Forest Ecology and Management 281 (2012) 100–110

Contents lists available at SciVerse ScienceDirect

Forest Ecology and Management

journal homepage: www.elsevier .com/ locate/ foreco

Three-dimensional characterization of pine forest type and red-cockadedwoodpecker habitat by small-footprint, discrete-return lidar

L.S. Smart ⇑,1, J.J. Swenson, N.L. Christensen, J.O. Sexton 2

Nicholas School of the Environment, Box 90328, Duke University, Durham, NC 27708, USA

a r t i c l e i n f o

Article history:Received 19 January 2012Received in revised form 13 June 2012Accepted 14 June 2012Available online 17 July 2012

Keywords:Forest structureLidar remote sensingLongleaf pinePinus palustrisLoblolly pineRed-cockaded woodpecker

0378-1127/$ - see front matter � 2012 Elsevier B.V. Ahttp://dx.doi.org/10.1016/j.foreco.2012.06.020

⇑ Corresponding author. Tel.: +1 919 484 7857x128E-mail address: Lindsey_Smart@NatureServe.org (

1 Present address: NatureServe, 6114 Fayetteville R27713, USA.

2 Present address: Global Land Cover Facility, GeograMaryland, College Park, MD, USA.

a b s t r a c t

Accurate measurement of forest canopy structure is critical for understanding forest-wildlife habitat rela-tionships. Although most theory and application have been based on in situ measurements, imaging tech-nologies such as Light Detection and Ranging (lidar) provide measurements that are both verticallyaccurate and horizontally extensive. We use small-footprint, multiple-return lidar from a state-widedataset (1-m footprint, 0.11 point/m2) to characterize the vertical and horizontal structure of succes-sional loblolly pine (Pinus taeda) and mature, fire-maintained longleaf pine (Pinus palustris) forests onthe coastal plain of North Carolina, USA. The relationship between these characteristics and the feder-ally-endangered red-cockaded woodpecker’s (Picoides borealis, Vieillot) habitat preferences wereassessed; as this species has a strong affinity for mature longleaf pine forests. Vertical structure was char-acterized by lidar-derived metrics (e.g., average and standard deviation of canopy height) and horizontalpatterns of vertical structure were quantified by semivariograms and lacunarity analysis. Lidar metricswere compared with field measurements of stand structure and with woodpecker habitat use. We pre-dicted woodpecker distribution using the Maxent species distribution modeling algorithm with eleva-tion, landcover, and hydrography geospatial variables, with and without lidar-derived structuralvariables. Lidar successfully quantified canopy variation and differentiated between the structural char-acteristics of these two similar coniferous evergreen forest types (e.g. significant differences in maximumheight, canopy cover, and size classes). Loblolly stands were found to have the tallest trees on averagewith a higher canopy cover. Both semivariograms and lacunarity analyses clearly differentiated betweenevergreen forest structural types (e.g. semivariogram range was 18.7 m for longleaf, 32.3 m for loblolly).By examining the immediate area around cavity nesting sites we found taller trees than those foundacross broader foraging sites. The species distribution model accurately predicted woodpecker distribu-tion (tested with woodpecker presence, AUC > .85). The addition of lidar-derived variables improved themodel’s predictive power by 8% compared to the model based only on elevation, landcover, and hydrog-raphy environmental variables. We show that relatively low density lidar data are valuable for wildlifestudies by characterizing and separating similar canopy types, describing different use zones (foragingvs. nesting), and for use in species distribution models.

� 2012 Elsevier B.V. All rights reserved.

1. Introduction

Vegetation canopies are complex, three-dimensional volumes.The vertical and horizontal structure of vegetation characterizesphysiognomic types and biomes (e.g., savanna, shrubland, forest)and varies with plant community composition (e.g., Weldenet al., 1991; Kruger et al., 1997). Vertical and horizontal canopystructure also impact the distribution of wildlife (Matlack, 1994;

ll rights reserved.

.L.S. Smart).

oad, Suite 109, Durham, NC

phy Department, University of

Didham and Lawton, 1999). Because of their great mobility in bothhorizontal and vertical dimensions, birds are especially affected bystructural attributes in their selection and use of habitat (MacAr-thur and MacArthur, 1961; Erdelen, 1984; Desrochers and Hannon,1997; others).

Vertical canopy structure also directly influences the scatteringof radiation and therefore provides a potential means of relatingvegetation structure to wildlife habitat (Vierling et al., 2008). Pas-sive, optical remotely sensed imagery (e.g., Landsat Thematic Map-per) has been essential in the development of vegetation maps,habitat models, and site prioritization efforts, but these data pro-vide limited information on vertical structure and typically havehorizontal resolutions considerably coarser than the dimensionsof individual trees or stands of trees (Woodcock and Strahler,1987; Waring et al., 1995; Treuhaft et al., 1996; Lefsky et al.,

L.S. Smart et al. / Forest Ecology and Management 281 (2012) 100–110 101

2002; Vierling et al., 2008). Alternatively, fine-scale measurementsof forest structure from active sensors such as radar and lidar canbenefit forest inventories and characterizations of habitat for wild-life species influenced by forest structure (Chiao, 1996; Imhoffet al., 1997; Hyyppa et al., 2000; Lim et al., 2003).

Lidar (Light Detection and Ranging) provides fine-grain infor-mation about three-dimensional forest canopy structure and ishorizontally extensive (Maclean and Krabill, 1986; Treuhaft et al.,1996; Lefsky et al., 2002; Maltamo et al., 2005; Ørka et al., 2009).It has been used for quantifying many aspects of forest structurebecause of its accuracy and scalability. In some cases, lidar mea-surements can be even more accurate than interferometric syn-thetic aperture radar measurements with respect to in situ treemeasurements and also more accurate than field measurementsof heights themselves (Sexton et al., 2009). Using discrete-return,small-footprint lidar data and field measurements, Hall et al.(2005) estimated stand height, total above ground biomass, foliagebiomass, basal area, and other variables for a fire-prone ponderosapine (Pinus ponderosa, Dougl.) forests. Discrete-return lidar mea-surements have been successfully compared to field measures ofevergreen conifer forest height with RMSE (root mean square er-ror) of <1 m (Sherrill et al., 2008). In conifer forests, field-measuredstand attributes such as mean stand height, tree density, and basalarea were significantly correlated with lidar estimates (Hudaket al., 2006).

