GF-1 Wide-Field Multispectral Multi-temporal Water Body Extraction Dataset

The data used in this study (Figure 1) were obtained from the Wide-Field Multispectral Camera (WFV) of the China High-Resolution Earth Observation System's "Gaofen-1" (GF-1) satellite, with a spatial resolution of 16 meters. The spectral range covers 450–890 nm, including four bands: blue (0.45–0.52 μm), green (0.52–0.59 μm), red (0.63–0.69 μm), and near-infrared (0.77–0.89 μm). The satellite has a revisit cycle of approximately four days, enabling efficient acquisition of large-scale surface data, making it suitable for monitoring water bodies, land use, and environmental changes. The dataset covers three typical lake regions—Poyang Lake, Yangcheng Lake, and Nam Co—with each image having a pixel size of 256×256 and containing four bands. Additionally, it includes a single-band water body annotation image of the same size, featuring the following characteristics:(1) Multi-Temporal CharacteristicsThe dataset comprehensively covers the spring, summer, autumn, and winter remote sensing images of Poyang Lake and Nam Co, which exhibit high dynamic variability, as well as Yangcheng Lake, which shows relatively low variability. Using time-series images with a 16-meter spatial resolution, Figure 2 illustrates the intra-annual dynamic changes in the proportion of water body samples in the dataset. In 2022, Poyang Lake experienced a severe drought, with most areas showing moderate to severe drought conditions from August onward. Nam Co, primarily influenced by plateau climate conditions, entered a distinct ice period in winter, with large areas of the lake covered by snow and ice, leading to a significant decline in the proportion of water body samples. As temperatures rose in spring, the lake gradually thawed, resulting in dynamic changes in water boundaries. In contrast, Yangcheng Lake maintained a stable water coverage throughout the year. The multi-temporal nature of this dataset provides diverse scenarios for model training, including floodplain water bodies during flood seasons, transitional zones during intermediate periods, main river channels during dry seasons, and plateau lakes during ice periods.(2) Heterogeneity in Water Body MorphologyThis study selected three typical lakes to construct a comparative observation system: Poyang Lake, the largest freshwater lake in China, exhibits significant changes in water boundaries due to the backwater effect of the Yangtze River, featuring complex texture characteristics of natural wetlands. Nam Co, located on the Tibetan Plateau, represents a typical plateau lake dominated by natural temperature variations, with notable seasonal snow and ice coverage. Yangcheng Lake, situated in the urban agglomeration of the Yangtze River Delta, is regulated by sluices and interspersed with aquaculture ponds, presenting a fragmented pattern under artificial intervention. The combination of these three lakes ensures a sufficient sample size for large-scale dynamic water bodies while encompassing challenges such as water-ice morphological changes in high-altitude regions and the identification of dense, small water bodies.(3) Fine-Grained AnnotationThis study employed automated annotation and manual visual interpretation, combined with rigorous accuracy verification, to achieve high-precision annotation results. Given the dense distribution of small water bodies and the complexity of water types in Yangcheng Lake, the annotation results for this region were refined by distinguishing between three types of water bodies: lakes, rivers, and ponds. The differentiation between lakes and ponds was primarily based on statistical features such as shape area and width-to-length ratio after segmentation, supplemented by manual visual confirmation. For the Nam Co region, considering its extensive snow and ice coverage in winter, annotations were applied to snow and ice areas, further enriching the water body annotation information provided by the dataset.

本研究采用中国高分辨率对地观测系统中高分一号（GF-1）卫星搭载的宽幅多光谱相机（WFV）数据（图1）。该数据集空间分辨率为16米，涵盖蓝（0.45-0.52 μm）、绿（0.52-0.59 μm）、红（0.63-0.69 μm）及近红外（0.77-0.89 μm）四个光谱波段。卫星重访周期为4天，可高效获取大范围地表数据，适用于水体监测、土地利用分析及环境变化检测。

数据集包含两个典型湖区：鄱阳湖（289幅影像）与阳澄湖（100幅影像）。每幅256×256像素影像均包含四个光谱波段及对应的单波段水体掩膜。

**多时间特征**
数据集全面覆盖两湖的季节变化（春、夏、秋、冬）。图2展示了年内水体覆盖动态：2022年鄱阳湖遭遇严重干旱，自8月起超80%区域呈中到极端干旱状态，而阳澄湖水体比例保持稳定。这种时间多样性为模型训练提供了洪泛区水体、过渡带及枯水期河道等差异化场景。

**水文形态异质性**
研究选取两类对比鲜明的湖泊：作为中国最大淡水湖的鄱阳湖呈现自然湿地纹理，其边界波动受长江回水效应显著影响；长三角城市群中的阳澄湖则显示人工碎片化格局，包含闸门调控水体与交错养殖池塘。该配置既提供了大范围动态水体样本，也为识别高密度小型水体带来挑战。

**精确标注**
结合自动化处理、人工目视解译及严格精度验证，数据集实现了高精度标注。针对阳澄湖复杂水文景观，特别对湖泊、河流及池塘三类水体进行区分性标注，丰富了水文信息。

综上，本数据集整合了两个形态迥异湖泊的季节性水体分布模式与高精度标签，提供多样化训练样本以提升深度学习模型在跨水文条件下的泛化能力与鲁棒性。其高分辨率多光谱特性进一步支撑精细化水体提取与环境监测研究。