Scholarlink

<p>Efficient crop fertilizer management is critical for promoting sustainable agriculture, enhancing food security, and minimizing environmental impact. Despite recent advances in machine learning (ML) and remote sensing (RS), existing systems often lack precision, scalability, and contextual adaptability for fertilizer recommendation. This study presents an advanced, data-driven fertilizer prediction framework that integrates remote sensing with multiple machine learning techniques to optimize fertilizer application for key crops, including maize, rice, and wheat. The research addresses the gap in context-aware, scalable solutions by combining the Hybrid Multi-Layer Slow Sand Filter (HML-SSF) for environmental filtering, the Filter-Based Decision Tree (FBDT) for feature selection, Context-Aware Graph Attention Networks (CA-GAT) for spatial learning, and Graph Neural Networks with Spatial-Temporal Analysis (GNN-STA) for dynamic modelling. The system utilizes the publicly available Kaggle Crop Yield Prediction dataset, incorporating multi-dimensional features such as climate, soil nutrients, and crop-specific variables. Pre-processing includes RS data calibration, missing value treatment, and normalization. The FBDT improves model efficiency by selecting high-impact features, while CA-GAT captures spatial dependencies. GNN-STA models temporal and contextual evolution in agricultural conditions. Hyperparameters are optimized through grid search, and model performance is evaluated using MAE, RMSE, and R², supported by cross-validation. Results show that integrated models achieve prediction accuracies ranging from 70% to 95%, with CA-GAT and GNN-STA delivering superior performance in capturing spatial-temporal fertilizer needs. The system demonstrates practical relevance for real-time applications in precision agriculture, with improved accuracy, reduced fertilizer waste, and environmental sustainability. This work contributes a novel hybrid approach to crop fertilizer management, validated on real-world agricultural data, and sets a foundation for extending the model across different geographies and crop types.</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

Effective Prediction System for Crop Fertilizer Management in Sustainable Agriculture Using Remote Sensing and Machine Learning Techniques

  • 期刊论文

  • 发表日期：

    2026年3月9日

查看全文
  • Sneha Suhas Koli,
  • Tushar H. Ghorpade,
  • Vanita Mane

摘要

Efficient crop fertilizer management is critical for promoting sustainable agriculture, enhancing food security, and minimizing environmental impact. Despite recent advances in machine learning (ML) and remote sensing (RS), existing systems often lack precision, scalability, and contextual adaptability for fertilizer recommendation. This study presents an advanced, data-driven fertilizer prediction framework that integrates remote sensing with multiple machine learning techniques to optimize fertilizer application for key crops, including maize, rice, and wheat. The research addresses the gap in context-aware, scalable solutions by combining the Hybrid Multi-Layer Slow Sand Filter (HML-SSF) for environmental filtering, the Filter-Based Decision Tree (FBDT) for feature selection, Context-Aware Graph Attention Networks (CA-GAT) for spatial learning, and Graph Neural Networks with Spatial-Temporal Analysis (GNN-STA) for dynamic modelling. The system utilizes the publicly available Kaggle Crop Yield Prediction dataset, incorporating multi-dimensional features such as climate, soil nutrients, and crop-specific variables. Pre-processing includes RS data calibration, missing value treatment, and normalization. The FBDT improves model efficiency by selecting high-impact features, while CA-GAT captures spatial dependencies. GNN-STA models temporal and contextual evolution in agricultural conditions. Hyperparameters are optimized through grid search, and model performance is evaluated using MAE, RMSE, and R², supported by cross-validation. Results show that integrated models achieve prediction accuracies ranging from 70% to 95%, with CA-GAT and GNN-STA delivering superior performance in capturing spatial-temporal fertilizer needs. The system demonstrates practical relevance for real-time applications in precision agriculture, with improved accuracy, reduced fertilizer waste, and environmental sustainability. This work contributes a novel hybrid approach to crop fertilizer management, validated on real-world agricultural data, and sets a foundation for extending the model across different geographies and crop types.