Lidar studies have been used to characterize habitat for can-opy-dwelling avian species (e.g. Hinsley et al. 2002; Hyde et al.2005, 2006; Broughton et al., 2006; Clawges et al., 2008; see re-view by Vierling et al., 2008; Marinuzzi et al., 2009; Mülleret al., 2010). However, most of these studies were based uponmore specialized lidar datasets with either full waveform abilityor very high point/posting density. Airborne lidar data was usedto examine differences in the structural components betweenhabitat and non-habitat for the marsh tit (Poecile palustris), findinga statistically significant difference of 1.6 m in canopy height be-tween occupied and unoccupied areas (Broughton et al., 2006).Optical and lidar measurements were compared to quantify struc-tural heterogeneity for predicting bird species richness in thePatuxent National Wildlife Refuge, Maryland. Canopy vertical dis-tribution measured by lidar was consistently found to be thestrongest predictor of species richness (Goetz et al., 2007). Anensemble decision tree modeling approach (random forests) wasused to assess lidar-derived metrics as predictors of habitat usefor the Neotropical migrant songbird, the Blackthroated Blue War-bler (Dendroica caenulescens). Results showed that canopy struc-ture variables consistently provided unique and complimentaryinformation that systematically improved model predictions(Goetz et al., 2010).

Counties, states and national governments are increasinglyusing lidar to produce precise bare-earth digital elevation models(DEMs) over large regions (Stoker et al., 2008), and there is consid-erable interest in a complete US national lidar data set (Ibid.). Cur-rently, these data are collected at a lower spatial density, and havebeen relatively less explored for forest structure characterizations.Future systems for acquisitions may have the capacity to collect athigher densities, providing increased accuracies but more chal-lenges in terms of analyzing such large datasets. Until this occurs,it’s important to explore the utility of the current low density data-sets for forest characterization. With lidar measurements at a den-sity of less than 1 point/m2, biomass and forest structure weresuccessfully analyzed in Wisconsin, USA (Hawbaker et al., 2009).Lidar-derived measurements were used at a density of0.11 point/m2 from the statewide NC Floodplain Mapping Program(NCFMP, 2008) lidar dataset for an analysis of tree height accura-cies in the Piedmont region of North Carolina (Sexton et al.,2009). As the first published analysis with this dataset, Sexton

et al. (2009), found this low density lidar dataset to be a good rep-resentation of loblolly pine (Pinus taeda) canopy height.

The accuracy, precision, and scalability of lidar are particularlyapplicable to management of the endangered red-cockaded wood-pecker (Picoides borealis, Epting et al., 1995; Rudolph et al., 2002),which occurs across the southeastern USA and is dependent onthe structure of rare old-growth longleaf pine forests (Ligonet al., 1986; Walters, 1991; US Fish and Wildlife Service, 2003).The longleaf pine communities upon which red-cockaded wood-peckers depend have a savanna canopy structure that is uniqueamong southeastern forests. Frequent fires in mature longleafstands maintain open, low-density stands of relatively large treescompared to surrounding stands of loblolly pine. The open, maturelongleaf pine woodlands provide old flat-topped pines that are re-quired for roosting and nesting. The open park-like characteristicof this habitat aids in prevention of nest predation on the red-cockaded woodpecker, promotes uninhibited flight paths, andprovides them with high quality foraging areas (Christensen,1981; US Fish and Wildlife Service, 2003).

Populations of red-cockaded woodpeckers have declined drasti-cally due to widespread conversion of old growth longleaf foreststo naturally-regenerated loblolly pine or short-rotation loblollypine plantations (Jackson, 1971; Lennartz et al., 1983; Ligonet al., 1986). Decreased fire frequency has also resulted in midstoryencroachment by shade-tolerant trees in many places (Beckett,1971; Hooper, 1988; Connor and Rudolph, 1989). Studies have sug-gested that hardwood midstory encroachment leads to site aban-donment by red-cockaded woodpeckers due to increases in nestpredation, negative effects on flight paths to active cavities, and de-creases in foraging habitat quality (Beckett, 1971; Jackson, 1971;Costa and Escano, 1989; Wood, 1983; Kelly et al., 1994; Connorand Rudolph, 1991). Woodpecker conservation efforts involvemaintaining and restoring the longleaf ecosystem as well as iden-tifying new candidate sites for woodpecker release (US Fish andWildlife Service, 2003).

Using the North Carolina Floodplain Mapping Program lidardataset, our goals in this study were to evaluate the ability oflow density multiple-return, small-footprint lidar at 0.11 point/m2 to characterize structural attributes in both vertical and hori-zontal dimensions for red-cockaded woodpecker habitat quality.Specifically, we: (1) assess the utility of lidar for detecting differ-ences between pine forest types by differences in their canopystructures and (2) examine the ability of lidar-based canopy struc-ture metrics to map the distribution of red-cockaded woodpeckernesting and foraging habitat.

2. Methods

2.1. Study area

The primary area of focus for this study was the 38,445-ha USMarine Corps Base at Camp Lejeune (MCBCL) located in Jackson-ville, North Carolina (Fig. 1). Camp Lejeune lies within the WhiteOak River Basin in Onslow County and encompasses habitat forseveral federally listed endangered, threatened, and rare species,including the red-cockaded woodpecker (US Marine Corps BaseCamp Lejeune, 2006). Forest management is an important compo-nent of the Marine Corps’ land stewardship responsibilities atCamp Lejeune, and restoration of historic native communities oflongleaf pine and breeding colonies of red-cockaded woodpeckeris especially important (US Marine Corps Base Camp Lejeune,2006). Restoration and maintenance of longleaf communities typ-ically involves thinning midstory trees and frequent prescribedfires (Brockway et al., 2005). Camp Lejeune also seeks to promoteconservation and compatible land use in the larger region outside

Fig. 1. US Marine Corps Base Camp Lejeune in Jacksonville, North Carolina.

102 L.S. Smart et al. / Forest Ecology and Management 281 (2012) 100–110

its borders (US Marine Corps Base Camp Lejeune, 2006), and so thelarger region was included for extrapolating woodpecker distribu-tion models.

