A New Foundation for LANDFIRE- Current and Future Innovations LANDFIRE has been a foundational dataset for countless applications within the Forest Service and beyond. The machine learning methodology largely used in the creation of LF National (circa 2001) and LF 2016 Remap was rooted in supervised classification of imagery using classification and regression tree models. Training data consisted of plot data over many years and sampling of lidar tree cover and height. Although these methods were innovative for their time and stood the test of time over the last 20 years, the amount and speed of data that can be used for modeling and mapping with more accuracy, detail and frequency has increased to unimaginable levels. For LF2023 (to be released in 2024), LANDFIRE will be moving towards the idea of a ‘continuous remap’ by using machine learning algorithms on an annual basis to map lifeform, existing vegetation cover, and existing vegetation height for herbaceous, shrub and tree lifeforms.
The LANDFIRE program is aware that our stakeholders are interested in obtaining more comprehensive 3D vegetation structure information to inform vegetation and species mapping, carbon accounting, and physics-based fire behavior models. Understanding how to provide annually updated 3D vegetation and fuel metrics in a way that is useful to the most stakeholders and accounts for the logistical and resource constraints within the program is a LANDFIRE goal over the next few years. LANDFIRE wants to connect with innovators who are motivated by the shared challenge of pulling together disparate data sources across scales and dimensions into logical machine learning or deep learning classification methodologies that are accurate, repeatable, and usable by managers. We hope to build these relationships by increasing our engagement across research and management communities to create a new foundation for LANDFIRE.