Airborne Radar Altimeter Echo Dataset for Terrain Classification

This dataset contains processed radar altimeter echo returns from nine airborne flights over two regions of Pakistan (Chenab, 4 flights; Swat, 5 flights). Each record is one radar echo of 600 samples, stored in .npy files, with a matching row in a CSV file that holds the onboard radar settings, navigation fields, waveform features and a terrain label. The original raw binary files were decoded, aligned with ground-processed radar and navigation data, labelled using public land-cover maps (ESA WorldCover 2021 and Google Dynamic World, with an SRTM slope rule for mountainous or rocky ground), and cleaned by noise reduction. Labels are weak labels from public maps and are not ground truth. The dataset has 188,780 records in five terrain classes: Vegetation (72,042), Built (39,426), Mountainous/Rocky (29,903), Water (26,721) and Bare (20,688). The dataset supports research on terrain classification from radar altimeter echoes. Please split train and test sets by flight or record group, because records from one flight are very similar. Files: 9 waveform .npy files, merged_coarse_header.csv, README.md. See the README for the structure and column descriptions. Acknowledgements: The data was collected with the support of the Ministry of Science and Technology (MoST) and COMSATS University Islamabad.