Unveiling the performance and influential factors of GEDI L2A for building height retrieval

Estimating building heights is essential for urban planning, disaster assessment, and sustainable development. While the Global Ecosystem Dynamics Investigation (GEDI) Light Detection and Ranging (LiDAR) was primarily designed for forest measurements, it also holds potential for large-scale building height retrieval. This study evaluates the performance and influential factors of GEDI L2A version 2 (V2) data for building height retrieval by comparing it with the airborne LiDAR-derived normalized digital surface model (nDSM). To ensure data reliability, we refined the GEDI dataset by excluding footprints outside buildings, filtering out low-quality footprints, removing footprints failing to detect ground elevation using the interquartile range (IQR) detection method, and excluding footprints with geolocation errors through an eight-direction offset approach. We assessed the effectiveness of different relative height (RH) metrics and systematically analyzed key influential factors in building height retrieval. Results indicate that GEDI RH96 achieves the highest correlation with reference building heights (R2 = 0.82, MAE = 1.67 m, RMSE = 4.40 m, rRMSE = 34.46%). GEDI demonstrates the highest accuracy for mid- and high-rise buildings, whereas low-rise buildings (<5 m) exhibit lower accuracy and tend to be overestimated (RMSE = 2.17 m, rRMSE = 49.79%). Sensitivity and slope are the most significant factors influencing the accuracy of building height retrieval. GEDI data with sensitivity above 0.95 showed a 4.66% decrease in rRMSE compared to data with sensitivity above 0.90. Slope negatively affects building height retrieval accuracy. Building roof type has a moderate impact; flat-roof buildings exhibit a slight advantage over pitched- and curved-roof buildings, with rRMSE reductions of 1.86% and 4.74%, respectively. Neither GEDI beam type nor data acquisition time significantly affect the accuracy of height retrieval. Overall, this study provides valuable insights for optimizing GEDI data in building height retrieval, contributing to large-scale building height mapping.