RiverScope
收藏资源简介:
RiverScope是一个高分辨率河流掩模数据集,由计算机科学和水文学专家合作开发。该数据集包含1145张高分辨率图像(覆盖2577平方公里),由专家手动标注河流和地表水掩模,耗时超过100小时。每张图像都与Sentinel-2、SWOT和SWOT河流数据库(SWORD)进行配准,从而能够在传感器之间评估成本-精度权衡,这是对操作水资源监测的关键考虑。该数据集还为河流宽度估计建立了第一个全球性高分辨率基准,实现了7.2米的平均误差,显著优于现有的卫星衍生方法。RiverScope提供了宝贵的资源,用于细粒度和多传感器水文建模,支持气候适应和可持续水资源管理。
RiverScope is a high-resolution river mask dataset developed collaboratively by experts in computer science and hydrology. This dataset comprises 1,145 high-resolution images covering a total area of 2,577 km², with river and surface water masks manually annotated by experts over 100 cumulative hours of work. Each image is co-registered with Sentinel-2, SWOT, and the SWOT River Database (SWORD), enabling the evaluation of cost-accuracy tradeoffs across sensors—a critical consideration for operational water resources monitoring. This dataset also establishes the first global high-resolution benchmark for river width estimation, achieving an average error of 7.2 meters and significantly outperforming existing satellite-derived methods. RiverScope provides a valuable resource for fine-grained and multi-sensor hydrological modeling, supporting climate adaptation and sustainable water resources management.
RiverScope: 高分辨率河流掩码数据集
数据集概述
RiverScope是一个高分辨率河流掩码数据集,专注于河流水域分割和河流宽度估算任务。
数据集获取
- 数据集下载地址:https://github.com/cvl-umass/riverscope
- 数据集发布年份:2025年
- 数据集发布机构:Zenodo
- DOI:10.5281/zenodo.15376394
模型架构
数据集支持多种分割模型架构:
- DeepLabv3
- DPT
- FPN
- UNet
骨干网络
支持多种骨干网络结构:
- MobileNet_v2
- ResNet50
- Swin-B
- Swin-T
- ViT-B/16
- ViT-L/16
预训练权重
提供多种预训练配置:
- ImageNet1k预训练
- MoCoV3预训练
- SeCo预训练
- CLIP预训练
- DINO预训练
- Prithvi预训练
- SatlasPretrain预训练
性能指标
水域分割性能
最佳F1分数模型配置:
- UNet + ResNet50 + SeCo预训练:F1=94.39
- UNet + ResNet50 + MoCoV3预训练:F1=94.20
- DeepLabv3 + ResNet50 + MoCoV3预训练:F1=93.92
宽度估算性能
最佳宽度估算模型配置:
- UNet + ResNet50 + SeCo预训练:中值误差=7.50米,平均误差=16.26米
- FPN + ResNet50 + SeCo预训练:中值误差=8.62米,平均误差=15.36米
- UNet + ResNet50 + MoCoV3预训练:中值误差=8.45米,平均误差=18.06米
预训练模型
提供33个预训练模型检查点,涵盖不同模型架构、骨干网络和预训练配置组合,每个模型都提供了最优阈值参数。
评估方法
- 水域分割评估:支持自动阈值搜索和手动阈值设置
- 河流宽度估算:基于分割结果进行河流宽度估算
- 测试集包含约400个节点的标注数据
基准模型
提供Sentinel-2训练的基准模型,使用SatlasPretrain数据集进行训练。
引用信息
bibtex @dataset{daroya_2025_15376394, author = {Daroya, Rangel and Rowley, Taylor and Flores, Jonathan and Friedmann, Elisa and Bennitt, Fiona and An, Heejin and Simmons, Travis and Hughes, Marissa Jean and Kluetmeier, Camryn L and Kica, Solomon and Vélez, J. Daniel and Esenther, Sarah E. and Howard, Thomas E. and Ye, Yanqi and Turcotte, Audrey and Gleason, Colin and Maji, Subhransu}, title = {RiverScope: High-Resolution River Masking Dataset}, month = may, year = 2025, publisher = {Zenodo}, doi = {10.5281/zenodo.15376394}, url = {https://doi.org/10.5281/zenodo.15376394}, }

- 1RiverScope: High-Resolution River Masking DatasetUMass Amherst, UNC Chapel Hill, Brown University, CU Boulder · 2025年



