Abstract
We present a high-resolution spatial dataset delineating individual rural household courtyards and building rooftops in a representative region of northern China, extracted from sub-meter remote sensing imagery using deep learning techniques. This dataset captures the fine-scale layout of rural homestead compounds, including courtyard boundaries and rooftop footprints, with high accuracy validated against field observations. The data were generated by applying convolutional neural network models to high-resolution aerial imagery, enabling automated identification of residential courtyards and associated structures over large areas. The resulting dataset provides unprecedented detail on rural settlement structure, which can be used to analyze food-energy-water systems and support sustainable rural development. For example, the rooftop layer can inform solar photovoltaic potential assessments and infrastructure planning, while the courtyard layer yields insights into household-level agricultural practices and land management. By filling a critical gap in micro-scale rural data, this openly available dataset offers significant value for researchers and policymakers focused on rural revitalization, resource management, and environmental planning.