Abstract
The TreeMap 2016 dataset provides detailed spatial information on forest characteristics including number of live and dead trees, biomass, and carbon across the entire forested extent of the continental United States at 30 × 30m resolution, enabling analyses at finer scales where forest inventory is inadequate. We used a random forests machine learning algorithm to assign the most similar Forest Inventory Analysis (FIA) plot to each pixel of gridded LANDFIRE input data. The TreeMap 2016 methodology includes disturbance as a response variable, resulting in increased accuracy in mapping disturbed areas. Within-class accuracy was over 90% for forest cover, height, vegetation group, and disturbance code when compared to LANDFIRE maps. At least one pixel within the radius of validation plots matched the class of predicted values in 57.5% of cases for forest cover, 80.0% for height, 80.0% for tree species with highest basal area, and 87.4% for disturbance. A new feature of the dataset is that it includes linkages to select FIA data in an attribute table included with the TreeMap raster, allowing users to map summaries of 21 variables in a GIS. TreeMap estimates compared favorably with those from FIA at the state level for number of live and dead trees and carbon stored in live and dead trees. Study Implications: TreeMap 2016 provides a 30 × 30 m resolution gridded map of the forests of the continental United States. Attributes of each grid cell include a suite of forest characteristics including biomass, carbon, forest type, and number of live and dead trees. Users can readily produce maps and summaries of these characteristics in a GIS. The TreeMap also includes a database containing, for each pixel, a list of trees with the species, diameter, and height of each tree. TreeMap is being used in the private sector for carbon estimation and by land managers in the National Forest system to investigate questions pertaining to fuel treatments and forest productivity as well as Forest Plan revisions.