The cumulative area of LiDAR collections across multiple ownerships in the northwestern United States has reached the point that land managers of the U.S. Forest Service (USFS) and other stakeholders would greatly benefit from a strategy for how to utilize LiDAR for regional aboveground biomass inventory. The need for Carbon Monitoring Systems (CMS) can be more robustly addressed by using not only available NASA satellite data products, but also commercial airborne LiDAR data collections.
The reference database of field and lidar observations of initial conditions is in a format ready for ingestion into the latest version of the Forest Vegetation Simulator with climate change projection capabilities.
We are using multiple airborne lidar datasets previously acquired at the project level in conjunction with field plot datasets to predict aboveground biomass across the diverse vegetation types of the northwestern United States.
Project-level biomass maps will serve as training areas for predicting regional biomass carbon annually from Landsat time series imagery processed through LandTrendr.
Regional maps will be validated with the U.S. Forest Service’s Forest Inventory and Analysis (FIA) data summarized annually at the county level.
Annual (2000-2012) regional biomass maps will be published on the Oak Ridge National Laboratory’s Data Active Archive Center along with county-level biases.
We have assembled and consistently processed field plot and LiDAR datasets at >53 landscape-level project areas distributed along a broad climate gradient across the northwestern U.S. from temperate rainforest to cold desert (Fig. 1). Twenty of the lidar collections we have assembled to date have accompanying field plot data.
1984-2012 Landsat image time series have been processed through LandTrendr across the entire study region. Landsat image time series have been found to explain more structural variation than can a single scene.
We are first developing our prototype modeling approach over a preliminary focal area of northern Idaho (Fig. 2).
We are using the Random Forests machine learning algorithm as our predictive modeling approach. The models are explaining approximately 2/3 of variance in above-ground biomass at both the project and regional levels.
Our annually mapped above-ground biomass predictions summarized at the county level are currently about 1.5 times higher than reported independently by the USFS Forest Inventory and Analysis Program. The overestimation bias appears to be related to the proportion of non-forest cover within the county (e.g., Fig. 3).