The potential of Sentinel-2 and -1 for upscaling GEDI LiDAR sampling of vegetation height at global and ecosystem-level

Abstract:

Global mapping of forest structure is important to better understand the carbon cycle, estimate aboveground biomass, and monitor biodiversity. The recent launch of the Global Ecosystem Dynamics Investigation (GEDI), a high-resolution spaceborne LiDAR, is providing new opportunities in forest monitoring by measuring the vegetation structure and providing datasets of canopy height measurements, canopy cover, leaf area index, and vertical vegetation profiles. GEDI will sample around 4% of the global land surface between ±51.6^° latitude. To achieve complete coverage of the land surface, an extrapolation based on other continuous variables is needed, like optical and radar satellite images.

In this study, we evaluated the potential of Sentinel-2 in combination with Sentinel-1 to upscale GEDI LiDAR samples of vegetation height at global and ecosystem-level (14 broad ecosystem types). We processed the entire GEDI archive for the first 6 months (April-October 2019) to extract vegetation height. We ran multiple Random Forest (RF) scenarios by fusing Sentinel-2’s spectral information and vegetation indices with Sentinel-1’s radar information for estimating vegetation height at 1 km spatial resolution. We tested the influence of the number of GEDI samples available within a 1 km pixel for training the RF on the final accuracy of vegetation height estimations. Accuracy increased when using 1 km pixels with an increasing number of GEDI samples for training the RF. For example, the validation R² for the tropical moist forest ecosystem increased from 0.59 to 0.73 when a minimum of 10 or 50 GEDI samples were available within a 1 km training pixel. The R² for the ecosystem-based analysis varied between 0.53 for mangroves and 0.82 for temperate grasslands, while the global analysis yielded an R² of 0.80. We discuss the challenges and opportunities of using Sentinel-2 and 1 within an RF workflow for upscaling GEDI vegetation height and present a global map of canopy height estimation at a 1 km spatial resolution. Since the GEDI mission is in its early stages, our in-depth analysis will help users in their future choices when combining GEDI with Sentinel-2 for a large-scale vegetation height assessment, a critical monitoring tool for better understanding of the global forests.