Scholarlink

<p>The increasing industrial and economic activities due to urbanization make the populous metropolitan areas susceptible to a plethora of environmental problems, which can be catastrophic when compounded with high-impact weather events like heat waves and floods. This is an ordeal for the citizens of the urban regions and requires a robust methodology for weather nowcasting. This paper focuses on examining a rainfall-laden region of the Indian Subcontinent, Mumbai, and developing an Artificial Intelligence-based model to predict the cloud morphology corresponding to the heavy rainfall pockets that can trigger severe urban flooding. To achieve this goal, we propose CloudMorphGAN, a CNN-GAN capable of processing geospatial data as well as learning the temporal behavior of the cloud morphology. In order to verify its effectiveness, the CloudMorphGAN is compared with a Numerical Weather Prediction model and a popular deep learning model on three metrics evaluating different aspects of the prediction, which are crucial to our objective of cloud morphological and cloud density prediction. This systematic implementation and performance evaluation using data captured by&#xa0;the Doppler Weather Radar of India Meteorological Department shows promising results by CloudMorphGAN, with mean absolute error ranging between 2.16 and 2.36 dBZ, structural similarity index metric between 91.33% and 93.89% and universal image quality from 88.54% to 90.45%. This experimentation and analysis indicate the contribution of CloudMorphGAN for nowcasting and issuing warnings of heavy rainfall and floods.</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

A Radar Data-Driven AI Approach for Rainfall Nowcasting: Towards Flood Preparedness in Urban Regions

  • 期刊论文

  • 发表日期：

    2025年2月15日

查看全文
  • Sharvil Dandekar,
  • Taksha Limbashia,
  • Om Parab,
  • Radhika Kotecha,
  • Kaustav Chakravarty,
  • Suresh Ukarande,
  • Krishnanand Hosalikar

摘要

The increasing industrial and economic activities due to urbanization make the populous metropolitan areas susceptible to a plethora of environmental problems, which can be catastrophic when compounded with high-impact weather events like heat waves and floods. This is an ordeal for the citizens of the urban regions and requires a robust methodology for weather nowcasting. This paper focuses on examining a rainfall-laden region of the Indian Subcontinent, Mumbai, and developing an Artificial Intelligence-based model to predict the cloud morphology corresponding to the heavy rainfall pockets that can trigger severe urban flooding. To achieve this goal, we propose CloudMorphGAN, a CNN-GAN capable of processing geospatial data as well as learning the temporal behavior of the cloud morphology. In order to verify its effectiveness, the CloudMorphGAN is compared with a Numerical Weather Prediction model and a popular deep learning model on three metrics evaluating different aspects of the prediction, which are crucial to our objective of cloud morphological and cloud density prediction. This systematic implementation and performance evaluation using data captured by the Doppler Weather Radar of India Meteorological Department shows promising results by CloudMorphGAN, with mean absolute error ranging between 2.16 and 2.36 dBZ, structural similarity index metric between 91.33% and 93.89% and universal image quality from 88.54% to 90.45%. This experimentation and analysis indicate the contribution of CloudMorphGAN for nowcasting and issuing warnings of heavy rainfall and floods.