International Journal of Applied Earth Observations and Geoinformation (Nov 2024)
Intercomparison of the DART model and GEDI simulator for simulating GEDI waveforms in forests
Abstract
The launch of GEDI opens a new era of forest structure monitoring using full-waveform LiDAR from space. Simulation of GEDI waveform is of great importance for the algorithm design and forest structure metric estimation. DART is a universal 3D radiative transfer model for simulating remote sensing signals by modeling light propagation in 3D landscape, with DART-RC adopting forward ray tracing and DART-Lux adopting bidirectional path tracing to model multi-mode multi-platform LiDAR signals. GEDI simulator is a semi-empirical model for simulating GEDI waveform with point clouds. In view of their different solutions, this study aims to evaluate and inter-compare the accuracy of DART-RC, DART-Lux, and GEDI simulator for simulating GEDI waveforms in forests. Because there are possible geolocation errors and 3D modeling uncertainty in real forests, we propose to adopt virtual forests (turbid medium, RAMI scenes) for consistency evaluation and real forests for accuracy evaluation. Metrics including R2 (coefficient of determination), RMSD (root mean square deviation), RMSE (root mean square error), and bias were calculated from the normalized waveform intensities from different simulation methods.According to the results, it is demonstrated that DART-RC/Lux, when only considering first order scattering, can simulate waveforms that are highly consistent compared to GEDI simulator in virtual forests. In real forests, DART-RC/Lux which considers first order and multiple scattering can simulate more accurate waveforms than GEDI simulator in sparse canopies (ePAI 1.2). The challenge in accurate forest 3D modeling is the most likely reason causing decreasing accuracy of DART-RC/Lux in dense forests. We recommend the use of GEDI simulator when airborne point cloud is available, while the use of DART-RC and DART-Lux for mechanism studies or theoretical analysis. The results of this study contribute to a better understanding of LiDAR waveform simulation methodologies in forests.
