A 30 m annual land cover dataset of Chinese Loess Plateau from 1990 to 2020

Due to current published land cover products cannot accurately captured authentic vegetation type dynamic change induced by ecological restoration projects on Chinese Loess Plateau, the annual land cover of Chinese Loess Plateau from 1990 to 2020 was produced by Yellow River Conservancy Commission. For the non-terrestrial areas, water bodies and snow/ice are directly from CLCD product (Yang and Huang, 2021). For the terrestrial areas, training samples were firstly collected by combining pixel samples with stable land cover type extracted from different land cover products at 30m scale, and visually-interpreted samples from time-series Landsat NDVI, cumulative high resolution satellite imagery, Google Earth and prerequisite knowledge. Change detection algorithm was then developed to detect the change time of forest and grassland, cropland and barren land and classify these three land covers before and after change based on interannual and intra-annual NDVI change characteristics derived from all available Landsat time-series stacks. Consequently, Random forest classifier was employed to obtain vegetation types in the domain of forest and grassland based on spectral, texture and temporal features derived from time-series Landsat imagery. After validation using field survey data and comparison with other products, this product has better performance to demonstrate significant change of cropland and forest and grassland induced by ecological restoration projects on the Loess Plateau.

鉴于现有公开的土地覆盖产品无法精准捕捉中国黄土高原生态修复工程引发的真实植被类型动态变化，黄河水利委员会制作了1990年至2020年中国黄土高原逐年土地覆盖数据集。对于非陆地区域，水体与积雪/冰数据直接取自CLCD产品（CLCD product）。对于陆地区域，训练样本首先通过两类方式采集：一是结合30米分辨率下从多套土地覆盖产品中提取的稳定土地覆盖类型得到的像素样本，二是基于时序Landsat归一化植被指数（Normalized Difference Vegetation Index, NDVI）、累积高分辨率卫星影像、Google Earth及先验知识获取的目视解译样本。随后研发变化检测算法，基于所有可用Landsat时序影像栈提取的年际与年内NDVI变化特征，识别森林草地、耕地与裸地的变化时点，并对变化前后的该三类土地覆盖类型进行分类。在此基础上，针对森林与草地范畴内的植被类型，借助时序Landsat影像提取的光谱、纹理及时间特征，采用随机森林分类器完成植被类型提取。经野外调查数据验证并与其他同类产品对比后，该数据集能够更精准地展现黄土高原生态修复工程引发的耕地、森林与草地的显著变化。