Scholarlink

<p>Potato late blight (PLB), caused by <i>Phytophthora infestans</i>, remains a major constraint to potato cultivation in West Bengal, India. This study examined the influence of key weather variables on PLB severity under early, normal, late, and overall planting conditions, and developed predictive models for disease severity (Percentage Disease Incidence, PDI) and disease initiation (First Appearance of potato late blight in days after planting, FADAP). Correlation analysis revealed minimum temperature as the most critical factor, exhibiting strong positive associations with PDI across all planting conditions (r up to 0.68), followed by maximum temperature and sunshine hours. Variable importance rankings from the Random Forest model consistently identified minimum temperature as the primary predictor, while maximum temperature or sunshine hours emerged as secondary determinants depending on planting time. Five modeling approaches, viz., ARIMA, ARIMAX, Support Vector Regression (SVR), Random Forest, and Time-Delay Neural Network with exogenous variables (TDNNx), were compared for PDI prediction. SVR achieved the highest accuracy for early planting, whereas TDNNx provided superior performance for normal, late, and overall planting conditions. Forecasts for the 2024–25 season indicated a sharp rise in disease severity under late planting, potentially reaching 100% within 28&#xa0;days after first appearance, highlighting the importance of control measures before 21&#xa0;days. FADAP prediction using the GM(1,1) grey model yielded the best performance for early planting (RMSE = 6.27; MAPE = 7.81%). Overall, the results emphasize the dominant role of temperature in driving PLB dynamics and demonstrate the effectiveness of combining statistical and machine learning approaches for accurate, condition-specific forecasting. Such integrated models can support timely, data-driven disease management strategies and enhance resilience in potato production systems.</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

Weather-based Forecasting of Potato Late Blight Disease Initiation and Its Severity Using Statistical and Machine Learning Models under Different Planting Conditions in West Bengal, India

  • 期刊论文

  • 发表日期：

    2026年6月15日

查看全文
  • Debraj Mitra,
  • Pradip Basak,
  • Deb Sankar Gupta,
  • Satyajit Hembram,
  • Litan Das

摘要

Potato late blight (PLB), caused by Phytophthora infestans, remains a major constraint to potato cultivation in West Bengal, India. This study examined the influence of key weather variables on PLB severity under early, normal, late, and overall planting conditions, and developed predictive models for disease severity (Percentage Disease Incidence, PDI) and disease initiation (First Appearance of potato late blight in days after planting, FADAP). Correlation analysis revealed minimum temperature as the most critical factor, exhibiting strong positive associations with PDI across all planting conditions (r up to 0.68), followed by maximum temperature and sunshine hours. Variable importance rankings from the Random Forest model consistently identified minimum temperature as the primary predictor, while maximum temperature or sunshine hours emerged as secondary determinants depending on planting time. Five modeling approaches, viz., ARIMA, ARIMAX, Support Vector Regression (SVR), Random Forest, and Time-Delay Neural Network with exogenous variables (TDNNx), were compared for PDI prediction. SVR achieved the highest accuracy for early planting, whereas TDNNx provided superior performance for normal, late, and overall planting conditions. Forecasts for the 2024–25 season indicated a sharp rise in disease severity under late planting, potentially reaching 100% within 28 days after first appearance, highlighting the importance of control measures before 21 days. FADAP prediction using the GM(1,1) grey model yielded the best performance for early planting (RMSE = 6.27; MAPE = 7.81%). Overall, the results emphasize the dominant role of temperature in driving PLB dynamics and demonstrate the effectiveness of combining statistical and machine learning approaches for accurate, condition-specific forecasting. Such integrated models can support timely, data-driven disease management strategies and enhance resilience in potato production systems.