The region’s forests are dominated by loblolly pine (P. taeda)and longleaf pine (P. palustris). Both species have similar heightsand crown structures but can differ in the age strata and horizontalpattern of canopy gaps. Structural differences between loblolly-and longleaf-dominated stands are due to differences in theirshade tolerance and demographic characteristics. Loblolly standsare typically even-aged and relatively dense, and they are oftenundergoing succession after agricultural abandonment or past log-ging. The mature longleaf stands preferred by the red-cockadedwoodpecker have more open canopies with larger gaps amongindividual trees or between small clumps of trees.

2.2. Data

2.2.1. In situ forest measurementsIn situ measurements of forest composition and structure were

recorded in 14 forest stands (seven dominated by longleaf pine andseven by loblolly pine) selected within Camp Lejeune (Fig. 1). Thesestands were selected from among stands in the MCBCL forest mon-itoring network to be representative of either mature longleaf pineforest or successional (40–70 year-old) loblolly pine forest and tobe sufficiently large to provide ample samples (4–8 ha) of lidarpoints (MCBCL, Natural Resource Management Division, unpub-lished data). These two different pine types cover approximately

half of the MCBCL land area (primary pine comprises approxi-mately 40% and pine/hardwood mixed forest comprises 15%). Ineach stand, all woody stems >1 cm DBH had been recorded by spe-cies and diameter in a permanently marked 0.1 ha plot betweenthe year 2003 and 2005 (methodology described in Peet et al.,1998). Estimates of stand basal area, tree size and tree diameterdistribution were based on these data. The longleaf stands werepure, whereas loblolly-dominated stands had extensive hardwoodmidstories.

Additional analyses were performed on longleaf and loblollystands derived from forest stand areas that are updated continuallyby Forestry Program personnel at Camp Lejeune (last updated in2005). From these stand areas, 16 longleaf pine 0.1 ha plots and48 loblolly pine 0.1 ha plots were randomly selected with equalinclusion probabilities. The primary species within the standsneeded to either be longleaf pine or loblolly pine with no co-dominants in order to be selected. Measures within these standswere recorded by the Forestry program and included pine basalarea, age, age class, site index, quality, and management method,

2.2.2. Red-cockaded woodpecker locationsRed-cockaded woodpecker locations at cavity trees and foraging

sites were recorded by GPS across Camp Lejeune from 1990 to2005. Cavity sites are nesting sites for the red-cockaded wood-pecker. They are recorded by personnel from the Threatened andEndangered Species Program at Camp Lejeune. An active cavitytree is a tree that exhibits fresh resin as a result of red-cockaded

L.S. Smart et al. / Forest Ecology and Management 281 (2012) 100–110 103

woodpecker activity or a tree at which one or more red-cockadedwoodpeckers have been observed. All potentially active cavitiesare checked for evidence of red-cockaded woodpecker activity.Potentially active cavities are all cavities that have been activewithin the last 5 years and all inactive cavities that have under-gone restoration of appropriate habitat structure and/or cavityinstallation within that time. All cavities that have been activewithin the last 5 years are evaluated until an active cavity is lo-cated or birds are observed. If cavities are inactive that were previ-ously active, a thorough search for new cavity trees is conducted insuitable habitat within 0.4 km (0.25 mi). There is high site fidelityamong nest cavities and therefore, many cavity sites remain ‘‘ac-tive’’ for several years. The data is updated quarterly and the lastupdate occurred in 2005. A total of 735 nesting cavities were iden-tified and buffered by 40 m (vegetation composition is critical be-tween 17 and 66 m; approximately 40 m on average) for thecomparison with lidar-derived structural metrics. A total of 547cavity sites occurred within the White Oak River sub-basin andwere used for the species distribution modeling.

A total of 51 foraging areas were used for the comparison withlidar-derived structural metrics. Foraging areas are delineated foreach active cluster site. An active cluster consists of several cavitysites. The cluster sites are determined by the Threatened andEndangered Species Program personnel. Foraging sites occur with-in these foraging areas. Foraging sites depict sites, where specieshave been observed in foraging function. These observations usu-ally occur within a 5–8 h period within a single day. These foragingsites do not preclude bird appearances in other areas within theirforaging partitions. Foraging partitions or foraging areas, whichwere used in this analysis, were generated through a process incor-porating buffers with Thiesson polygons by the Threatened andEndangered Species Program personnel. These were last updatedin 2005 and any timber removal within these areas requires con-sultation with the USFWS (see Supplemental Fig. 1 for a red-cockaded woodpecker data collection schematic).

2.2.3. Lidar measurements of forest structureThe N.C. Floodplain Mapping Program lidar dataset was col-

lected between 2000 and 2003 to create updated digital elevationmodels (DEMs) for flood hazard mapping and floodplain manage-ment (North Carolina Flood Mapping Program, 2008), with the On-slow County portion acquired in February of 2001. To favor thecollection of ‘‘bare earth’’ data for DEM creation, the data were col-lected during leaf-off conditions. Data were collected with a dis-crete-return sensor with four returns per pulse, from a platformaboard an aircraft operated by a commercial vendor. The pointdensity was approximately 0.11 points/m2. The first (top of can-opy) returns were processed to create three-dimensional ‘pointclouds’ of planimetric coordinates (x,y) and elevation values (z)for the New River sub-basin.

A canopy surface model and ground surface model were createdby calculating the maximum and minimum values across a movingwindow and creating continuous surfaces from these values. Theground surface model and canopy surface model were differencedto produce a surface of canopy heights. An output resolution of 3 mwas used to preserve the original sampling density. To estimatecanopy cover for the 14 plots, the 3-m resolution canopy heightmodel was reclassified to create a binary image of gaps and trees.A threshold of 2.1 m (7 ft) was chosen to separate gaps from can-opy based on the Red-cockaded Woodpecker Recovery Plan’s guidefor managing longleaf forest (US Fish and Wildlife Service, 2003),which suggests that trees smaller than 2.1 m (7 ft) do not inhibitred-cockaded woodpecker foraging activities. For the plot-levelanalysis, all pixels were averaged to obtain a single proportion ofcanopy cover for each plot. Using the canopy height model, the li-dar heights for all 14 plots were divided into three general height

categories: short: 3.0–9.1 m (10–30 ft), medium: 9.4–16.8 m (31–55 ft), and tall: 17.1–30.5 m (56–100 ft) and proportions of eachheight class were calculated for the loblolly and longleaf stands.

