ACCORDING TO THE IT HOUSE ON February 6, according to the website of the Institute of Botany of the Chinese Academy of Sciences, high-resolution forest canopy height products at the national and global scales are crucial for estimating forest carbon storage, understanding forest ecosystem processes, and formulating forest management policies. The spatial resolution of existing forest canopy height products is generally 500 meters or 1000 meters, which is difficult to meet the application requirements.
The newly launched spaceborne lidar sensors GEDI and ICESat-2 ATLAS can acquire footing data on forest canopy heights at decimeter-level spatial resolution on a global scale, providing the possibility of high-resolution mapping of large-scale forest canopy heights. However, the foot point data obtained by spaceborne lidar is discretely distributed along the satellite orbit and cannot directly generate a spatially continuous forest canopy height product. Previous studies have usually used spaceborne lidar footpoint data and remote sensing images and environmental factors to establish regression models to generate spatially continuous large-scale forest canopy height products, but this method is difficult to make full use of the dense foot data of GEDI and ICESat-2 ATLAS, and is susceptible to the saturation effect of remote sensing images. How to fully and efficiently use the new generation of spaceborne lidar data to generate large-scale high-resolution forest canopy height products needs to be further explored.
The Research Group of Su Yanjun of the Institute of Botany of the Chinese Academy of Sciences designed a deep learning-guided spatial interpolation model NNGI (Neural Network Guided Interpolation). This model uses the ability of deep neural network to automatically learn weights, solves the weight contribution of how to balance the consideration of space, environment and spectroscopy in the interpolation model, and breaks through the limitation of using only spatial distance in commonly used interpolation models. Using the 140 square kilometers of UAV lidar data accumulated by the research group, the NNGI model was trained, and the GEDI and ICESat-2 ATLAS spaceborne lidar footpoint data were interpolated to successfully map the national 30-meter resolution forest canopy height product. The products obtained were more accurate than three independent validation datasets—more than 1 million GEDI pin data, 33 square kilometers of UAV lidar data, and nearly 60,000 forest inventory data. In addition, thanks to spatial interpolation strategies, NNGI-generated products are barely saturated in areas with higher forest canopy. This set of high-precision, high-spatial resolution national forest canopy height products shows that the NNGI model has good application potential in monitoring forest canopy height at national and global scales, and the resulting national forest canopy height products can help improve the accuracy of large-scale forest biomass estimation and provide data support for the formulation of climate mitigation policies and the realization of "carbon neutrality" goals.
▲ National forest canopy height distribution map with 30-meter spatial resolution | Source: Website of the Institute of Botany, Chinese Academy of Sciences
IT House learned that the research results were published online in December 2021 in the international academic journal Remote Sensing of Environment.
Article links:
https://www.sciencedirect.com/science/article/pii/S0034425721005642?dgcid=coauthor