Multimodal Dataset of 583 Traditional Villages for Spatial Morphology Analysis in Jiangxi Province, China

This dataset accompanies the study “Multi-Modal Feature Fusion for Spatial Morphology Analysis of Traditional Villages via Hierarchical Graph Neural Networks.” It contains multimodal data for 583 traditional villages in Jiangxi Province, China, and was constructed to support the quantitative analysis and classification of traditional village spatial morphology. Each village is represented by three major types of input data: Image data, including village texture maps, satellite images, and topographic maps; Textual data, including brief descriptions of village history, culture, spatial characteristics, and local context; Structured multi-source variables, covering humanities, geography, and socioeconomic conditions. The structured variables include folklore, architectural characteristics, clan information, village history, notable figures, administrative divisions, geographic coordinates, altitude, slope, water-network density, distance to the nearest waterbody, precipitation, sunshine duration, temperature, proportion of arable land, normalized difference vegetation index (NDVI), population density, nighttime light intensity, urbanization rate, points of interest (POIs), gross domestic product (GDP), road length, distance to ancient stage roads, village texture dispersion, and village texture density. The dataset also provides ground-truth labels for 17 spatial morphology classification tasks, organized into three groups: Settlement landscape spatial structure (S1–S6); Settlement and parcel morphology patterns (V1–V5); Settlement road network patterns (R1–R6). These labels describe relationships among villages, mountains, water systems, agricultural fields, settlement boundaries, parcels, and road networks. They were established according to existing theories and classification principles of traditional village spatial morphology. The dataset can be used for multimodal data fusion, graph neural network modeling, spatial morphology classification, rural settlement analysis, interpretable machine learning, and traditional village conservation research. In the associated study, the dataset was divided into an 80% training set and a 20% testing set to develop and evaluate a hierarchical graph neural network framework.