Integrating national forest inventory, airborne lidar, and satellite imagery for wall-to-wall mapping of forest structure with computer vision

Effective forest management and wildfire risk mitigation require accurate, up-to-date, and spatially consistent data across large landscapes. Existing remote sensing products often suffer from inconsistencies due to disparate data sources, varying vintages, and different modeling objectives. The authors introduce the VibrantForests framework, a system designed to provide wall-to-wall, annual estimates of fundamental forest attributes—canopy cover, canopy height, aboveground biomass, basal area, and quadratic mean diameter—at a 10-meter resolution across the contiguous United States.

To achieve this, the researchers employed a two-stage modeling pipeline. First, they developed an allometric model using tabular data from the US Forest Service’s Forest Inventory and Analysis (FIA) program. This model learns the relationships between field-measured forest attributes and vegetation structure metrics. Second, they applied this allometric model to high-resolution airborne lidar data to create spatially explicit training tiles. These tiles then served as the ground-truth targets for a satellite-based computer vision model, which uses Sentinel-2 imagery to predict forest structure across the entire country.

The VibrantForests framework demonstrates robust predictive capability across a wide spectrum of forest conditions, from sparse, low-biomass landscapes to dense, high-biomass forests. A key advantage of this approach is its ability to mitigate common modeling artifacts. Specifically, the model extends the range at which saturation typically occurs in passive-sensor models, where the signal from dense canopies becomes indistinguishable. Furthermore, it reduces the regression-to-mean behavior that often leads to the overestimation of attributes in sparse conditions and underestimation in dense conditions, providing more reliable data for operational planning.

For natural resource managers and wildfire planners, the ability to access coherent, annually updated forest data is critical for decision-making at the stand and landscape levels. By standardizing the input data and modeling process, VibrantForests provides a foundation that avoids the confounding discrepancies found when mixing disparate datasets. This framework offers a scalable solution for monitoring forest health, planning restoration treatments, and assessing wildfire risk, ultimately supporting more effective stewardship of forest landscapes.