significant implications for food security, resource optimization, and climate change adaptation. In this direction, Deep learning (DL) models and environmental data enable more precise and reliable crop yield predictions, which can inform policy decisions and help optimize resource allocation for agricultural productivity. However, the effectiveness of these models for wheat yield estimation based on stage-wise environmental factors (rainfall, min. and max. temperature) has not been thoroughly investigated. This study fills literature gaps by determining minimal monthly data needs for precise wheat yield estimation and identifying the best DL model for district-wise estimation among the five state-of-the-art models (TCN, LSTM, GRU, Transformer and FFNN) using Full-factorial Design-based statistical assessment. In addition, the study systematically analyses district-level wheat yield data spanning a decade (2011-2021) and the corresponding countrywide weather data for the same period. The study results indicate that DL models based on Temporal Convolutional Networks exhibit higher precision in estimating wheat yield using environmental data, with a reported RMSE of 0.67 t/ha and MAE of 0.44 t/ha. By determining minimal data requirements and showcasing comparisons of the models, the study informs policymakers and agricultural practitioners, facilitating more informed decisions for resource allocation and improved productivity. Moreover, the accurate crop yield prediction methods described in the study can help provide early warning of potential crop failures, which can help ensure food security and improve agricultural productivity.