2.3. Analysis

2.3.1. Characterization of vertical structure and horizontal patternWe compared lidar-derived metrics to in situ measurements of

forest structure and used the lidar-derived metrics to comparestructural attributes of longleaf- vs. loblolly-dominated pinestands. Lidar-derived layers were correlated with field measure-ments of forest stand structure (stand age, basal area, and site in-dex), and the distribution of heights and their relative skewacross sample plots of different forest types were examined to dis-tinguish among forest structures and to characterize the variabilityof vertical foliage distribution, respectively. We had a larger set offield sites (n = 32), where we calculated more general variables ofheight and canopy cover, and a smaller field data set (n = 14),where we calculated the more intensive spatial statistics describedbelow.

We compared horizontal canopy-height patterns of loblolly vs.longleaf pine stands in 1-ha image subsets using semi-variance.Lacunarity analyses (Allain and Cloitre, 1991; Frazer et al., 2005)were used to quantify the horizontal heterogeneity found in theupper surface of the forest canopy. Calculating lacunarity requiresthe application of a gliding box or ‘moving window’ algorithm overbinary data to describe the distribution of gap sizes in a fractal se-quence. Geometric objects appear more lacunar if they contain awide range of gap sizes. Lacunarity statistics were calculated forall stands. The lacunarity statistic is a simple integrated measureof cross-scale spatial heterogeneity, computed as the sum of thenormalized lacunarity statistics estimated at each discrete box sizer: kTOTAL ¼ ð1=kð1ÞÞRkðrÞ, where kðrÞ is the lacunarity statistic com-puted at the various box sizes and k (1) is the lacunarity statisticestimated for box size r = 1. Small values of kTOTAL < 3:25 generallyindicate randomness in canopy heights at coarser spatial scaleswhile larger values of kTOTAL > 4:25 denote the presence of spatialstructure at coarser scales (Frazer et al., 2005).Examination of thegraphs produced by an analysis of lacunarity reveals the overallfraction of the study area occupied by the canopy, the level of con-tagion between trees at a particular scale, the scale at which thestudy area approximates a random pattern, and the range of scalesover which a study area exhibits self-similarity (Plotnick et al.,1993).

2.3.2. Canopy structure for red-cockaded woodpecker habitatWe summarized forest structure at observed red-cockaded

woodpecker localities and compared structural characteristics incavity sites vs. foraging areas and then modeled the woodpecker’sgeographic distribution from these field locations based on spatialdata layers across the study area. Based on recommendations ofthe Red-Cockaded Woodpecker Recovery Plan (US Fish and Wild-life Service, 2003), structural variables chosen for inclusion in mod-els were: canopy cover, canopy height, and the standard deviationof height; these variables were measured within 3-m pixels andresampled to 30-m resolution to correspond to other environmen-tal layers. Other variables included (all at 30-m resolution): the de-tailed North Carolina Gap Analysis land cover map which includesland units (McKerrow et al., 2006) containing 59 natural and semi-natural vegetation categories, the more general 2001 National LandCover Dataset (NLCD; Homer et al., 2007; Vogelmann et al., 1998)the National Elevation Dataset (US Geological Survey SeamlessData Warehouse, URL: http://seamless.usgs.gov/ned13.php), anddistance-to-water based on the 1:100,000-scale National Hydrog-raphy Dataset (US Geological Survey, http://nhd.usgs.gov/da-ta.html). Correlations between the continuous predictor variables

104 L.S. Smart et al. / Forest Ecology and Management 281 (2012) 100–110

were evaluated before modeling. To model woodpecker distribu-tion, we used the Maximum Entropy algorithm Maxent (Phillipset al., 2006). The algorithm assumes no parametric distributionof the data and has been tested successfully in the past (Elithet al., 2006; Franklin, 2009). Maxent as a ‘presence-only’ methodis an appropriate choice because no woodpecker absence datawere recorded for the study area. We relied on the heuristic anal-ysis of variable contributions generated by Maxent, permutationimportance, and the jackknife procedures for an estimation of the‘‘gain’’ of each input variable. The jackknife procedures provideestimates by excluding each variable in turn and creating a modelwith the remaining variables to identify critical environmentalvariables for the species.

We constructed several different variable combinations in orderto test the effect on model prediction accuracy. We first selectedgeospatial variables that are available across all locations of theUS and will be referred to as the ‘DEM-NLCD-Hydro’ model (Eleva-tion, NLCD land cover and hydrography). Elevation and hydrogra-phy are ecologically meaningful because the red-cockadedwoodpecker is affiliated with upland longleaf pine savannas andnot lower elevation riparian areas or seepage swamps. The NLCDprovides coarse land cover classes, of which some are importantfor the red-cockaded woodpecker. We then successively addedadditional variables: the ecologically detailed GAP analysis mapand the lidar-derived information on canopy structure. For eachunique variable combination we ran 15 replicate (permutation)models, where 30 random woodpecker presence points were with-held for each run. Comparisons among different Maxent runs withdifferent variable combinations were based upon the ROC (Recei-ver Operating Characteristic) and AUC (Area Under the ROC Curve)(Phillips et al., 2006).

Each model’s performance was evaluated for its ability to accu-rately predict suitable habitat for the test data. As well, the predic-

Table 1Summary statistics for stands of two different pine types at U.S. Marine Corps Base Camp

Longleaf pine stands

(n = 32) Height (max.) 19.0 mHeight (mean) 8.13Standard deviation (Ht) 5.79Canopy cover (n = 14) 40.0%Percentage of small-sized trees 12.7%Skew (50th percentile of heights) 0.6 m Or lessLacunarity Statistic <3.25 (Average = 2.77)Semivariogram range 18.7 m

Fig. 2. Density distributions of canopy heights for the seven longleaf pine stands (a) and swithin the stands.

tor variable contributions were recorded along with permutationimportance. These were ranked according to the influence on thecontribution of unique information to the model. The same wasdone for relative contributions to the model sensitivity and AUC.The highest ranking or contributing variables were retained forthe final model.

