Estimation of Fractional Plant Lifeform Cover for the Conterminous United States Using Landsat Imagery and Airborne LiDAR

In this study, we produced annual 30-m resolution maps of vegetation lifeform cover (tree, shrub, herb, and an additional class denoted as “other”) and their spatially explicit uncertainties for the conterminous United States for the period of 1984–2018, modeled using a quantile regression forest calibrated from a fusion of airborne light detection and ranging (LiDAR)/multispectral data and Landsat imagery. The resulting predictions were compared to available fractional cover products with the same temporal and spatial resolution and to different types of documented land cover change to determine the relative accuracy of the percent cover predictions. We found that our approach yielded a higher accuracy of the shrub, herb, and “other” cover estimates than existing products and had a similar performance of the resulting tree cover estimate compared to other products. Our results indicate that the use of spatially dense training and validation data from airborne multisensor acquisitions facilitates the incorporation of a large range of variability in the response variables into model construction. The produced percent lifeform cover maps, which are publicly available through the Oak Ridge National Laboratory Distributed Active Archive Center, represent a consistent dataset of medium spatial resolution and long temporal record that can be a valuable input into a variety of applications such as the analysis of vegetation change processes, as well as ecosystem conservation and management.