Interpreting the Spatial Characteristics of the Dike-Pond System through Deep Learning and Digital Mapping Techniques: A Case Study of Foshan Sangyuanwei
Keywords: Dike-Pond System, Spatial Characteristics, Digital Mapping Techniques, Deep Learning, Agricultural Heritage
Abstract. The Dike-Pond System in China’s Pearl River Delta is a distinctive form of agricultural heritage, renowned for its integrated land-water production, ecological adaptability, and embedded cultural practices. Despite growing recognition of its heritage value, there remains a lack of a spatially grounded framework capable of decoding its internal structure and landscape heterogeneity. This study develops an intersubjective approach to identify, quantify, and interpret the spatial characteristics of the Dike-Pond System from a landscape perspective. Taking Sangyuanwei in Foshan as a case study, the research first extracts four core landscape characters, production and livelihood, ecological networks, water management, and transportation connectivity, through systematic literature review. A deep learning model, trained on high-resolution satellite imagery, was employed to detect pond morphologies and, together with hydrological, infrastructural, and land-use data, construct a comprehensive spatial database. Spatial indicators were then computed and visualized using digital mapping and geostatistical techniques, supporting the classification of five distinct landscape types. These typologies reflect the system’s coexisting patterns of resilience and transformation, offering insights into its spatial logic under urban-rural integration. The framework bridges qualitative interpretation and quantitative analysis, providing a replicable method for spatially grounded heritage evaluation and landscape-informed planning.