Development of a New Modeling Framework for LANDFIRE Vegetation Products

Submitted by tadamson on

LANDFIRE products are integral to fire management and research within the United States. An essential component of LANDFIRE is the development of detailed vegetation maps, which use 30-meter Landsat imagery as the principal source of remotely sensed data. Classified vegetation layers, including lifeform and Existing Vegetation Type (EVT), were modeled for LANDFIRE 2016 Remap using categorical and regression tree (CART) approaches, while continuous vegetation layers such as Existing Vegetation Cover (EVC) and Existing Vegetation Height (EVH) were modeled using thematic and regression tree (Cubist) techniques. Model-dependent variables relied on Forest Inventory and Analysis, Bureau of Land Management, and U.S. Department of Agriculture Natural Resources Conservation Service plot data. Landsat-derived bands and indices were the predominant model independent variables, with additional remotely sensed topographic and climatic variables used to a lesser extent. For LANDFIRE 2016 Remap, map accuracy metrics were produced only for EVT. To improve data processing, modeling, and post-modeling and provide internal model and map accuracy assessments for classified and continuous LANDFIRE data products, scientists developed open-source enhanced performance (multithreaded) scripted procedures that can be run on desktop, virtual desktop, high performance computing (HPC), and cloud computing systems. Although these procedures were specifically developed for LANDFIRE, their implementation in is generic enough to allow for the repurposing of scripting procedures and use by other modeling/mapping programs.

Overview of the LANDFIRE mapping process, which starts with determining variable importance and ends with the production of a model and output map with accompanying error analysis.

Disclaimer: Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.

Author Name
Joshua J. Picotte
Author Email