3. Results

3.1. Vertical and horizontal structure of coniferous canopies

The 14 plots had bimodal height distributions, sharing one peakat �0 m corresponding to the ‘‘ground return’’ (i.e., canopy gaps)and exhibiting another peak corresponding to ‘‘top-of-canopy’’heights, which varied among plots (Fig. 2). The strength of theground return was occasionally weakened by the dense, impene-trable canopies characteristic of unthinned loblolly pine stands.Overall, longleaf pine stands had higher numbers of ground returnsor open canopy gaps in comparison to loblolly dominated stands.

The seven longleaf pine plots were more similar to one anotherin their height distributions than the seven loblolly pine plots (Ta-ble 1). Longleaf pine stands typically had low densities of tall treesand consistent top-of-canopy peaks between 12 m and 24 m,whereas loblolly-dominated stands were characterized by densercanopies and more heterogeneous height distributions. Longleafstands had lower canopy cover and higher coefficients of variationof height in comparison to loblolly-dominated stands (Table 1).

Semivariogram and lacunarity analyses indicated fine-scalehorizontal patterns within longleaf stands that were largely absentin loblolly stands. Loblolly stands exhibited horizontal patternsthat only became apparent at coarser measurement scales. Differ-ences in the variogram ranges of loblolly and longleaf pine heights

Lejeune, Jacksonville, N.C.

Loblolly pine stands P-Value (where applicable)

24.54 m <0.00510.89 0.026.61 0.25755.1% 0.0737.5% 0.0477.9 m Or less n/a>4.25 (Average = 3.31) n/a32.3 m 0.044

even loblolly stands (b). The peaks at zero signify the ground returns or canopy gaps

Fig. 3. Log–log plot of normalized lacunarity values (K(r)) versus box size (r). Solidcurves depict loblolly plots, dotted curves depict longleaf plots. The black curve wascreated from a randomized model to represent complete spatial randomness (CSR).

Table 2Pearson’s product moment correlation coefficients for field measures (rows) versusLidar-derived metrics (columns). Pine and hardwood basal area is measured in ft2/acre. Compiled at 32 field sites (12 longleaf, 20 loblolly); unless otherwise noted.

Fieldmeasurements

Meanheight

Maximumheight

Standard deviationof heights

Canopycover

Stand age 0.038 0.2876 0.270 0.135Site index 0.543** 0.589** 0.422** 0.328**

Pine basal area 0.029 �0.408** �0.335* 0.050Hardwood

basal area0.5303** 0.6372** 0.3370* 0.5065**

* Significant at p-value = <0.10.** [Bold] = Significant at p-value <0.05.

Table 3Comparisons and tests of significance for lidar-derived metrics in red-cockadedwoodpecker foraging areas and longleaf cavity areas within Camp Lejeune.

Attributemeasured byLidar

LongleafCavity areas(n = 735)

Foragingareas(n = 51)

Difference between cavityand foraging areas P-value

Maximum treeheight (m)

16.36 25.4 <0.001

Average treeheight (m)

6.24 5.71 0.05

Standarddeviation ofheights (m)

3.81 4.54 <0.001

Coefficient ofvariation ofheights

0.67 0.83 <0.001

Average canopycover

0.25 0.24 0.33

Standard dev.canopy cover

0.20 0.27 <0.001

L.S. Smart et al. / Forest Ecology and Management 281 (2012) 100–110 105

were significantly different (p-value = 0.04), median ranges of18.5 m for longleaf stands and 35 m for loblolly stands.

Lacunarity statistics were calculated for all 14 plots. Lacunarityhas been used to quantify textural patterns in single trees (Martenset al., 1993) and to monitor change in rainforest landscapes (Wei-shampel et al., 1998). More recently it has been used with lidar todetermine the spatial pattern of gaps in forest canopies (Weisham-pel et al., 2000; Morsdorf et al., 2006). The lacunarity results in thisanalysis suggested that all longleaf stands, with the exception ofone, had lower lacunarity (<3.25) than loblolly stands (>4.25) indi-cating that gaps in mature longleaf stands were more randomlydistributed and lacked a wide range of gap sizes. Lower lacunaritymeasures and those that fall near the line of complete spatial ran-domness generally indicate randomness in canopy heights at coar-ser horizontal scales. It suggests that the pattern is approachingself-similarity; where the coarse-scale patterns are the same asthe fine-scale patterns, Loblolly pine stands have more variabilityin lacunarity curves. There are some stands above the completespatial randomness curve, suggesting that there are large clumpsof canopy separated by large clumps of non-canopy or gaps anddifferences between fine-scale patterns and coarse-scale patterns.There are also two loblolly stands that fall near the line of completespatial randomness which indicates that these stands have a morerandom pattern of gaps and canopy, similar to that of longleaf pinestands. Longleaf pine stands have a more narrowly defined patternthat is visible at all scales and loblolly pines have a more variablepattern of canopy.

Lacunarity curves suggest that loblolly stands exhibit greatervariability in their gap distribution at varying scales, though onaverage, show a tendency toward coarser horizontal structuring(e.g. clumping or aggregation of similar values) (Fig. 3). Lacunaritycurves for longleaf stands show much less variability, as evidencedby the proximity of longleaf stand curves to one another, and havea distribution of patches that can be quantified at finer scales. Thelongleaf pine pattern was not random but was closer to the ‘‘com-plete spatial randomness’’ (CSR) curve than most loblolly pine pat-terns. This suggests that longleaf pine pattern can be described as amore randomly distributed pattern of canopy patches compared tothe more clumped patterns exhibited by loblolly pine plots. Fur-ther details canopy structural results can be found in Smart (2009).

With a wider selection of field plots (n = 32), we found lidar suf-ficiently characterized canopy structure (Table 2). This analysiswas limited however because height measurements were not re-corded in the field, and therefore we were constrained to comparelidar height –based measurements to stand age, basal area and siteindex. We did however, find that maximum height as measured bylidar was explained statistically with the field variables of standage, pine basal area and site index (Adj R2 = 50.63; p-value < 0.001;MaxHt = �4.01 + 0.261 � Age � 0.201 � PINE_BA + 1.016 � SITE_INDEX). Lidar maximum stand height was highest correlated withsite index and both pine and hardwood basal area and (r = 0589,0.408, and 0.637, respectively), while lidar canopy cover was corre-lated with site index and hardwood basal area (0.328 and 0.507,respectively). Site index was correlated with lidar maximum andmean heights (r = 0.45 and 0.33 respectively).

