Scholarlink

<p>Accurate rain rate estimation remains a longstanding challenge in atmospheric science, with significant implications for disaster management and agricultural planning. In this work, we present a deep learning-based approach utilizing a conditional Generative Adversarial Network (cGAN) to estimate rain rates from satellite-derived Outgoing Longwave Radiation (OLR) data in the region from latitudes 0◦N to 40◦N and longitudes 60◦E to 100◦E, i.e. the Indian subcontinent and its surrounding regions. Quantitative evaluation across multiple rain rate thresholds over a five-year period demonstrates the competitiveness of our method compared to traditional algorithms such as Hydro-Estimator (HE) and INSAT Multispectral Rainfall Algorithm (IMSRA), particularly in detecting moderate to heavy rain events. The threat scores for the proposed method range from 0.377 at 0.5&#xa0;mm/hr to 0.021 at 20&#xa0;mm/hr, compared to 0.265 and 0.024 for HE and 0.270 and 0.023 for IMSRA. Thus, the proposed method results in substantial improvements of more than 35% at lower rain rates relative to operational algorithms but performs slightly worse at the highest rain rate thresholds.</p> Graphical abstract <p></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

Evaluation of a Conditional Generative Adversarial Network Model for Retrieval of Instantaneous Rain Rates from INSAT-3D Outgoing Longwave Radiation Observations

  • 期刊论文

  • 发表日期：

    2026年5月21日

查看全文
  • Atharva Deshpande,
  • Kaushik Gopalan

摘要

Accurate rain rate estimation remains a longstanding challenge in atmospheric science, with significant implications for disaster management and agricultural planning. In this work, we present a deep learning-based approach utilizing a conditional Generative Adversarial Network (cGAN) to estimate rain rates from satellite-derived Outgoing Longwave Radiation (OLR) data in the region from latitudes 0◦N to 40◦N and longitudes 60◦E to 100◦E, i.e. the Indian subcontinent and its surrounding regions. Quantitative evaluation across multiple rain rate thresholds over a five-year period demonstrates the competitiveness of our method compared to traditional algorithms such as Hydro-Estimator (HE) and INSAT Multispectral Rainfall Algorithm (IMSRA), particularly in detecting moderate to heavy rain events. The threat scores for the proposed method range from 0.377 at 0.5 mm/hr to 0.021 at 20 mm/hr, compared to 0.265 and 0.024 for HE and 0.270 and 0.023 for IMSRA. Thus, the proposed method results in substantial improvements of more than 35% at lower rain rates relative to operational algorithms but performs slightly worse at the highest rain rate thresholds.

Graphical abstract