Global Dam Image Dataset

The Global Dam Image Dataset (GDID) is a publicly available dataset developed for training and evaluating deep learning models for multispectral image generation and dam construction analysis. Since no public dataset previously provided paired multispectral images acquired before dam construction and after the first reservoir impoundment together with corresponding guidance masks, GDID was created to address this gap.The dataset was generated using information from the Global Dam Watch (GDW) database, Landsat Collection 2 Level-2 imagery (Landsat 4–9), Google Earth Engine (GEE), OpenStreetMap, the GLC_FCS30D and Dynamic World land-cover datasets, and a newly developed Detection of Initial Reservoir Impoundment (DIRI) algorithm. DIRI estimates the first reservoir impoundment year by analyzing long-term surface water dynamics within the reservoir extent.Each sample contains:• A six-band multispectral Landsat image acquired before dam construction.• A corresponding six-band multispectral image acquired after the first reservoir impoundment.• A guidance mask indicating the dam crest, reservoir area, and background.All images have a spatial size of 512 × 512 pixels (covering 15.36 × 15.36 km²) and include the Red, Green, Blue, NIR, SWIR1, and SWIR2 bands normalized to the range [0, 1]. The guidance masks were generated using NDWI-based water extraction and land-cover information, followed by manual refinement to ensure accurate reservoir and dam crest boundaries.After quality control and manual verification, the final dataset contains 2,188 training samples and 546 testing samples. GDID is intended for research on remote sensing image generation, conditional generative models, dam monitoring, reservoir analysis, and related Earth observation applications.