Year of Award


Document Type


Degree Type

Master of Science (MS)

Degree Name


Department or School/College

College of Forestry and Conservation

Committee Chair

Carl A. Seielstad

Commitee Members

Charles Palmer, LLoyd P. Queen


Forestry, Fuels, LiDAR, Remote Sensing


University of Montana


Quantifying surface fuels in forests is problematic for land managers due to the difficulty in measuring fuels of different sizes and spatial variability. Estimating fuel loads is important for identifying departures from historical fire regimes, predicting fire behavior and effects, and prioritizing parcels for fuels reduction. Current field methods of estimation are not always cost-effective nor can they be practical for full coverage at landscape scales. Several studies have examined remote sensing techniques for estimating fuel loads. One of the most promising is Light Detection and Ranging (LiDAR), which thus far has been applied primarily to forest canopies. Metrics derived from LiDAR include canopy base height, canopy bulk density, biomass, crown height, basal area, and tree stem location. This study focuses on the surface fuel bed, defined as the two meter stratum above ground. The relationships between LiDAR-derived surface roughness and fuels were explored in mixed-conifer forest using a relatively sparse LiDAR dataset (~1 point/m2). Surface roughness was imputed as the standard deviation of ground height distribution of laser pulse returns. Field data were derived from the nationally-scoped Fire-Fire Surrogate Study for 432 plots using two opposing azimuth Brown’s transects at each sample point. Fuel loading and surface roughness were both highly variable at plot level across the study area. Total biomass could be predicted at a nine ha resolution (R2 = 0.73). Relationships for total biomass in the fuelbed, analyzed at 2.25 ha and 0.07 ha resolutions, showed less correlation (R2 = 0.56 and 0.094, respectively). Individual surface fuel components were analyzed for correlation with surface roughness. A combination of forest floor mass and 1-hour fuels produced the highest correlation (R2 = 0.86). Additionally, LiDAR-derived data were used to derive fire behavior fuel models. Fuel models were classified by decision tree, CART analysis, and unsupervised classification using LiDAR-derived inputs. Results were validated using 101 gridded forest inventory plots. While LiDAR consistently characterized the plots at fine scale, the subjective nature of fuel model designation made statistical validation difficult.



© Copyright 2010 Tim E. Wallace