Scholarlink

<p>Accurate crop yield forecasting is essential for strategic agricultural planning, especially in climate-sensitive regions like the Kashmir Valley of India. This study evaluates and compares the performance of five individual models, their ensembled combinations and an Artificial Neural Network (ANN) for mustard yield prediction using time series data and agrometeorological indices across five key districts: Anantnag, Baramulla, Budgam, Kulgam, and Srinagar. Weekly weather data from local observatories, supplemented by NASA POWER (NASA Prediction Of Worldwide Energy Resources) datasets, and historical mustard yield records (1999–2023) were used to develop predictive models. Among all models, Random Forest (RF) emerged as the most consistently reliable across all districts, achieving R² values of 0.90–0.97 during training and maintaining strong testing performance, particularly in Budgam (R² = 0.99), Kulgam (0.97), and Anantnag (0.90). While ANN outperformed all models in Anantnag, Budgam, and Kulgam with R² ≥ 0.95 during testing, it failed to generalize in Baramulla and Srinagar, recording R² values below 0.05. Conversely, RF maintained stable predictions even in these challenging districts, reinforcing its robustness against climatic and data variability. The findings challenge the assumption that ensemble models always surpass individual learners, highlighting the superior performance and generalizability of RF. While ANN is recommended for localized use in Anantnag, Budgam, and Kulgam, RF is suitable for reliable mustard yield prediction across all five districts. This study underscores the importance of region-specific model evaluation in agricultural forecasting and advocates cautious deployment of complex models without validation.</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

Time series-based mustard yield prediction under climate variability: superior performance of individual and ANN models over ensembles

  • 期刊论文

  • 发表日期：

    2026年4月16日

查看全文
  • Sameera Qayoom,
  • Anurag Satpathi,
  • Neha Chand,
  • Latief Ahmad,
  • Mohd Ayoub Malik,
  • Afshan Tabassum,
  • K. Nirmal Ravi Kumar,
  • Nadhir Al-Ansari,
  • Ahmed Z. Dewidar,
  • Mohamed A. Mattar

摘要

Accurate crop yield forecasting is essential for strategic agricultural planning, especially in climate-sensitive regions like the Kashmir Valley of India. This study evaluates and compares the performance of five individual models, their ensembled combinations and an Artificial Neural Network (ANN) for mustard yield prediction using time series data and agrometeorological indices across five key districts: Anantnag, Baramulla, Budgam, Kulgam, and Srinagar. Weekly weather data from local observatories, supplemented by NASA POWER (NASA Prediction Of Worldwide Energy Resources) datasets, and historical mustard yield records (1999–2023) were used to develop predictive models. Among all models, Random Forest (RF) emerged as the most consistently reliable across all districts, achieving R² values of 0.90–0.97 during training and maintaining strong testing performance, particularly in Budgam (R² = 0.99), Kulgam (0.97), and Anantnag (0.90). While ANN outperformed all models in Anantnag, Budgam, and Kulgam with R² ≥ 0.95 during testing, it failed to generalize in Baramulla and Srinagar, recording R² values below 0.05. Conversely, RF maintained stable predictions even in these challenging districts, reinforcing its robustness against climatic and data variability. The findings challenge the assumption that ensemble models always surpass individual learners, highlighting the superior performance and generalizability of RF. While ANN is recommended for localized use in Anantnag, Budgam, and Kulgam, RF is suitable for reliable mustard yield prediction across all five districts. This study underscores the importance of region-specific model evaluation in agricultural forecasting and advocates cautious deployment of complex models without validation.