Transformer-based Reconstruction of Canopy Profiles from Large-Footprint Waveform LiDAR

This dataset contains paired waveform LiDAR and airborne laser scanning (ALS) reference profiles used to evaluate Transformer-based canopy-profile reconstruction over the Smithsonian Environmental Research Center (SERC), Maryland, USA. The data were derived from NASA’s Land, Vegetation, and Ice Sensor Facility instrument (LVIS-F) and the National Ecological Observatory Network’s (NEON) Airborne Observation Platform (AOP), both acquired over SERC in August 2021. The final dataset contains 101,110 spatially coincident LVIS-F waveform and ALS canopy-profile pairs after filtering and preprocessing. The dataset is provided as Python pickle (.pkl) files: LVIS-F_preprocessed_waveforms.pkl: Preprocessed large-footprint full-waveform LiDAR observations acquired by LVIS-F. LVIS-F shots were filtered to retain only footprints overlapping forested 30 m pixels in the National Land Cover Database (NLCD) 2021 land-cover product, thereby restricting the analysis to forested areas. The retained waveforms were subsequently smoothed and thresholded to suppress noise and isolate the signal-bearing portion of each waveform. AOP_ALS_canopy_profiles.pkl: Reference canopy profiles derived from NEON AOP ALS point clouds. Classified ALS point clouds were clipped into 7 × 7 m square vertical columns spatially coincident with the filtered LVIS-F shots, with each column centered on the geolocated ground coordinate of the corresponding LVIS-F footprint. Filtering steps were applied to retain high-quality forest canopy samples, after which ground, low-vegetation, and noise-classified points were removed. The remaining canopy returns were aggregated into 0.15 m vertical bins to match the vertical sampling interval of the LVIS-F waveforms. These binned ALS return-count profiles served as reference canopy profiles for training, validating, and testing the waveform reconstruction models.