Scholarlink

<p>Urban flash floods are an increasing challenge in Indian cities, driven by rapid urbanization, inadequate drainage, and extreme rainfall. This study presents an integrated framework that combines IoT-based sensing, edge computing, machine learning, and computer vision for real-time flood monitoring and prediction in Guwahati, India. Ten water-sensing nodes and two base stations were deployed with solar power, UPS backup, and LoRa communication to ensure continuous operation during disaster conditions. Validation using the NASA POWER API confirmed the accuracy of sensor data. A fuzzy fusion rainfall prediction model, deployed at the edge, achieved 92.4% accuracy, outperforming Random Forest, XGBoost, CatBoost, and KNN. In parallel, a blockage detection module employing a U-Net with an EfficientNetB0 backbone achieved 91% segmentation accuracy, surpassing ResNet50 and InceptionV3. Predictions reliably identified low, moderate, and high blockage levels. The proposed IoT–AI framework offers a scalable and reliable early warning system. Future work will focus on expanding datasets, incorporating additional sensing modalities, and extending deployment to other cities to enhance urban flood resilience.</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 Scalable Edge-Integrated Architecture for Urban Flood Monitoring and Prediction

  • 期刊论文

  • 发表日期：

    2026年5月29日

查看全文
  • Rupesh Mandal,
  • Bobby Sharma,
  • Dibyajyoti Chutia,
  • Nupur Choudhury

摘要

Urban flash floods are an increasing challenge in Indian cities, driven by rapid urbanization, inadequate drainage, and extreme rainfall. This study presents an integrated framework that combines IoT-based sensing, edge computing, machine learning, and computer vision for real-time flood monitoring and prediction in Guwahati, India. Ten water-sensing nodes and two base stations were deployed with solar power, UPS backup, and LoRa communication to ensure continuous operation during disaster conditions. Validation using the NASA POWER API confirmed the accuracy of sensor data. A fuzzy fusion rainfall prediction model, deployed at the edge, achieved 92.4% accuracy, outperforming Random Forest, XGBoost, CatBoost, and KNN. In parallel, a blockage detection module employing a U-Net with an EfficientNetB0 backbone achieved 91% segmentation accuracy, surpassing ResNet50 and InceptionV3. Predictions reliably identified low, moderate, and high blockage levels. The proposed IoT–AI framework offers a scalable and reliable early warning system. Future work will focus on expanding datasets, incorporating additional sensing modalities, and extending deployment to other cities to enhance urban flood resilience.