We used a random forests machine learning process to produce a tree-level model of the US by assigning forest plot data from Forest Inventory Analysis (FIA) to landscape data from LANDFIRE at 30x30 resolution. The resulting dataset has numerous applications. These include maps of the number of live and dead trees at pixel scale, which can be summarized to units like the state (state-level summaries of live trees had R2=0.95 and dead trees had R2=0.92 when compared with estimates from FIA). Pixel-level summaries of carbon in live and dead trees compare favorably with FIA estimates at the National Forest level. The TreeMap was also used to produce a map of Snag Hazard, based on the median height and density of snags. The Snag Hazard had an 86% within-class accuracy when compared with ground plots. The Snag Hazard map is used by the Forest Service’s Risk Management Assistance group during active fire incidents. We are currently leveraging measurements of litter, duff, fine and coarse woody debris made by FIA to produce a FuelMap. In the short term, we plan to use TreeMap and FuelMap in estimation of risk to forest carbon from wildland fire.
Mapping live and dead trees, fire responder hazard, carbon, and more with TreeMap, Snag Hazard, and FuelMap datasets