3.2. Canopy structure for red-cockaded woodpecker habitat

Red-cockaded woodpecker nesting and foraging sites wereassociated with several structural attributes detectable by lidar.Overall, cavity and foraging areas are both very open, with �25%canopy cover, although foraging areas have slightly more variablecanopy height (p < 0.001) and cover (p < 0.001) than nesting areas(Table 3). Forests immediately surrounding nesting sites were tal-ler on average than those surrounding foraging areas, although thetallest trees in foraging areas were significantly taller than those incavity areas, as shown by averaged maximum heights (p < 0.001).

All of our distribution models produced reasonable and accu-rate models (AUC = 0.87–0.93). When combined with a detailedland cover map, the lidar-derived metrics of canopy cover andstandard deviation of heights contributed more to the model interms of ‘gain’ than the other ‘DEM–NLCD–Hydro’ variables

Table 4Relative contributions of each environmental variable to the Maxent model: DEM–NLCD–Hydro + GAP + Lidar. Interpretation follows Phillips et al. (2006). Shown arepercent contribution of each variable to the model (depending on the path the modeltakes to get to the optimal solution) and permutation importance of the 15 modelruns (depends on the final model, not the path the model used).

Variable Percentcontribution

Permutationimportance

GAP land-cover 32.5 20.1Distance to streams 20.3 21.3Canopy cover (30ft) 14.9 13.4Elevation 13.7 25.6NLCD land-cover 9 6.9Standard deviation of

heights4.8 10.1

Maximum heights 4.8 2.6

Table 5Species distribution modeling accuracy results based on the AUC with different modelvariable inputs. The four models used for comparison were the (1) DEM–NLCD–Hydro, (2) DEM–NLCD–Hydro + GAP land cover map, (3) DEM–NLCD–Hydro + lidar-derived variables (canopy heights, canopy cover, and standard deviation of heights),and (4) DEM–NLCD–Hydro + GAP + lidar-derived variables. ROC curves for all uniquemodels can be found in Supplemental Figures.

Model AUC AUC difference

DEM–NLCD–Hydro 0.868 0.045DEM–NLCD–Hydro + Lidar 0.913DEM–NLCD–Hydro 0.868 0.046DEM–NLCD–Hydro + Gap 0.914DEM–NLCD–Hydro + Lidar 0.913 0.001DEM–NLCD–Hydro + GAP 0.914DEM–NLCD–Hydro + GAP 0.914 0.020DEM–NLCD–Hydro + GAP + Lidar 0.934

106 L.S. Smart et al. / Forest Ecology and Management 281 (2012) 100–110

(DEM, NLCD Land cover, and distance to streams). For the modelthat included all variables, the Gap Analysis land cover map andpercent canopy cover map contributed 32.5% and 14.9%, respec-tively (Table 4). Based upon multiple model permutations andthe use of woodpecker presence test data, we found that the addi-tion of canopy structure and/or the GAP land cover map improvedaccuracy (measured by the AUC) of predictions over DEM–NLCD–Hydro model variables (Table 5). The addition of either theGAP land cover map or canopy structure to the DEM–NLCD–Hydromodel improved prediction accuracy by 5%. When both lidar can-opy structure and the GAP land cover map were included withthe DEM–NLCD–Hydro model variables, predictive success in-creased 8% from the DEM–NLCD–Hydro model.

The geographical pattern of the woodpecker model including allvariables (DEM–NLCD–Hydro model, with canopy structure andGap land cover) revealed that the majority of predicted wood-pecker habitat occurred on the military base, with isolated patchesof suitable habitat to the north and northeast (Fig. 4). This coin-cides with the land use in the rest of the sub-basin—mainly rowcrops, pasture, or residential, with some managed pinelands andnatural ecosystems to the northeast. Isolated patches of suitablehabitat were predicted in the Hoffman Forest, which is mostlypocosin which is a palustrine wetland with extensive shrub coverand contains pond pine (Pinus serotina) as the dominant canopytree but does include some disturbed longleaf pine flatwoodswhich may have potential to serve as red-cockaded woodpeckerhabitat (Fig. 4).

4. Discussion

4.1. Lidar characterization of vertical structure and horizontal pattern

Low density lidar imaging can discriminate the three-dimensional structural attributes characteristic of dense loblollypine forests versus mature longleaf pine savannas. These structuraldifferences result from natural and artificial (i.e., management-related) processes impacting growth and age class distributions,as well as variation of the horizontal patterns of gaps within thestands. While canopy height—commonly extracted from lidar mea-surements—is an important tool for forest inventory and assess-ments, we found that with our analysis methods and datadensity, it alone could not discriminate between management re-gimes, forest structural types, or species though others have foundvarious lidar metrics indicative of successional status and struc-tural type (Falkowski et al., 2009). It is important to quantify ver-tical structure and horizontal heterogeneity with a combinationof canopy-height and gap metrics that are indicative of the speciesof interest (e.g. Fig. 5). We found this to be the case for these twopine species that have similar individual (i.e., crown) structural

characteristics and comparable maximum heights, and whichwould therefore be difficult to distinguish relying solely on maxi-mum height.

Horizontal patterns were most effectively quantified by metricsthat characterized the heterogeneity of canopy cover, e.g., horizon-tal standard deviation and coefficient of variation of heights. Scale-dependent measures of landscape pattern—i.e., lacunarity—wereespecially important in quantifying the distribution of canopypatches in ways that proportional canopy cover metrics alonecould not. For management, scale-dependent measures of land-scape pattern such as lacunarity could be used to characterize can-opy gaps or heterogeneity gradients in order to restore these sitesto the longleaf pine savannas that were historically characteristicof North Carolina’s coastal plain.

4.2. Canopy structure for wildlife habitat

Habitat use by many avian species is determined by forest com-munity composition and three-dimensional vegetation structure.Our results corroborate previous field-based studies relatingwoodpecker habitat to the structural components of forest cano-pies (Ligon et al., 1986; US Fish and Wildlife Service, 2003). Forthe red-cockaded woodpecker, mature longleaf pine stands arewidely recognized as optimal nesting habitat (US Fish and WildlifeService, 2003). Red-cockaded woodpeckers have been known toforage and even occasionally to nest in open, old-growth loblollystands (US Fish and Wildlife Service, 2003). Such stands could bedistinguished from dense loblolly stands with the methods pre-sented here.

