Evaluating the accuracy of ground elevation estimates from GEDI satellite LiDAR in forested terrain

Abstract

The goal of this study is to assess the accuracy of ground elevation estimates derived from data collected by the Global Ecosystem Dynamics Investigation (GEDI) full-waveform satellite LiDAR sensor, for a study area of approximately 246 km² located in northwestern Romania and with a significant broadleaved forest cover. GEDI L2A footprints were filtered, resulting in 32,666 valid footprints. The analysis involved comparing GEDI-derived ground elevation with reference data collected by Airborne Laser Scanning (ALS), which offers superior vertical accuracy. Outlier values, potentially caused by degraded pointing/positioning data from GEDI’s auxiliary systems or cloud reflections, were identified using modified <i>z</i>-score, leading to the removal of 2.4% of observations. The study further explored factors influencing GEDI performance of ground surface retrieval, such as: the presence of forest vegetation, its height and type, use of coverage vs. power beam data, the time and season of data collection and geomorphological conditions (slope and terrain ruggedness). Overall, the accuracy is reasonably good (with an RMSE of 6.75 and a bias of − 0.53 m, after outlier removal), with a significant reduction of ground surface estimation in forested areas (RMSE of 7.75, versus 5.29 m in open ground). Accuracy is mainly influenced by the season of acquisition (RMSE between 5.45 and 8.75 m, the presence and height of forest canopy (RMSE of 6.04 m for canopies under 9.2 m and RMSE of 9.81 m for canopies taller than 22.7 m) and geomorphological conditions (RMSE between 5.61 m and 9.18 m, depending on slope). Understanding the interaction of GEDI with these factors is essential for improving the utilization of GEDI data in forest ecosystem monitoring. This research underscores the importance of accurate ground modeling in forestry applications and contributes to the ongoing evaluation of GEDI performance.