GEDI Elevation Accuracy Assessment: A Case Study of Southwest Spain

Abstract

Information about forest structures is becoming crucial to Earth's global carbon cycle, forest habitats and biodiversity. The Global Ecosystem Dynamics Investigation (GEDI) provides 25-m diameter footprints of the surface for 3D structure measurements. The main goal of this study is to compare 12 031 footprints of GEDI data with other airborne and spaceborne Digital Elevation Models (DEMs) for Southwest Spain. Ground elevation differences (ELM) are analyzed by comparing GEDI measurements with ALS LiDAR-and TanDEM-X-derived DEMs. The vertical structure (RH100) is compared to the ALS LiDAR measurement. Ten zones are analyzed, considering different degrees of coverage and slopes. We achieved an RMSE of 6.13 m for the ELM when comparing GEDI and LiDAR data and an RMSE of 7.14 m when comparing GEDI and TanDEM-X data. For some of the studied areas, these values were considerably smaller, with RMSE values even lower than 1 m. For the RH100 metric, an RMSE of 3.56 m was achieved when comparing GEDI and LiDAR data, but again with a minimum value of 2.09 m for one zone. The results show a clear relation to coverage and slope, especially for the latter. This work also evaluates the positional uncertainty of GEDI footprints, shifting them ±10 and ±5 m along and across the track of the satellite orbit and their intermediate angular positions. The outcomes reveal a strong tendency to obtain better results in the ELM when setting the footprint to 270° and displacing it within 10 m of its positional uncertainty in comparison with the LiDAR and TanDEM-X data.

Assessment of terrain elevation estimates from ICESat-2 and GEDI spaceborne LiDAR missions across different land cover and forest types

“…An increasing residual error of GEDI terrain estimates on steep slopes is partly caused by geolocation error of GEDI samples of around 10 m for version 2 (Beck et al, 2021). Similar trends in increased GEDI residual errors with an increasing slope were observed in temperate forests of central Germany (Adam et al, 2020), in Mediterranean temperate forests of Southwest Spain (Quirós et al, 2021), in an alpine forest region of northern Italy (Kutchartt et al, 2022), in different ecozones across the United States (Liu et al, 2021) and across Eucalyptus plantations in Brazil (Fayad et al, 2021). Kutchartt et al (2022) reported that terrain slope is the most important factor influencing GEDI terrain height accuracy among other environmental (forest type, canopy cover) and sensor (beam sensitivity) parameters.…”

Fusing Sentinel-1 and -2 to Model GEDI-Derived Vegetation Structure Characteristics in GEE for the Paraguayan Chaco

“…In other words, only a fraction of the GEDI samples is captured by the model predictors, limiting the spectral representation of model predictors. Furthermore, the studies of Dorado-Roda et al, 2021 (european mediterranean forests) [69] and Quirós et al, 2021 (Southwest Spain) [70] highlight that there are certain limitations to GEDI-derived canopy height estimates and georeference. But limitations in GEDI-derived canopy height are specifically related to highly multilayered forest structures [69], which are only a minor proportion of the forests of the Paraguayan Chaco (Figure 7).…”

Quantifying and correcting geolocation error in spaceborne LiDAR forest canopy observations using high spatial accuracy data: A Bayesian model approach

“…While (2) subsumes many submodels, some of which are discussed in Section 5, the one we consider here is a model-based analog to several GEDI geolocation correction approaches, see, for example, Quirós et al (2021), Roy et al (2021), and, Blair and Hoften (1999). Again using the local coordinate system across all n observed locations, we estimate…”

“…Persistent geolocation errors may undermine forest canopy height retrievals in areas of complex or heterogeneous forest structure (Frazer et al, 2011; Milenković et al, 2017; Roy et al, 2021) and thus hinder identification of coincident measurements with field plot and ALS datasets (Duncanson et al, 2022). To this end, previous studies have implemented a variety of methods to characterize or even correct the geolocation uncertainty of spaceborne LiDAR data, including GEDI, and improve estimates of terrain and canopy heights, which are key covariates for forest AGB estimation (Liu et al, 2021; Quirós et al, 2021; Roy et al, 2021; Wang et al, 2022). In these studies, coincident ALS data with smaller geolocation errors are compared with spaceborne LiDAR measurements to assess disagreements between reported spaceborne RH metrics and observed ALS data, often via simulation to allow direct comparison.…”

“…This waveform matching approach resulted in slight improvements in both RMSE and mean absolute error (MAE) values for GEDI terrain height estimates. Similarly, Quirós et al (2021) assessed the positional error of reported GEDI footprint locations using ALS data by considering 16 positional adjustments at distances of 5 or 10 m in the along and across track directions, as well as at intermediate angular positions. Here, the position resulting in the lowest RMSE was considered optimal, and the authors observed a consistent improvement in RMSE of estimated ground elevations for positional adjustments of $$$ 27{0}^{\circ } $$$ within 10 m. These studies highlight the dual sources of geolocation error associated with spaceborne LiDAR measurements such as GEDI, which might result from both biases in the instrument itself, along with random noise at the level of individual footprints.…”