Many bird species in temperate forests recognize different veg-etation layers and gap patterns and select habitat accordingly (e.g.MacArthur et al., 1962; Rosenberg and Anthony 1992; Mills et al.,1993; Newell and Rodewald 2011). These woodpeckers in particu-lar depend on the old growth characteristics of southeastern pineforests (Engstrom and Sanders, 1997), namely tall open canopies,which we were able to capture with lidar. Additional old growthcharacteristics include canopy gaps, patchy understories and oldtrees (larger height classes), which we detected by variation of can-opy heights at local sites as well as across the watershed.

We found that woodpecker habitat use was associated with un-ique aspects of forest canopy structure characterized by lidar. Byfocusing on structural characteristics at nest cavities and foragingsites, we found similarities and differences among the two useareas. Woodpecker foraging areas had greater maximum heightmeasures than the cavity areas. This is most likely the result ofthe species’ preference for very open loblolly stands for foraging;loblolly pines have greater maximum heights than longleaf pines(US Fish and Wildlife Service, 2003). Habitat requirements aremore specific for nesting cavities than for foraging areas, and

Fig. 4. Map of predicted distribution of red-cockaded woodpecker habitat which included all variables (DEM-NLCD-Hydro + GAP + Lidar; AUC = 0.934).

L.S. Smart et al. / Forest Ecology and Management 281 (2012) 100–110 107

therefore we would expect a greater variability in both heights andcanopy cover in foraging areas, as supported by our results. Bothcavity and foraging sites exhibited very similar open canopies(�25% cover averaged across sites). Red-cockaded woodpeckersappear to be sensitive to the horizontal distribution of canopyand understory patches, as represented by scale-dependent mea-sures of lacunarity and correlation length. The longleaf stands ana-lyzed had a substantial proportion of overstory patches, but these

patches—though connected—are interspersed with the understoryand have been shown to be very important for woodpeckers (USFish and Wildlife Service, 2003). The loblolly stands were muchmore variable and tended to show coarser-scale horizontal pat-terns rather than the fine-scale interspersion of over- and under-story patches prevalent in longleaf pine stands. For the loblollystands, there were large patches of overstory separated by equallylarge patches of understory. Some of the loblolly stands had almost

Fig. 5. Graphic displays of typical horizontal patterns for two example pine stands, loblolly at left, longleaf at right. (a) Pine overstory (grey) and gaps (black) and (b) canopyheight profiles.

108 L.S. Smart et al. / Forest Ecology and Management 281 (2012) 100–110

completely closed canopies with few gaps and understory. Thesepatterns are not suitable for red-cockaded woodpecker, especiallynot for foraging behaviors that require sufficient open areas as wellas a connected canopy patch distribution. Our results support longheld ecological theory on the importance of horizontal heterogene-ity for avian species (e.g. MacArthur et al., 1962; Roth, 1976) andcomplement recent work focusing on fine-scale canopy detailsmodeling snags, and understory for woodpecker habitat (e.g. Mari-nuzzi et al., 2009).

The addition of indices of landscape pattern and verticalstructure improved the accuracy of the species distribution mod-els compared to models using only elevation, the National Land-cover Database dataset, and hydrography geospatial layers.Lidar-derived variables, when combined with all other geospatialvariables, contributed 28.4% towards the woodpecker distribu-tion model and improved the overall model accuracy. Of thelidar-derived indices, canopy cover was the greatest contributorto the model goodness-of-fit, followed by standard deviation ofheights and then canopy heights. When all variables wereincluded in the model, the Gap Analysis land-cover map contrib-uted the most unique information to the distribution modelbecause of its delineation of various evergreen forest types, spe-cifically longleaf pine, which is strongly associated with red-cockaded woodpecker presence. Although a region-wide spatiallyexplicit dataset for vegetation classification, the NC Gap dataset

has a maximum accuracy of 60–70% for coarser vegetation classgroupings, and lower for finer vegetation classes (McKerrowet al., 2006). Of the Map’s 69 mapped land cover categories, 59are of natural or semi-natural vegetation types. The NC Gapdataset is one of the highest currency and quality, but this isnot available everywhere in the US and therefore Gap maps forother areas may not be as reliable nor contribute as substantiallyto wildlife habitat modeling (see McKerrow et al., 2006 for de-tailed NC accuracy assessment results and see regional Gap Anal-ysis projects such as Lowry et al. (2007)). Lidar datasets supplyinformation that is lacking from the Gap datasets, and this un-ique data can complement Gap datasets in analyses, both forNorth Carolina and nation-wide.

The geospatial results of the species distribution model showedthat most of the potential habitat for the red-cockaded wood-pecker occurred within Camp Lejeune. This is consistent with theland-use in the rest of the sub-basin which is mostly agriculturaland residential. This emphasizes the importance of Camp Lejeuneand their involvement in natural resource management for theconservation and restoration of the longleaf pine habitat. Their ef-forts to restore converted land to natural longleaf pine with thin-ning and prescribed burns are essential for the conservation ofthe red-cockaded woodpecker because the Marine Corps Baseserves as a primary area of intact longleaf pine habitat within thecoastal plain. Their efforts at maintaining suitable longleaf pine

L.S. Smart et al. / Forest Ecology and Management 281 (2012) 100–110 109

habitat are evident in the canopy structure metrics and the speciesdistribution model.

In our case, lidar canopy metrics provided more detail on thefine-grain pattern of habitat suitability than could be obtainedfrom standard geospatial variables. Lidar provides continuouscharacterization of canopy structural traits across a landscapeand therefore is capable of characterizing the canopy within a gi-ven vegetation class of even a relatively detailed categorical vege-tation map such as the Gap Analysis land cover map. To improveregional mapping efforts for various applications, there is increas-ing interest to incorporate canopy structure from lidar (Gouldet al., 2010).

