Scholarlink

<p>Agricultural forecasting faces complex interactions between crop traits, environmental conditions, and shifting consumption patterns, requiring adaptive predictive models. In this paper, we propose <Emphasis FontCategory="NonProportional">AFFAC</Emphasis> (Advanced Forecasting Framework for Agricultural Crops), a data-driven framework designed to enhance production forecasts, with a focus on olive cultivation a cornerstone of Mediterranean agriculture. <Emphasis FontCategory="NonProportional">AFFAC</Emphasis> integrates tailored parameters for olive groves and leverages deep learning architectures (LSTM, CNN-LSTM, GRU, and Transformer) to improve decision-making. Experimental results demonstrate an 18% reduction in prediction error versus ARIMA (RMSE=7.4 vs. 12.5), with sensitivity analysis identifying 17 key parameters (95% cumulative importance). The framework excels in modeling non-linear patterns (MAE &lt; 0.5 for critical months) and offers a modular design for customizable model integration, providing actionable insights for sustainable farming practices such as dynamic yield prediction under drought scenarios, aiding policymakers in risk management.</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

A framework for modeling and testing forecasting in agricultural production systems

  • 期刊论文

  • 发表日期：

    2025年8月2日

查看全文
  • Souhila Chadli,
  • Abdelkader Ouared,
  • Abdelkader Dellel,
  • Nour Islam Bachari,
  • Abdelhafid Chadli

摘要

Agricultural forecasting faces complex interactions between crop traits, environmental conditions, and shifting consumption patterns, requiring adaptive predictive models. In this paper, we propose AFFAC (Advanced Forecasting Framework for Agricultural Crops), a data-driven framework designed to enhance production forecasts, with a focus on olive cultivation a cornerstone of Mediterranean agriculture. AFFAC integrates tailored parameters for olive groves and leverages deep learning architectures (LSTM, CNN-LSTM, GRU, and Transformer) to improve decision-making. Experimental results demonstrate an 18% reduction in prediction error versus ARIMA (RMSE=7.4 vs. 12.5), with sensitivity analysis identifying 17 key parameters (95% cumulative importance). The framework excels in modeling non-linear patterns (MAE < 0.5 for critical months) and offers a modular design for customizable model integration, providing actionable insights for sustainable farming practices such as dynamic yield prediction under drought scenarios, aiding policymakers in risk management.