Assessment of GEDI's LiDAR Data for the Estimation of Canopy Heights and Wood Volume of Eucalyptus Plantations in Brazil

Abstract

Over the past two decades spaceborne LiDARsystems have gainedmomentumin the remote sensing community with their ability to accurately estimate canopy heights and aboveground biomass. This article aims at using the most recent global ecosystem dynamics investigation (GEDI) LiDAR system data to estimate the stand-scale dominant heights (H-dom), and stand volume (V) of Eucalyptus plantations in Brazil. These plantations provide a valuable case study due to the homogenous canopy cover and the availability of precise field measurements. Several linear and nonlinear regression models were used for the estimation of H-dom and V based on several GEDI metrics. H-dom and V estimation results showed that over low-slopped terrain the most accurate estimates of H-dom and V were obtained using the stepwise regression, with an root-mean-square error (RMSE) of 1.33m(R-2 of 0.93) and 24.39 m(3).ha(-1) (R-2 of 0.90) respectively. The principal metric explaining more than 87% and 84% of the variability (R-2) of H-dom and V was the metric representing the height above the ground at which 90% of the waveform energy occurs. Testing the postprocessed GEDI metric values issued from six available different processing algorithms showed that the accuracy on H-dom and V estimates is algorithm dependent, with a 16% observed increase in RMSE on both variables using algorithm a5 vs. a1. Finally, the choice to select the ground return from the last detected mode or the stronger of the last two modes could also affect the H-dom estimation accuracy with 12 cm RMSE decrease using the latter.