Major European Wetland Types at 10-meter resolution
收藏DataCite Commons2024-07-15 更新2025-04-17 收录
下载链接:
https://erda.ku.dk/archives/c9c747a5c5c93883597a86548f644f81/published-archive.html
下载链接
链接失效反馈官方服务:
资源简介:
### Dataset Metadata Description
Dataset Title: Major European Wetland Types at 10-meter resolution
Summary:
In 2018, we utilized 10-meter resolution satellite data and machine learning to map six wetland types across Europe, achieving a 94±0.5% accuracy. The dataset includes optical data from Sentinel-2 MSI and SAR imagery from Sentinel-1, processed via Google Earth Engine. We employed the CORINE Land Cover (CLC) dataset covering 38 European countries. For a pixel-based classification, we used the XGBoost model, which effectively handles multi-class problems and noisy labels. Evaluation on 10,075 samples showed an accuracy of 94±0.5%, with a mean precision of 85% and recall of 88%. Precision ranged from 67% (inland marshes) to 96% (salines), and recall from 75% (moors & heathland) to 87% (inland marshes). F1 scores ranged from 73% to 95%.
Data Sources:
- Optical Data: Sentinel-2 MSI, Level-1C TOA reflectance, 10-meter resolution.
- SAR Imagery: Sentinel-1, 10-meter resolution.
- Land Cover Data: CORINE Land Cover (CLC) dataset, covering 38 European countries.
Ancillary Data:
- Land surface temperature from MOD11A1 V6.1 product, 1-kilometer resolution, 2000-2020.
- Total precipitation data from ERA5, 27-kilometer resolution, 2000-2020.
- Digital Elevation Model (EU-DEM 1.1), 25-meter resolution and derivatives.
- Proximity layer to significant soil types and parent materials from the European Soil Database v2.0, including dystric histosols organic parent materials from Level-3 FAO Soil Classification.
- Spatial information on terrestrial biomes from the RESOLVE Ecoregions dataset.
Machine Learning Model: XGBoost, optimized for multi-class classification and robust to noisy labels.
Performance Metrics:
- Model Accuracy: 94±0.5%
- Mean Precision: 85%
- Mean Recall: 88%
- Precision Range: 67% (inland marshes) to 96% (salines)
- Recall Range: 75% (moors & heathland) to 87% (inland marshes)
- F1-Score Range: 73% (moors & heathland) to 95% (salines)
### 数据集元数据描述
数据集标题:10米分辨率欧洲主要湿地类型数据集
摘要:
2018年,本研究借助10米分辨率卫星数据与机器学习技术,对欧洲范围内的六大湿地类型开展空间制图,最终实现了94±0.5%的分类精度。本数据集包含哨兵-2号多光谱仪器(Sentinel-2 MSI)光学数据与哨兵-1号合成孔径雷达(Sentinel-1 SAR)影像,并通过谷歌地球引擎(Google Earth Engine)完成数据处理。研究采用了覆盖38个欧洲国家的CORINE土地覆盖数据集(CORINE Land Cover, CLC)。在基于像元的分类任务中,我们使用了可有效处理多分类问题与噪声标签的XGBoost模型。基于10075个样本的评估结果显示,模型整体精度为94±0.5%,平均精确率为85%,平均召回率为88%。各类别精确率区间为67%(内陆沼泽)至96%(盐渍湿地),召回率区间为75%(荒原与石楠灌丛)至87%(内陆沼泽),F1分数区间为73%至95%。
数据来源:
- 光学数据:Sentinel-2 MSI,Level-1C顶层反射率产品,10米分辨率。
- 合成孔径雷达(SAR)影像:Sentinel-1,10米分辨率。
- 土地覆盖数据:CORINE Land Cover (CLC)数据集,覆盖38个欧洲国家。
辅助数据:
- 来自MOD11A1 V6.1产品的地表温度数据,分辨率为1千米,时间跨度为2000年至2020年。
- 来自ERA5的总降水量数据,分辨率为27千米,时间跨度为2000年至2020年。
- 数字高程模型(EU-DEM 1.1),分辨率为25米及其衍生产品。
- 来自欧洲土壤数据库v2.0(European Soil Database v2.0)的典型土壤类型与母质邻近度图层,其中包含联合国粮农组织(FAO)三级土壤分类中的dystric histosols有机母质。
- 来自RESOLVE生态区数据集(RESOLVE Ecoregions)的陆地生物群系空间信息。
机器学习模型:
针对多分类任务优化且对噪声标签具有鲁棒性的XGBoost模型。
性能指标:
- 模型整体精度:94±0.5%
- 平均精确率:85%
- 平均召回率:88%
- 精确率区间:67%(内陆沼泽)至96%(盐渍湿地)
- 召回率区间:75%(荒原与石楠灌丛)至87%(内陆沼泽)
- F1分数区间:73%(荒原与石楠灌丛)至95%(盐渍湿地)
提供机构:
University of Copenhagen
创建时间:
2024-07-15



