Scholarlink

<p>Agriculture is a fundamental source of livelihood in developing and underdeveloped nations and a key contributor to global food security. However, increasing climate volatility and agricultural stress enormously affect agrarian practices. Farmers face chronic stress from droughts, temperature extremes, and land degradation. Without precise identification of the most vulnerable zones, adaptation and resource allocation remains ad hoc, perpetuating yield losses and economic constraints. This study considered 26 pertinent variables categorised into Exposure, Sensitivity, and Adaptive Capacity. The key drivers of agricultural distress identified drought proneness, changes in maximum temperatures, the dimensionless precipitation anomaly index, and land degradation. The multi-layer perceptron neural network (MLPNN) was used as a benchmark model, followed by the DenseNet neural network. Despite these, an innovative Hybrid Deep Ensemble Learning model was developed with CNN, LSTM, and DFNN dynamically weighted meta-learner to reduce overfitting and to predict the agriculture distress-prone region. The findings show that MLPNN defined 24.31% of the study region as distress-prone, DenseNet recognised 17.17%, and the Hybrid model identified 16.37%. Furthermore, the Hybrid Deep Ensemble model had the best predictive performance, with an ROC-AUC of 93.8% and an F1 score of 88.7%. The methodology tested in this study may assist policymakers in developing an adaptation strategy by implementing effective risk mitigation techniques, increasing agricultural resilience, and ensuring sustainable growth in the context of climatic uncertainty in the same Geographical settings of the different regions.</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

Analysing agricultural distress in the eastern plateau of West Bengal's Rarh Region: integrating hybrid deep ensemble and GIS-based soft computing

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年6月18日

查看全文
  • Gopal Chowdhury,
  • Ashis Kumar Saha

摘要

Agriculture is a fundamental source of livelihood in developing and underdeveloped nations and a key contributor to global food security. However, increasing climate volatility and agricultural stress enormously affect agrarian practices. Farmers face chronic stress from droughts, temperature extremes, and land degradation. Without precise identification of the most vulnerable zones, adaptation and resource allocation remains ad hoc, perpetuating yield losses and economic constraints. This study considered 26 pertinent variables categorised into Exposure, Sensitivity, and Adaptive Capacity. The key drivers of agricultural distress identified drought proneness, changes in maximum temperatures, the dimensionless precipitation anomaly index, and land degradation. The multi-layer perceptron neural network (MLPNN) was used as a benchmark model, followed by the DenseNet neural network. Despite these, an innovative Hybrid Deep Ensemble Learning model was developed with CNN, LSTM, and DFNN dynamically weighted meta-learner to reduce overfitting and to predict the agriculture distress-prone region. The findings show that MLPNN defined 24.31% of the study region as distress-prone, DenseNet recognised 17.17%, and the Hybrid model identified 16.37%. Furthermore, the Hybrid Deep Ensemble model had the best predictive performance, with an ROC-AUC of 93.8% and an F1 score of 88.7%. The methodology tested in this study may assist policymakers in developing an adaptation strategy by implementing effective risk mitigation techniques, increasing agricultural resilience, and ensuring sustainable growth in the context of climatic uncertainty in the same Geographical settings of the different regions.