How can UAV bridge the gap between ground and satellite observations for quantifying the biomass of desert shrub community?

Abstract

Accurate estimation of the shrub above-ground biomass (AGB) is an essential basis for determining carbon storage and monitoring desertification risk in arid ecosystems. However, significant uncertainties and biases exist in large-scale monitoring of desert AGB due to the scale mismatch and spatio-temporal topological errors between the coarse resolution of satellite images and the limited area that can be surveyed via ground measurements. The rapid development of unmanned aerial vehicle (UAV) technology has the prospective advantage of multi-scale vegetation observation. However, its potential to bridge the gap between satellites and the ground in desert AGB estimation has not yet been explored. This study developed a procedure to fill the gap between satellite and ground measurements based on low-cost and easy-to-use UAV visible-light technology in typical desert shrub communities in Inner Mongolia, China. First, canopy area (CA), canopy maximum height (CH), and canopy volume (CV) metrics derived from UAV-RGB (Red, Green, Blue) images, coupled with structure-from-motion photogrammetry, were used to invert the UAV-based AGB. Then, the UAV-based AGB data were aggregated to different scales to align with satellite data. Here, we focused on examining the performance of generalized additive models between the upscaled UAV-based AGB and vegetation indices (VIs) generated from PlanetScope (resolution: 3 m), Sentinel-2A MSI (resolution: 10 m, 20 m), and Landsat 8 OLI (resolution: 30 m). Finally, we investigated the effects of scale and spectra in the upscaling and modeling process. Results showed that the UAV-based AGB linear prediction models developed by the CV metric performed best for Reaumuria soongarica (R2 = 0.749, RMSE = 65.5 g) and Salsola passerina (R2 = 0.919, RMSE = 56.7 g), which allowed for accurately mapping the desert AGB at the individual plant level with ultra-high resolution (2 cm). The performance of satellite biomass models was excellent based on the upscaled UAV-based AGB and VIs (best-performing models for different satellites: adjusted R2 = 0.625–0.934, RMSE = 35.1–119.1 g/m2). As the resolution of satellite data increases, the increase in variance is a big scale-related challenge in AGB model performance evaluation. It is worth noting that the UAV-based AGB data are unmodifiable due to their area-independent nature, which can avoid the scale-related effect. Moreover, the satellite VIs most closely related to the desert AGB were associated with Red, NIR, and Red-Edge bands. Specifically, we recommend the Red-Edge band when using Sentinel-2A data. This study proves that UAV technology can not only provide exhaustive information on vegetation biomass but also can train and validate satellite data beyond individual UAV observations, which can significantly improve large-scale biomass monitoring if widely applied in the future.