satellite-derived spatial continuous, consist time series of annual urban impervious surface, green vegetation and bare soil fractions of China

We utilized a total of 201,143 multi-temporal orthorectified images from Landsat 5, 7, and 8, spanning the past three decades, on the Google Earth Engine platform to generate an annual and continuous Vegetation–Impervious–Soil (V-I-S) fraction product for urban areas across China.A diverse set of urban landscape scenes was chosen from six representative regions covering all urban ecoclimatic zones in China, which vary by climatic conditions, land use/cover categories, urban morphology, and economic development. The selected imagery corresponds to multiple dates during the study period (including 6 TM5, 2 ETM+, and 4 OLI acquisitions). For each scene, efforts were made to use cloud-free images and perform atmospheric correction to reduce random noise from Rayleigh scattering during fraction estimation. All sub-scenes were merged into sensor-specific composite images, forming a nationally representative spectral mixing space. This composite captures both the spectral variability and nationwide consistency of urban landscapes as observed by Landsat TM, ETM+, and OLI sensors.Principal Component Analysis (PCA) was applied to assess the intrinsic dimensionality of the urban spectral mixing space based on variance distribution. After identifying endmember types and their corresponding principal components across sensors, we selected 200–400 “pure pixels” from the extremities of the pixel cloud bounded by key PCs to define candidate endmember spectra. These pixels were validated using high-resolution imagery available on Google Earth. This endmember selection process was repeated for each regional subset within the national composite, enabling the extraction of region-specific endmember spectra. Average endmember spectra were then computed by aggregating individual candidate spectra for high- and low-albedo features, green vegetation, and bare soil for each sensor. These standardized national endmember spectra served as inputs to a fully-constrained Linear Spectral Mixture Analysis (LSMA) model.As a result, we produced an annual and continuous national dataset of urban landscape components from 1990 to 2019, characterizing green vegetation (GV), impervious surface area (ISA), and bare soil (BS) coverage at a 30×30-meter sub-pixel resolution. Each urban pixel for a given year is described by the proportional coverage of ISA, GV, and BS, reflecting the urban landscape composition at the peak of the growing season. This data represents the results of a multi-year average of China. Specifically, B1-B3 correspond to:ISA, GV, BS