A UAV Multimodal Remote Sensing Dataset for Wind Damage and Flooding Induced by Typhoon Hagas in Coastal Guangdong, China (2025)

This dataset was collected during the 72-hour post-typhoon “Huasasha” golden rescue window from September 26 to 28, 2025, covering seven sample areas in Hailing Island, Yangjiang City and two areas in Enping City, Jiangmen, with a total coverage of approximately 10 km². Data acquisition was conducted using a DJI M300 RTK UAV platform equipped with the GreenValley LiAir X4 airborne LiDAR system, synchronously capturing high-resolution RGB imagery. The UAV flew at an altitude of 80–100 m, with forward overlap exceeding 70% and side overlap exceeding 60%, and a laser pulse frequency of 200 kHz to ensure dense point cloud coverage over buildings, vegetation, and water-land interfaces in the post-disaster areas.Raw point clouds were generated by tightly coupled differential processing of onboard GNSS/IMU data to achieve centimeter-level trajectory accuracy and projected into WGS 84 / UTM Zone 49N coordinate system. Subsequent denoising, classification, and ground point extraction were performed in LiDAR360, using a progressive densified TIN algorithm (maximum building size 100 m, grid step 0.5 m) for complex terrain with fallen trees. A 0.5 m resolution digital elevation model (DEM) was generated from ground points, and a 0.5 m resolution digital surface model (DSM) from the full point cloud. RGB images were processed in Pix4Dmapper for bundle adjustment and orthorectification, with high-precision DEM used for geometric correction, producing digital orthophoto maps (DOM) at better than 0.1 m resolution, ensuring precise alignment with the LiDAR point clouds.The dataset contains nine typical sample areas, each stored in a folder named following the logic “ID_CityAbbr_CoreScene” (YJ for Yangjiang, EP for Enping), with a total size of approximately 91.7 GB. Each folder includes four standardized data types: colored point clouds (.las, density >200 pts/m², containing 3D coordinates and RGB), DEM (_DEM.tif, 0.5 m resolution, representing bare-earth elevation), DSM (_DSM.tif, 0.5 m resolution, representing top surface of objects), and DOM (_DOM.tif, <0.1 m resolution, representing surface textures and disaster details). Table records have row labels as sample IDs and column labels as geographic location, core disaster features, and observed targets, with units in meters or points/m². The dataset is mostly complete, with only minor sparsity in areas under high building shadows, which is mitigated by DEM/DSM interpolation.Quality control was applied throughout acquisition, processing, and product generation. RTK fixed solution ensured centimeter-level spatial reference, high-precision strip adjustment removed misalignment, and a combination of automated filtering and manual editing ensured accurate point cloud classification, smooth DEM surfaces, and complete DSM features. Point clouds and DOM are strictly co-registered, with feature alignment errors under 5 cm. Data are compatible with mainstream GIS and point cloud software including ArcGIS Pro, QGIS, ENVI, LiDAR360, and CloudCompare, and can be read in Python using GDAL, Rasterio, Laspy, and Open3D for batch processing and AI model training.This dataset extensively captures post-typhoon scenarios including high-rise building flooding, roadside fallen trees, collapsed crop structures, and normal waterbody terrain in reservoirs and rivers. It is suitable for 3D post-disaster damage assessment, fine-scale hydrological simulation, building extraction under complex backgrounds, and multimodal AI algorithm development, providing high-precision, multimodal benchmark data for disaster mitigation research and practical applications.