Scholarlink

<p>Crop yield prediction is a key issue that can contribute to food security by enabling better decision-making in agricultural planning, supporting food consumption, and reducing the effects of climate variability. However, the majority of current soil–vegetation modelling methodologies are either limited to single-source data or utilise linear statistical models, which do not capture the nonlinear interactions among soil health, time-varying atmospheric conditions, and fine-scale vegetation attributes derived from remote sensing. In recent years, Generic deep learning approaches have had shortcomings, such as the inability to generalise across regions, inefficiency, and low interpretability, which have discouraged their successful deployment in the field (on farms) as a workable tool in real-life agricultural settings. To fill these gaps, this paper presents AgriYieldAI, a decision-support system based on Artificial Intelligence (AI) for predicting crop yields using a newly developed deep learning method, YieldFusionNet, described in this work. Our architecture consists of three branches with three related modules (MLP, LSTM, and CNN) for soil data, time-series weather data, and satellite images (features from three sources). The model architecture is performance-, generalizability-, and explainability-centric—SHAP-based feature importance visualisation. SHAP-based analysis is applied to visualise feature importance based on NDVI, day5 rain, day30 rain, and soil pH from the day30 database. It was assessed on a range of real-world public datasets. As a result, it achieved an RMSE of 0.431, an MAE of 0.292, and an R² of 0.91, surpassing traditional state-of-the-art models such as Random Forest and Support Vector Regression (SVR). Additionally, a regional case study in the Guntur region of India illustrated the practical use of this method. AgriYieldAI may be a powerful, explainable crop yield forecasting solution that farmers, policymakers, and agronomic planners can use to obtain interpretable, data-driven insights for sustainable agricultural practices.</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

AgriYieldAI an AI driven decision support system for crop yield prediction using YieldFusionNet

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月21日

查看全文
  • M. Snehalatha,
  • Anitha Patil

摘要

Crop yield prediction is a key issue that can contribute to food security by enabling better decision-making in agricultural planning, supporting food consumption, and reducing the effects of climate variability. However, the majority of current soil–vegetation modelling methodologies are either limited to single-source data or utilise linear statistical models, which do not capture the nonlinear interactions among soil health, time-varying atmospheric conditions, and fine-scale vegetation attributes derived from remote sensing. In recent years, Generic deep learning approaches have had shortcomings, such as the inability to generalise across regions, inefficiency, and low interpretability, which have discouraged their successful deployment in the field (on farms) as a workable tool in real-life agricultural settings. To fill these gaps, this paper presents AgriYieldAI, a decision-support system based on Artificial Intelligence (AI) for predicting crop yields using a newly developed deep learning method, YieldFusionNet, described in this work. Our architecture consists of three branches with three related modules (MLP, LSTM, and CNN) for soil data, time-series weather data, and satellite images (features from three sources). The model architecture is performance-, generalizability-, and explainability-centric—SHAP-based feature importance visualisation. SHAP-based analysis is applied to visualise feature importance based on NDVI, day5 rain, day30 rain, and soil pH from the day30 database. It was assessed on a range of real-world public datasets. As a result, it achieved an RMSE of 0.431, an MAE of 0.292, and an R² of 0.91, surpassing traditional state-of-the-art models such as Random Forest and Support Vector Regression (SVR). Additionally, a regional case study in the Guntur region of India illustrated the practical use of this method. AgriYieldAI may be a powerful, explainable crop yield forecasting solution that farmers, policymakers, and agronomic planners can use to obtain interpretable, data-driven insights for sustainable agricultural practices.