Scholarlink

<p>Understanding long-term rainfall variability is critical for hydrological planning and disaster mitigation, particularly in monsoon-dependent regions like Hyderabad, India. This study utilizes India Meteorological Department gridded data (1981–2023) to analyze seasonal rainfall trends, frequency-domain characteristics, anomalous precipitation events, regime transition pathways, extreme rainfall probabilities, lag-based forecasting accuracy, and clustering-based seasonal regimes. Using Fourier transform analysis, dominant low-frequency magnitudes were detected, confirming seasonal signal stability and intensity across South West Monsoon (SWM), North East Monsoon (NEM), Hot Weather Period, and Cold Weather Period. Anomaly detection methods highlighted extreme precipitation events occurring in 1988, 1996, and 2020, aligning with atmospheric disturbances and ENSO impacts. Statistical probability estimations revealed the highest likelihood of extreme rainfall (&gt; 200&#xa0;mm) during SWM (62.79%), while NEM exhibited greater variability for rainfall exceeding 300&#xa0;mm (32.55%). Lag-based forecasting demonstrated superior accuracy using LSTM models with a 7-day history, improving RMSE by 18%, while clustering methods identified distinct low, moderate, and extreme rainfall regimes within seasonal classifications. Advanced regime metrics such as the Rainfall Regime Acceleration Index and Multi-Year Regime Momentum Grid further revealed intra-seasonal volatility and persistence patterns. These findings contribute to regional hydrological planning and flood risk mitigation strategies, emphasizing the importance of sequence-aware, data-driven forecasting in climate variability assessment.</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

Decoding monsoon dynamics: machine learning and regime analytics for seasonal rainfall in Hyderabad

  • 期刊论文

  • 发表日期：

    2026年2月14日

查看全文
  • V. Guhan,
  • A. Dharma Raju,
  • K. Nagaratna

摘要

Understanding long-term rainfall variability is critical for hydrological planning and disaster mitigation, particularly in monsoon-dependent regions like Hyderabad, India. This study utilizes India Meteorological Department gridded data (1981–2023) to analyze seasonal rainfall trends, frequency-domain characteristics, anomalous precipitation events, regime transition pathways, extreme rainfall probabilities, lag-based forecasting accuracy, and clustering-based seasonal regimes. Using Fourier transform analysis, dominant low-frequency magnitudes were detected, confirming seasonal signal stability and intensity across South West Monsoon (SWM), North East Monsoon (NEM), Hot Weather Period, and Cold Weather Period. Anomaly detection methods highlighted extreme precipitation events occurring in 1988, 1996, and 2020, aligning with atmospheric disturbances and ENSO impacts. Statistical probability estimations revealed the highest likelihood of extreme rainfall (> 200 mm) during SWM (62.79%), while NEM exhibited greater variability for rainfall exceeding 300 mm (32.55%). Lag-based forecasting demonstrated superior accuracy using LSTM models with a 7-day history, improving RMSE by 18%, while clustering methods identified distinct low, moderate, and extreme rainfall regimes within seasonal classifications. Advanced regime metrics such as the Rainfall Regime Acceleration Index and Multi-Year Regime Momentum Grid further revealed intra-seasonal volatility and persistence patterns. These findings contribute to regional hydrological planning and flood risk mitigation strategies, emphasizing the importance of sequence-aware, data-driven forecasting in climate variability assessment.