Fine-Scale Small Water Body Uncovered by GF-2 Remote Sensing and Multifeature Deep Learning Model

Small water bodies are essential for hydrological connectivity, efficient use of water resources, and ecological conservation. Existing water body extraction models face issues such as detail loss, misclassification, and incomplete coverage. The lack of fine-scale remote sensing imagery and well-performing models has hindered the ability to track small water body dynamics, especially in heterogeneous urban areas. To address this, the Multi-Feature Joint Perception Convolutional Network (MSFCN) is proposed for the fine-scale extraction of small water bodies using GF-2 satellite data. Spatial characteristics of small water bodies in different urban zones and their relationship with overall urban water resources are then analyzed. Results show that MSFCN performs well with an overall accuracy of 0.91 in extracting small water bodies. Spatial analysis indicates that small water bodies are mainly found in suburban areas (80%), followed by the expansion zone (14%) and the city core (6%). In the city core, small water bodies are predominantly artificial, often linked to construction activities, while in suburban areas, they are mainly agricultural, serving irrigation purposes. The analysis also reveals a negative correlation between small and large water bodies in all urban areas, suggesting the ecological sensitivity of these water bodies and their crucial role in maintaining hydrological balance. These findings provide important technical support for large-scale extraction of small water bodies and offer valuable insights for urban planning and ecological management.