Delivering innovation in annual tree canopy maps to the American people

Maps of a 2016 wildfire event in Gatlinburg Tennessee. The upper left map is TCC in 2015, before the wildfire. The upper right map is TCC in 2017, the year after the fire. The lower left map is TCC in 2021, showing TCC gain ln the years following the fire.

The National Land Cover Database (NLCD) Tree Canopy Cover (TCC) is a federally designated National Geospatial Data Asset within the Land Use-Land Cover Theme, where the Forest Service is one of two national co-lead agencies alongside USGS.

A new manuscript, the first of its kind since the Forest Service began producing NLCD TCC data in 2011, documents the redesigned and rebuilt TCC production workflow. The team's work on this project prioritized scaling new techniques and research,This new peer-reviewed methods manuscript, the first of its kind since the Forest Service began producing NLCD Tree Canopy Cover (TCC) data in 2011, documents the redesigned and rebuilt TCC production workflow. The work prioritized scaling new techniques research, temporal cohesion of annual continuous canopy cover measurements and maps, robust QA/QC, and delivering rapid statistical map accuracy metrics along with Tau — a true measure of model uncertainty. 

Federal, state, and research entities rely on TCC data for many applications across the nation's geospatial data infrastructure. However, it's challenging for long-term federal data projects like TCC to remain responsive to and help drive new knowledge discovery. 

Research findings are meant to inform ongoing development and, where appropriate, can be scaled, tested and validated for operations and applications across the agency. Meanwhile, ongoing tech transfer with wider management and user communities creates shared dialogue about real-world product performance and emerging needs. These results can sometimes prove that a new approach fails to solve the problem. This project's structure supports a robust innovation pipeline, without asking production to take on the risk of experimentation that could endanger the regular delivery of consistent, quality time-series products that users depend on.

This project relies on a research-to-operations and operations-to-research feedback chain, with development as a critical link, by integrating Forest Inventory and Analysis (FIA) and Field Services and Innovation Center Geospatial Office (FSIC-GO) staff and contractors. 

For this work, an independent accuracy assessment against 16,607 FIA reference plots produced a national mean absolute error of 8.0 percent TCC (with a weighted root mean square deviation of 12.8 percent). Smaller updates continue to be documented in technical white papers with each annual time-series data release reflecting the different, yet overlapping, needs of tech transfer for research and operations.

A 2023 paper by Greenfield and Schleeweis provides guidance to managers and i-Tree users by demonstrating statistically sound methods and results, comparing TCC time-series datasets to estimates generated from high-resolution photointerpretation.

Beyond the data and maps, this project's techniques and applications research feed back into large-area inventory and monitoring science more broadly, testing core assumptions like temporal model transfer and the reliability of crowd-sourced reference data. One outcome, described in the paper, is that TCC annual time-series maps now ship with statistical accuracy assessment results, more robust and repeatable quality control, and a built-in confidence indicator (rather than a probability score), which enables a consistent, repeatable way for users to judge how much confidence to place in the map at any given point.