Ultra-high spatial resolution mapping of urban vegetation heights with multimodal remote sensing data and deep learning method

Urban forests are an integral component of urban ecosystems, which play several critical roles in improving the quality of life in cities and towns. Accurate estimation of urban forest canopy height is pivotal for quantifying forest carbon storage and understanding forest ecosystem processes as well as shaping effective forest management police to mitigate global climate change. Although spaceborne or airborne LiDAR can provide the height information, there is often a trade-off between the spatial resolution and spatial coverage. On the synergism of the above two issues, we aim to fuse the multimodal remote sensing data and digital elevation model (DEM) data for ultra-high spatial resolution vegetation canopy height estimation over large urban area. In this study, we introduce a novel deep learning model, ARFCNet, designed for vegetation canopy height mapping employing unmanned aerial vehicle (UAV) imagery, Sentinel data, and DEM data as model inputs. We compare the potential of vegetation canopy height mapping under two strategies: the first involving RGB imagery, Sentinel-1 data, and DEM data with a spatial resolution of 1m, and the second with DEM spatial resolution of 30m. We assessed the model performance and compared with existing canopy height products and ground-based measurements. Results show that the ARFCNet model, under the first strategy, exhibits superior accuracy in estimating vegetation height across different regions, with the R² and RMSE value of 0.98 and 1.33m, respectively. We also mapped the 1-m vegetation canopy height in Guangzhou, China based on ARCFNet model, and compared it with three existing tree height products in Guangzhou (ETHGCH: Lang et al. (2023), GLIGCH: Potapov et al. (2021) and NNGIFCH: Liu et al. (2022)), with R² of 0.72, 0.61, and 0.45, and RMSE of 3.94, 6.04, and 4.81, respectively. In comparison, our ultra-high spatial resolution (1m) vegetation canopy height provides detailed measurements especially in the land cover type of urban land. It holds promise for national or global vegetation heights monitoring, enhancing biomass mapping accuracy, and contributing to carbon neutrality goals.