This dissertation explores the efficacy of large-footprint, waveform-digitizing lidar for the inventory and mapping of canopy fuels for utilization in fire behavior simulation models.  Because of its ability to measure the vertical structure of forest canopies lidar is uniquely suited among remote sensing instruments to observe the canopy structure characteristics relevant to fuels characterization and may help address the lack of high-quality fuels data for many regions, especially in more remote areas.  Lidar data were collected by the Laser Vegetation Imaging Sensor (LVIS) over the Sierra National Forest in California.  Various waveform metrics were calculated from the waveforms.  Field data were collected at 135 plots co-located with a subset of the lidar footprints.  The field data were used to calculate groundbased observations of canopy bulk density (CBD) and canopy base height (CBH).  These observed values of CBD and CBH were used as dependent variables in a series of regression analyses using the derived lidar metrics as independent variables. 

Comparisons of observed and predicted resulted in an r2 of 0.71 for CBD and an r2 of

0.59 for CBH.  These regression models were then used to generate grids of CBD and CBH from all of the lidar waveform data in the study area.  These grids, along with lidar-derived grids of canopy height, were then used as inputs to the FARSITE (Fire Area Simulator-Model) fire behavior model in a series of simulations.  Comparisons between conventionally derived and lidar-based model inputs showed differences between the two sets of data.  Specifically, the lidar-derived inputs contained much more spatial heterogeneity.  Outputs from FARSITE using the lidar-derived inputs were also compared to outputs using input maps of CBD and CBH generated from field observations.  There were significant differences between the two sets of outputs, especially in the frequency and spatial distribution of crown fire.  Experiments in manipulating the effective resolution of the lidar-based inputs confirmed that FARSITE outputs are affected by the spatial variability of the input data.  Furthermore, a sensitivity analysis demonstrated that FARSITE is sensitive to potential errors in the canopy structure input grids.  The results of this dissertation show that lidar can be used effectively to predict CBD and CBH for the purpose of fire behavior modeling and that investment in these lidar-based canopy structure data is worthwhile, especially for forests characterized by significant heterogeneity.  This work affirms that lidar is a useful tool for future canopy fuels inventory and mapping.

Chapter 1: Introduction

Canopy Fuels: The Evolving Need for Data

Recent years have been marked by severe fire seasons in the western United

States, especially the summers of 2000, 2002 and 2003.  According to the National Interagency Fire Center (NIFC) in 2004 77,534 fires burned 6,790,692 acres; these fires burned 315 primary residence dwellings, 18 commercial buildings and 762 outbuildings and the fire suppression costs for Federal agencies totaled over $880 million (  These fires have an enormous impact ecologically and economically (Butry et al. 2001; Graham et al. 2004).  They damage or kill a large portion of vegetation, produce large amounts of smoke and exacerbate soil erosion (Graham et al. 2004).  They also can cause significant economic devastation to an area, by destroying homes and businesses, negatively impacting recreational use of an area and burning marketable timber (Butry et al. 2001).   

It is now widely accepted that altered fuel loads due to past management practices (i.e. fire suppression over the last 60-70 years) have increased fire risk and promoted the occurrence of large and intense wildland fires (Weatherspoon 1996; Skinner and Chang 1996; Chang 1996).  For example, fire suppression combined with the selective felling of large trees has altered the state of Sierra Nevadan forests so that they currently are denser, smaller, more homogenous in structure and of a different species composition than they have been historically (i.e. pre-European) (McKelvey et al.  1996; Weatherspoon 1996).  These changes have caused an

increase in live and dead fuel loads and can make the affected stands more vulnerable to damage from fires (McKelvey et al. 1996; Weatherspoon 1996).  According to Keane et al. (2001) fuels are typically defined as the physical characteristics (e.g. loading, size and bulk density) of the live and dead biomass that affect the spread, intensity and severity of wildland fire.  

Awareness of changing risks has spawned growing interest in mapping fire hazard potential (Sapsis et al. 1996, Scott and Reinhardt 2001).  Of the different wildland fire types canopy fire behavior is of particular interest to forest managers because canopy fires are difficult to control, spread rapidly and their post-fire effects can be severe (Scott and Reinhardt 2001).  Furthermore, there has been an increase in the incidence of canopy fires in areas not typically susceptible to them in the past (Scott and Reinhardt 2001).  Though they are often addressed separately canopy fires are dependent on surface fires and typically occur when surface fuels are sufficient to ignite ladder fuels or the lower crowns of trees (Scott and Reinhardt 2001; Graham et al. 2004).

