Scholarlink

<p>Rainfall variability and extremes pose major challenges to agriculture and water management in semi-arid regions. This study investigates rainfall dynamics across twelve districts of Southern Telangana using India Meteorological Department (IMD) daily rainfall data for 1981–2023. The dataset contained less than 2% missing values, which were interpolated using nearest-neighbor methods, ensuring reliability for subsequent diagnostics. A diagnostic framework was applied, integrating climatological statistics, extreme value theory, persistence analysis, entropy measures, synchrony diagnostics, and forecast skill evaluation with machine learning explainability. Results reveal pronounced heterogeneity. Annual rainfall ranges from ~ 670&#xa0;mm in Jogulamba Gadwal to ~ 1,288&#xa0;mm in Mahabubnagar, with coefficients of variation between 18 and 32%. Extreme rainfall analysis using Peaks Over Threshold–Generalized Pareto Distribution (POT–GPD) indicates district-wise return levels of ~ 160–215&#xa0;mm (RL10), ~ 190–245&#xa0;mm (RL25), and ~ 210–275&#xa0;mm (RL50), with confidence intervals reported. RL10 is emphasized for short-term planning relevance (e.g., irrigation scheduling, crop insurance), while RL25 and RL50 provide longer-term insights for infrastructure and water resource management. Persistence diagnostics suggest long-range dependence (indices 0.69–0.77, with confidence intervals reported), while entropy distinguishes irregular regimes such as Narayanpet (0.43) from more predictable districts like Mahabubnagar (0.22). Synchrony analysis highlights moderate coherence in Hyderabad (0.69) but weaker coupling in Mahabubnagar (0.40), consistent with negative Moran’s I values. Forecast skill metrics show median absolute errors of 12–19&#xa0;mm, prediction interval coverage above 0.87, and Continuous Ranked Probability Scores (0.081–0.112). PCA and SHAP analysis identify district-specific drivers, including JJAS rainfall, seasonality indices, entropy, and long memory properties. The integrated framework moves beyond conventional rainfall studies, delivering district-specific insights that are both rigorous and operationally relevant. Findings provide actionable guidance for drought preparedness, flood risk management, and agricultural resilience in semi-arid monsoon regions, while contributing to global efforts to understand rainfall variability under climate uncertainty.</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

Multi-dimensional diagnostics of rainfall variability, extremes, and forecast skill in Southern Telangana

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月17日

查看全文
  • Guhan Velusamy,
  • Dharma Raju Akasapu,
  • Nagaratna Kopparthi,
  • Sheshakumar Goroshi

摘要

Rainfall variability and extremes pose major challenges to agriculture and water management in semi-arid regions. This study investigates rainfall dynamics across twelve districts of Southern Telangana using India Meteorological Department (IMD) daily rainfall data for 1981–2023. The dataset contained less than 2% missing values, which were interpolated using nearest-neighbor methods, ensuring reliability for subsequent diagnostics. A diagnostic framework was applied, integrating climatological statistics, extreme value theory, persistence analysis, entropy measures, synchrony diagnostics, and forecast skill evaluation with machine learning explainability. Results reveal pronounced heterogeneity. Annual rainfall ranges from ~ 670 mm in Jogulamba Gadwal to ~ 1,288 mm in Mahabubnagar, with coefficients of variation between 18 and 32%. Extreme rainfall analysis using Peaks Over Threshold–Generalized Pareto Distribution (POT–GPD) indicates district-wise return levels of ~ 160–215 mm (RL10), ~ 190–245 mm (RL25), and ~ 210–275 mm (RL50), with confidence intervals reported. RL10 is emphasized for short-term planning relevance (e.g., irrigation scheduling, crop insurance), while RL25 and RL50 provide longer-term insights for infrastructure and water resource management. Persistence diagnostics suggest long-range dependence (indices 0.69–0.77, with confidence intervals reported), while entropy distinguishes irregular regimes such as Narayanpet (0.43) from more predictable districts like Mahabubnagar (0.22). Synchrony analysis highlights moderate coherence in Hyderabad (0.69) but weaker coupling in Mahabubnagar (0.40), consistent with negative Moran’s I values. Forecast skill metrics show median absolute errors of 12–19 mm, prediction interval coverage above 0.87, and Continuous Ranked Probability Scores (0.081–0.112). PCA and SHAP analysis identify district-specific drivers, including JJAS rainfall, seasonality indices, entropy, and long memory properties. The integrated framework moves beyond conventional rainfall studies, delivering district-specific insights that are both rigorous and operationally relevant. Findings provide actionable guidance for drought preparedness, flood risk management, and agricultural resilience in semi-arid monsoon regions, while contributing to global efforts to understand rainfall variability under climate uncertainty.