Land cover–wetness combination maps

What the data product is This data product is the end product of a workflow of decision rules to inform a classification problem for the combined land cover and wetness of peatlands and mineral wetlands across ten European catchment areas. It used high-resolution land cover data products in combination with the water table depth model developed in this project. The 10 m raster products from ESA WorldCover (2021), the Copernicus Land Service Monitoring Programme CLCplus Backbone (2023) and Google Dynamic World (GDW; March-September modal composite, 2023) were selected, and their land cover classes harmonised as follows: each source contributes a per-pixel score to each harmonised class – GDW and ESA weighted by per-class accuracy derived from their respective confusion matrices, and CLCplus weighted by its per-pixel confidence value so that lower-confidence CLC assignments contributed less to the final decision. The class with the highest combined score was assigned as the output land cover. Pixels where no class achieved a positive score were assigned as no data. For the Scottish catchments, where the balancing contributions of the CLCPlus Backbone data set were not available for download, we replaced the CLCPlus data with the UKCEH Land Cover Map product, combining it with GDW and ESA in the same manner. This decision tree resulted in a single output land cover class per 10 m pixel. In cases where no agreement could be found, an “uncertain” class assignment was used. This was combined with wetness levels based on reducing the complexity of the data product in https://zenodo.org/records/16753826. We used a functional classification of “wet” = mean annual predicted water level between 0 and -0.2 metres; for “shallow drained” we used thresholds of -0.2 to -0.5 metres; and for “deep-drained” areas we used -0.5 or deeper, to allow for the uncertainties of ~20 cm in the water table predictions. Each single land cover class from the decision rule set was therefore subdivided into three wetness classes. A lookup table (lookup_combo.csv) is provided to decode the combined class values in each output raster. What it is useful for At present, the condition or land cover of wetland and peatland areas, as defined by the European Wetland Map across several European catchment areas, can only be defined by either global land cover data products, which tend to have a very limited number of assigned classes; CORINE, which has limited temporal steps; or higher resolution national data that do not use cross-compatible classification systems. This data product formed a first attempt at producing a harmonisation across European wetland and peatland condition classification and will be used for a scientific publication that compares the utility of the underlying parent products. As such, this is a purely academic exercise with limited immediate use cases. Other uses in derived products should take note of the caveats. Important caveats and use notes Although we tested whether the predicted classes matched available ground observations from the EVA database and our own compilation for Wet Horizons, Deliverable D1.4, this was more of a proof of principle, as there was insufficient data available from the same time period as the Earth Observation data. In addition, only a very low number of the available vegetation community observations fell within some of the predicted decision-rule combination categories. In addition, the modelling effort produced some incongruent output classes, such as 'water (deep drained)', meaning that the land cover product combination suggested water as the most likely land cover, but the water table depth model suggested a mean annual water table depth below 0.5 metres. We therefore suggest caution in further use as the model output is likely to be of only moderate predictive quality. Contacts for further queries or feedback on this product For further information or feedback, please contact the author team, either directly or via [e-mail].数据产品概况 本数据产品为一套决策规则工作流的最终成果，用于支撑欧洲10个流域内泥炭地与矿质湿地的土地覆被及湿润度联合分类任务。本产品结合了高分辨率土地覆被数据集与本项目自研的地下水位深度模型。 本次选用了欧空局世界覆被（ESA WorldCover，2021）、哥白尼陆地服务监测计划CLCplus骨干数据集（Copernicus Land Service Monitoring Programme CLCplus Backbone，2023）以及谷歌动态世界（Google Dynamic World，简称GDW；2023年3-9月模态合成数据）的10米分辨率栅格（raster）产品，并对其土地覆被类别进行统一化处理，流程如下：各源数据集为每个统一化类别提供逐像素得分——GDW与ESA数据集基于各自混淆矩阵（confusion matrix）导出的类别级准确率进行加权，CLCplus则基于逐像素置信度值加权，以此让低置信度的CLC类别分配对最终决策的影响更小。将综合得分最高的类别作为输出土地覆被类别；若某像素无任何类别获得正得分，则将其标记为无数据（no data）。 针对苏格兰流域，由于无法下载CLCplus骨干数据集的均衡贡献数据，我们采用英国生态与水文中心土地覆被图（UKCEH Land Cover Map）产品替代CLCplus数据，并以相同方式与GDW、ESA数据集进行融合。该决策树可为每个10米像素输出唯一的土地覆被类别；若无法达成一致判定，则将其标记为"uncertain"（不确定）类别。 本产品结合了来自https://zenodo.org/records/16753826的简化版数据产品所导出的湿润度等级：我们采用如下功能分类规则——"wet（湿润）"对应年平均预测水位介于0至-0.2米之间；"shallow drained（浅排）"对应水位介于-0.2至-0.5米之间；"deep-drained（深排）"对应水位≤-0.5米，以此预留地下水位预测中约20厘米的不确定性空间。因此，决策规则集输出的每个单一土地覆被类别均被细分为3个湿润度子类别。本次提供了查找表（lookup table，lookup_combo.csv），用于解码各输出栅格中的组合类别值。 产品应用场景 当前，欧洲湿地地图（European Wetland Map）所定义的多个欧洲流域内的湿地与泥炭地状况或土地覆被，仅能通过以下三类数据进行界定：一是类别数量极为有限的全球土地覆被数据集；二是时间分辨率不足的CORINE土地覆被数据；三是未采用跨兼容分类体系的高分辨率国家级数据。本数据产品首次尝试实现欧洲湿地与泥炭地状况分类的统一化处理，将用于一篇对比各原始父数据集效用的学术论文。因此，本产品仅为纯学术探索，当前可用场景有限。基于本产品衍生的各类应用需注意其附带的限制说明。 重要限制与使用说明 尽管我们验证了预测类别与EVA数据库以及"Wet Horizons"项目可获取的地面观测数据（交付件D1.4）是否匹配，但该验证仅为原理性证明——由于缺乏与地球观测数据同期的足量地面观测数据。此外，仅有极少部分现有植被群落观测数据可归入部分预测得到的决策规则组合类别中。同时，建模过程中生成了部分矛盾的输出类别，例如"water (deep drained)（水体（深排））"：即土地覆被融合结果判定水体为最可能的土地类型，但地下水位深度模型却显示年平均地下水位低于0.5米。因此，我们建议后续使用时保持谨慎，本模型输出的预测质量仅为中等水平。 产品咨询与反馈联系方式 如需获取更多信息或反馈意见，请直接联系作者团队，或通过邮箱[e-mail]联络。