Accurate measurement of canopy fuel loads over large areas is essential for predicting fire hazard potential.  An understanding of the spatial distribution of wildland fuels is critical to evaluating fire hazard and risk over the landscape and to how management options should be prioritized (Chuvieco and Congalton 1989).  Forest managers’ overall picture of where canopy fires are likely to occur and how they might behave becomes clearer when they are provided with detailed fuels information.  This is because fire behavior is predicted based on three variables: fuels, weather and static topography.  While weather and topography are beyond their control, forest managers can exert influence over fuel loads (e.g. through prescribed fire and mechanical thinning).  Therefore, information about canopy fuel conditions can lead to better management decisions and mitigation practices.  Unfortunately, in many cases the quality of spatial data regarding fuels distribution needed to make informed decisions is simply not available (Keane et al. 2001; Miller and Landres 2004).  Miller and Landres (2004), for example, examine the results of a workshop designed to identify the information needs for wildland fire and fuels management.  They found that most managers described the availability of data regarding the fuels complex as well as fuels maps as ‘low’.  Furthermore, the authors found that while most managers rely on computerized tools or models for planning purposes the data needed to make full use of these tools is still lacking at the landscape scale.  

Because detailed quantitative fuels data are difficult to obtain generalized descriptions to the fuel load in a given area are applied (Keane et al. 2001); these descriptions are called ‘fuel models’ (Anderson 1982; Scott and Burgan 2005).  A fuel model essentially serves as a template of fuel distributions and loadings for a forest stand, focusing on surface fuels.  The fuel models are predefined (e.g. Anderson 1982) though custom models can be created for unique situations.  The model that best describes a forest stand is then used to represent that area.  However, applying the fuel models correctly requires a great amount of experience and skill.

Forest managers have various tools available to them to aid the development of policy and strategy regarding fire in the landscape, however such tools are typically dependent on accurate fuels data for them to yield good results.  Fire behavior models represent one such tool.  As policy has changed from one of total suppression of wildfires to letting fire resume a role in maintaining the health of forest ecosystems there has been a growing interest in improving knowledge of fire behavior – especially at landscape scales.  Understanding how fire behaves under different conditions is vital to developing appropriate management plans (van Wagtendonk 1996).  Fire behavior models have been developed to promote a better understanding of wildland fire and its effect on ecosystems.  Progress in computer technology has fostered the development of spatially explicit fire growth models which has significantly advanced fire management planning and decision making (Keane et al. 2001).  

Fire behavior models can be used to predict the behavior of ongoing wildfires and also to study the effects of potential mitigation strategies.  In the latter case, models can be used to test different fuel treatments (e.g. prescribed burning, biomassing, cutting and scattering) (van Wagtendonk 1996, Stratton 2004).  However, for the models to be used appropriately the existing fuels complex must be quantified.  This requires information about the amount and location (both horizontal and vertical) of available fuel in the canopy.  Canopy fuels are defined as the aerial live and dead biomass located within tree crowns (Keane et al. 2001).  Two canopy structure characteristics have been identified that help quantify these fuel loads: canopy bulk density (CBD) and canopy base height (CBH) and have been adopted for fire behavior modeling (Sando and Wick 1972, Scott and Reinhardt 2001). CBD is the mass of available canopy fuel per unit canopy volume and CBH is the lowest height in the canopy where there is sufficient fuel to propagate fire vertically into the canopy (Scott and Reinhardt 2001). This dissertation examines the use of largefootprint, waveform-digitizing lidar data to predict and create maps of CBD and CBH as well as the use of lidar-derived products to run a fire behavior model.  Lidar metrics are compared to field-based estimates of CBD and CBH and, based on the regression models resulting from these comparisons, maps of CBD and CBH are generated that are then tested as inputs into a fire behavior model.  

            The remainder of this introductory chapter is organized as follows.  The next sections provide (1) a background to fire behavior models, focusing specifically on FARSITE (Fire Area Simulator-Model); (2) a review of field- and remote-sensingbased collection of wildland fuels data; (3) a synopsis of the previous use of remotesensing-derived fuels data as input into fire behavior models; (4) a summary of the current state of the utilization of lidar to predict fuels and (5) an overview of the research objectives addressed in the remaining chapters. 


Background to Fire Behavior Modeling

Fire behavior has been studied for decades (van Wagner 1969; Albini 1976;

Rothermel 1972; Rothermel 1991; van Wagner 1977; van Wagner 1993; Xu and Lathrop 1994).  Early research studied surface fires independent of canopy fires while later studies explored the links between the two (Finney 1998; Scott and Reinhardt 2001).   Current research is leading to a better understanding as to how all the different factors affecting fire behavior are interlinked.  Advances in computer technology and increased speed with which large data sets and complex computations can be processed stimulated the development of fire behavior models (Keane et al. 2000; Keane et al. 2001).  However, these models are based on the fundamental principles governing fire behavior that were laid forth by earlier work (Finney 1998). 

Given uniform conditions (e.g. fuels, terrain, winds) fire is assumed to spread in a predictable pattern or shape (Finney 1998).  The most common shape model assigned to fire spread is the ellipse (Finney 1998).  Computer models have been developed that automate the application of fire shape models to realistic, non-uniform fire conditions (Finney 1998).  This is done by assuming that conditions at points along the fire perimeter are uniform (though point-to-point conditions can vary) and govern fire spread (Finney 1998).  One method of doing so is the vector approach to fire growth modeling – or Huygens’ principle.  In this approach the perimeter of the fire front is represented as a series of two-dimensional vertices (Finney 1998).