Measurements, Models and Maps: Large-area Forest Inventory from Airborne LiDAR Data

Sponsor: University of Maine Cooperative Forestry Research Unit

Daniel Hayes, Associate Professor, School of Forest Resources - University of Maine
David Sandilands, Remote Sensing Technician, School of Forest Resources - University of Maine
Stephanie Willsey, M.S. Student, School of Forest Resources - University of Maine
Anthony Guay, Remote Sensing Specialist, School of Forest Resources - University of Maine
Ian Prior, Inventory Analyst - Seven Islands Land Company
Steve West, GIS Specialist - Seven Islands Land Company
Kyle Burdick, Woodlands Manager - Baskahegan Company
Adam Dick, Forest Research Project Leader, Canadian Wood Fibre Centre - Natural Resources Canada
Aaron Weiskittel, Professor of Forest Biometrics and Modeling, School of Forest Resources - University of Maine

10/01/2018 - 9/30/2023

LiDAR point cloud clipped to a field-measured inventory plot

This project has organized and carried out initial investigations into the use of LiDAR remote sensing analysis to enhance the design and operation of inventory programs for Maine’s forest industry stakeholders. The research conducted here is evaluating ground-based inventory plot designs together with existing, publicly-available Airborne Laser Scanning (ALS) data sets processed in a high performance computing environment for workflow efficacy in generating geospatial data products useful for forest management. For these initial investigations, we have partnered with the Seven Islands Land Company in using their Ashland West property (~150,000 acres) to evaluate the impact of plot type, size and location accuracy on model prediction of forest inventory attributes derived from relating field data sampling with wall-to-wall LiDAR measurements across the study area.


  • We are building a set of workflows for generating gridded maps of forest inventory attributes from ALS data sets using an area-based modeling approach calibrated on ground-based plot data.
  • The first step in the workflow is to acquire and organize the plot data locations and measurements from the ground-based inventory that are used to ‘clip out’ the associated locations and metrics generated from the LiDAR point clouds.
  • Next, a statistical model is developed that relates the LiDAR metrics to the plot-based inventory measurements for each concurrent location. The model is evaluated in terms of its explanatory power, average error and bias in matching the predictions to the observations.
  • Once the model is calibrated (and verified) at the plot locations, it is then applied over a wall-to-wall, gridded raster of LiDAR metrics to “predict” the inventory attributes for each grid cell in the study area. The results of the model application are evaluated against a held-out subset of plots, stand-level information and/or parcel-level summaries.
  • A designated set of alternative models are then developed, applied and evaluated to investigate applied research questions on the impacts of plot design, location accuracy, stratification and sampling intensity, along with ALS density and other acquisition specifications.


Our initial results have highlighted some of the challenges in linking plot data with the LiDAR models – particularly with variable radius plots with large locational error – but also suggest opportunities to improve results with alternative plot designs and ALS data sets that will be the focus of investigations going forward. Here, we are expanding our scientific collaboration on this project to include the Canadian Forest Service and NASA to analyze a larger set of diverse data sets (both plot data and LiDAR acquisitions) to be able to investigate a greater number of applied research questions on plot design, placement and LiDAR data density for their impacts on EFI model performance.


Daniel Hayes

Watch the Video

Go to Project Website

Skip to content