Scholarlink

<p>Sandy beaches are important coastal features with ecological, social, and economic significance, yet comprehensive national-scale datasets mapping their extent in India remain scarce. Here, we present a high-resolution, geo-referenced dataset of sandy beach extents along the Indian coastline, derived from IRS ResourceSat-2/2 A LISS-IV multispectral imagery (5.8 m resolution) collected between 2021 and 2024. The dataset was derived from cloud-free imagery acquired under low-tide conditions to ensure consistent delineation of exposed sandy beach extents. Fourteen representative coastal sites were manually annotated to train and validate a U-Net deep learning model, which was subsequently applied to the entire Indian coastline. The model achieved robust performance on independent test tiles, with an intersection-over-union of 0.84, precision of 0.90, recall of 0.94, and F1-score of 0.92. Accuracy was further validated at two independent beaches with low tidal influence, yielding an RMSE of 11.8 m<sup>2</sup> (&lt;0.05% relative deviation) between U-Net–derived and manually digitized areas. The final dataset is distributed in Shapefile format to ensure compatibility with standard GIS workflows, supporting applications such as coastal monitoring, habitat assessment, and shoreline change analysis, as well as serving as training data for future machine learning models. An accompanying Google Earth Engine web application is provided for interactive visualization and exploration of the mapped sandy beaches. This dataset represents the first comprehensive, high-resolution mapping of sandy beaches across India, enabling improved coastal management, conservation planning, and research into coastal dynamics.</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 national-scale sandy beach dataset for India derived from high-resolution satellite imagery and deep learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月9日

查看全文
  • Giribabu Dandabathula,
  • Subham Roy,
  • Omkar Shashikant Ghatage,
  • Sagar Subhashrao Salunkhe,
  • Durgesh Nandani,
  • Apurba Kumar Bera,
  • Sushil Kumar Srivastav

摘要

Sandy beaches are important coastal features with ecological, social, and economic significance, yet comprehensive national-scale datasets mapping their extent in India remain scarce. Here, we present a high-resolution, geo-referenced dataset of sandy beach extents along the Indian coastline, derived from IRS ResourceSat-2/2 A LISS-IV multispectral imagery (5.8 m resolution) collected between 2021 and 2024. The dataset was derived from cloud-free imagery acquired under low-tide conditions to ensure consistent delineation of exposed sandy beach extents. Fourteen representative coastal sites were manually annotated to train and validate a U-Net deep learning model, which was subsequently applied to the entire Indian coastline. The model achieved robust performance on independent test tiles, with an intersection-over-union of 0.84, precision of 0.90, recall of 0.94, and F1-score of 0.92. Accuracy was further validated at two independent beaches with low tidal influence, yielding an RMSE of 11.8 m2 (<0.05% relative deviation) between U-Net–derived and manually digitized areas. The final dataset is distributed in Shapefile format to ensure compatibility with standard GIS workflows, supporting applications such as coastal monitoring, habitat assessment, and shoreline change analysis, as well as serving as training data for future machine learning models. An accompanying Google Earth Engine web application is provided for interactive visualization and exploration of the mapped sandy beaches. This dataset represents the first comprehensive, high-resolution mapping of sandy beaches across India, enabling improved coastal management, conservation planning, and research into coastal dynamics.