Scholarlink

<p>The North West Himalayan (NWH) region consists of Himachal Pradesh (HP) and Uttarakhand (UK), which received extremely heavy rainfall due to sudden cloud bursts between 12th and 16th August 2023, resulting in more than 140 lives lost. In recent years, HP and the UK have faced massive devastation due to cloud bursts, flash floods, and landslides frequently during the monsoon season, causing catastrophic death and destruction in the region. However, traditional numerical weather models (NWP) substantially underestimate the intensity and timing of these short-lived heavy rainfall events with a specific lead time for effective disaster preparedness and mitigation strategies. This work aims to fill this major gap. Here, we proposed a dual-encoder cross-attention fusion transformer with a decoder-based deep learning (DL) model to enhance cloud-burst prediction at the district scale over mountainous terrain in HP and the UK, with a lead time of up to 72&#xa0;h. With a mean absolute error of less than 9&#xa0;mm, the suggested model demonstrated superior rainfall estimation, outperforming the Weather Research and Forecasting (WRF) model. More than 6 cloud bursts occurred over the NWH region between 13 and 14 August 2023, and our proposed model captures them remarkably well at the spatio-temporal scale. The rainfall skill score, i.e., the Equitable Threat Score (ETS) and Probability of Detection (POD) average values, are 0.6 (0.1) for the DL(WRF) model. The DL model successfully captures the temporal variation (3-hourly) of rainfall for all testing districts. Across Mandi, Dehradun, Haridwar, and Pauri Garhwal, the DL model’s accuracy for heavy rainfall events is 68.4%, 67.33%, 54.66%, and 77.7%, respectively, whereas the WRF model barely predicts any events. This is landmark research with direct implications for improving early warning, disaster preparedness, and mitigation.</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

Enhancing the cloud burst prediction over mountainous terrains of northwestern Himalaya of India: a deep learning approach

  • 期刊论文

  • 发表日期：

    2026年6月1日

查看全文
  • Hemant Kumar,
  • Dhananjay Trivedi,
  • Omveer Sharma,
  • Sandeep Pattnaik,
  • Niladri Bihari Puhan

摘要

The North West Himalayan (NWH) region consists of Himachal Pradesh (HP) and Uttarakhand (UK), which received extremely heavy rainfall due to sudden cloud bursts between 12th and 16th August 2023, resulting in more than 140 lives lost. In recent years, HP and the UK have faced massive devastation due to cloud bursts, flash floods, and landslides frequently during the monsoon season, causing catastrophic death and destruction in the region. However, traditional numerical weather models (NWP) substantially underestimate the intensity and timing of these short-lived heavy rainfall events with a specific lead time for effective disaster preparedness and mitigation strategies. This work aims to fill this major gap. Here, we proposed a dual-encoder cross-attention fusion transformer with a decoder-based deep learning (DL) model to enhance cloud-burst prediction at the district scale over mountainous terrain in HP and the UK, with a lead time of up to 72 h. With a mean absolute error of less than 9 mm, the suggested model demonstrated superior rainfall estimation, outperforming the Weather Research and Forecasting (WRF) model. More than 6 cloud bursts occurred over the NWH region between 13 and 14 August 2023, and our proposed model captures them remarkably well at the spatio-temporal scale. The rainfall skill score, i.e., the Equitable Threat Score (ETS) and Probability of Detection (POD) average values, are 0.6 (0.1) for the DL(WRF) model. The DL model successfully captures the temporal variation (3-hourly) of rainfall for all testing districts. Across Mandi, Dehradun, Haridwar, and Pauri Garhwal, the DL model’s accuracy for heavy rainfall events is 68.4%, 67.33%, 54.66%, and 77.7%, respectively, whereas the WRF model barely predicts any events. This is landmark research with direct implications for improving early warning, disaster preparedness, and mitigation.