Terrestrial ecosystem structure is linked to the flow and cycling of energy and nutrients, and to habitat and species richness of plant and animal species. Numerous remote sensing studies have quantified spatial variation in forest structure, but our understanding of the drivers of this variation is poorly understood, especially at regional to continental scales. In addition, forest structure is a three-dimensional property, but when studying the drivers, previous studies either use one single metric (e.g., maximum tree height or AGBD) that does not represent the vertical distribution of vegetation or employ a few metrics (e.g., stem density, basal area) that only reflect two-dimensional information. Here we use data from a spaceborne waveform lidar (the Global Ecosystem Dynamics Investigation) to isolate the importance of the physical environment and anthropogenic variables on worldwide terrestrial ecosystem structure using machine learning. Our analysis isolates the importance of physical climate, including temperature and precipitation, topographic and edaphic variables, and anthropogenic impacts on terrestrial ecosystem structure in 735 terrestrial ecoregions among 13 biomes on all continents except Antarctica. We partition variation in forest structure into components that are attributable to physical and anthropogenic drivers, and investigate how the influence of physical and anthropogenic drivers varies among terrestrial ecoregions. Results show that there is variation in the strength of human impacts in comparison to physical drivers among terrestrial ecoregions. The importance of human impacts is greatest in forested ecoregions and smallest in montane grasslands and shrublands. However, physical variables overwhelm the importance of anthropogenic drivers of terrestrial ecosystem structure in most terrestrial ecoregions.
