Reunion island - 2018, Land cover map (Pleiades) - 0.5m

CIRAD's TETIS research unit is developing an automated mapping method based on the Moringa chain that minimizes interactions with users by automating most image analysis and processing. The methodology uses jointly a Very High Spatial Resolution image (Spot6/7 or Pleiades) and one or more time series of High Spatial Resolution optical images such as Sentinel-2 and Landsat-8 for a classification combining segmentation and object classification (use of the Random Forest algorithm) driven by a learning database constituted from in situ collection and photo-interpretation. The land use maps are produced as part of the GABIR project (Gestion Agricole des Biomasses à l'échelle de l'Ile de la Réunion) and are downloadable below or on CIRAD's spatial data catalogue: https://geode.cirad.fr This Dataverse entry concerns the maps produced, for the year 2018, using a mosaic of Pleiades images to calculate segmentation (extraction of homogeneous objects from the image). We use a field database with a nested nomenclature with 3 levels of accuracy allowing us to produce a classification by level. The most detailed level 3 distinguishing crop types has an overall accuracy of 87% and a Kappa index of 0.85. Level 2, distinguishing crop groups, has an overall accuracy of 92% and a Kappa index of 0.90. Level 1, distinguishing major land use groups, has an overall accuracy of 97% and a Kappa index of 0.95. A detailed sheet presenting the validation method and results is available for download.

法国国际农业发展研究合作中心（Centre de coopération internationale en recherche agronomique pour le développement，简称CIRAD）下属的TETIS研究团队正基于Moringa处理链（Moringa chain）开发一套自动化制图方法，该方法通过自动化绝大多数影像分析与处理流程，最大限度降低了用户交互需求。该方法联合使用超高空间分辨率影像（Very High Spatial Resolution image，即Spot 6/7卫星或Pleiades卫星影像）以及Sentinel-2、Landsat-8等单时序或多时序高空间分辨率光学影像，开展结合影像分割与对象分类的分类任务（采用随机森林算法（Random Forest algorithm）），并由基于野外原位采集（in situ collection）与目视解译（photo-interpretation）构建的学习数据库驱动。本套土地利用制图产品作为GABIR项目（留尼汪岛尺度生物质农业管理项目，Gestion Agricole des Biomasses à l'échelle de l'Ile de la Réunion）的产出成果，可通过下方链接或CIRAD空间数据目录（https://geode.cirad.fr）下载获取。本Dataverse入库条目（Dataverse entry）对应2018年基于Pleiades影像镶嵌图完成分割计算、提取影像同质对象所生成的制图产品。本次研究采用具备三级精度嵌套命名体系（nested nomenclature）的野外调查数据库，支持按精度层级开展分类制图。其中区分作物类型的最精细三级分类，总体精度（overall accuracy）达87%，Kappa系数（Kappa index）为0.85；区分作物类群的二级分类总体精度达92%，Kappa系数为0.90；区分主要土地利用类群的一级分类总体精度达97%，Kappa系数为0.95。另有一份详细说明文档，涵盖验证方法与结果，可供下载查阅。

搜集汇总

背景与挑战

背景概述

该数据集是留尼汪岛2018年的土地覆盖图，基于Pleiades影像（0.5米分辨率）生成，采用自动化分类方法（随机森林算法）提供三级分类精度，最高精度达97%。数据由CIRAD的TETIS研究单元开发，属于GABIR项目，用于农业生物质管理。