Additional analysis of lidar data for the woodpecker could focuson the detection of key structural components such as potential cav-ity trees and hardwood encroachment. Though our study was re-stricted to conifers, recent work (e.g., Hawbaker et al., 2010) hasshown the value of low-density data for leaf-off mapping of timbervolume across wide areas. Further research could also be directedtoward the integration of fine-scale canopy attributes from high-density lidar with that of large area - low density lidar. Also, the syn-thesis of spectral indices from optical imagery and lidar may be ben-eficial for large scale analyses as it has been shown that lidar-derived heights are associated with different spectral indices (Pasc-ual et al., 2012). At the regional scale, further work could focus onthe development of a ‘wall-to-wall’ characterization of the NC coast-al pine ecosystem to detect structural patterns preferred by red-cockaded woodpecker. Mapping and management of other can-opy-dependent species could also benefit from regional analysisacross the state of North Carolina (�135,000 km2). Understandingthe habitat requirements of multiple species is leading towards abroader understanding of biodiversity patterns with respect to can-opy structure (e.g. Goetz et al., 2007). The increasing availability oflidar data and its capacity to provide unique information about hab-itat structure will improve accuracy and quality of habitat modelingand mapping, and will potentially replace many labor-intensive,field-based measurements across broad areas.

Acknowledgements

Lidar data were provided by the North Carolina Floodplain Map-ping Program. We thank D. Urban for his support and assistancewith landscape metrics, and A. Hudak, for technical advice on lidarprocessing. We appreciate the cooperation of the Forestry andWildlife Resources Staff at the Marine Corps Base at Camp Lejeune,Dr. Susan Cohen and Dr. Stephen Mitchell in providing access tofield data. We would also like to thank the reviewers that providedcareful review and extensive comments that greatly improved thequality of this paper.

Lidar manipulation and geospatial analyses were performedusing ArcGIS and ArcScene (Environmental Systems Research Insti-tute, 2008) for three-dimensional viewing of geospatial data, andthe Hawth’s Tools extension for ArcGIS (Beyer, 2004). Landscapemetrics and lacunarity measures were generated in QRULE (anextension of RULE, Gardner, 1999) and FragStats (McGarigalet al., 2002). Species distribution models were generated usingMaxent (Version 3.2.19, Phillips et al., 2006). All statistical analysesas well as analyses of semi-variance and autocorrelation were per-formed using R statistical computing software (R DevelopmentCore Team, 2009).

Appendix A. Supplementary material

Supplementary data associated with this article can be found, inthe online version, at http://dx.doi.org/10.1016/j.foreco.2012.06.020.

References

Allain, C., Cloitre, M., 1991. Characterizing the lacunarity of random anddeterministic fractal sets. Phys. Rev. Lett. 44, 3552–3558.

Beckett, T., 1971. A summary of red-cockaded woodpecker observations in SouthCarolina. In: Thompson, R.L. (Ed.), Symposium on the Red-CockadedWoodpecker. Bureau of Sport Fisheries and Wildlife and Tall TimbersResearch Station, Tallahassee, Florida, pp. 87–95.

Beyer, H.L., 2004. Hawth’s Analysis Tools for ArcGIS. (Extension for ArcGIS Version9.3). <http://www.spatialecology.com/htools>.

Brockway, D.G., Outcalt, K.W., Tomczak, D.J., Johnson, E.E., 2005. Restoration ofLongleaf Pine Ecosystems. General Technical Report SRS-83. USDA ForestService Southern Research Station, Asheville, NC, USA. <http://www.srs.fs.usda.gov/pubs/gtr/gtr_srs083.pdf>.

Broughton, R.K., Hinsley, S.A., Bellamy, P.E., Hill, R.A., Rothery, P., 2006. Marsh titPoecile palustris territories in a British broad-leaved wood. Ibis 148 (4), 744–752.

Chiao, K.M., 1996. Comparison of three remotely sensed data on forest crown closureand tree volume estimations. Int. Arch. Photogramm. Rem. S. (B), 123–130.

Christensen, N.L., 1981. Fire regimes in southeastern ecosystems. In: Mooney, H.A.,Bonnickson, T.M., Christensen, N.L., Lotan, J.E., Reiners, W.A., (Eds.), Fire Regimesand Ecosystem Properties. General Technical Report WO-26. U.S. Department ofAgriculture., Forest Service, Washington, DC, pp. 112–130.

Clawges, R., Vierling, K., Vierling, L., Rowell, E., 2008. The use of airborne lidar toassess avian species diversity, density, and occurrence in a pine/aspen forest.Rem. Sens. Environ. 112, 2064–2073.

Connor, R.N., Rudolph, D.C., 1989 Red-cockaded woodpecker colony status andtrends on the Angelina. Davy Crockett, and Sabine National Forests, USDA ForestService Research Paper SO-250, New Orleans, LA.

Connor, R.N., Rudolph, D.C., 1991. Effects of midstory reduction and thinning in Red-cockaded Woodpecker cavity tree clusters. Wildlife Soc. B. 19, 63–66.

Costa, R., Escano, R.E.F., 1989. Red-cockaded woodpecker: Status and Managementin the Southern Region in 1986. Tech. Publ. R8-TP 12. U.S. Department ofAgriculture., Forest Service, Atlanta.

Desrochers, A., Hannon, S.J., 1997. Gap crossing decisions by forest songbirds duringthe post-fledging period. Conserv. Biol. 11, 1204–1210.

Didham, R.K., Lawton, J.H., 1999. Edge structure determines the magnitude ofchanges in microclimate and vegetation structure in tropical forest fragments.Biotropica 31, 17–30.

Elith, J., Graham, C.H., et al., 2006. Novel methods improve prediction of species’distributions from occurrence data. Ecography 29, 129–151.

Engstrom, R.T., Sanders, F.J., 1997. Red-cockaded woodpecker foraging ecology in anold-growth longleaf pine forest. Wilson Bull. 109 (2), 203–217.

Environmental Systems Research Institute, 2008. ArcMap. Version 9.3. Redlands,CA.

Epting, R.J., Delotelle, R.S., Beaty, T., 1995. Red-cockaded woodpecker territory andhabitat use in Georgia and Florida. In: Kulhavy, D.L., Hooper, R.G., Costa, R.(Eds.), Red-Cockaded Woodpecker: Recovery, Ecology, and Management.College of Forestry, Stephen F. Austin State Univ., Nacogdoches, TX, pp. 270–276.

Erdelen, M., 1984. Bird communities and vegetation structure: I. Correlations andcomparisons of simple and diversity indices. Oecologia 61, 277–284.

Falkowski, M.J., Evans, J.S., Martinuzzi, S., et al., 2009. Characterizing forestsuccession with lidar data: an evaluation for the Inland Northwest. USA. Rem.Sens. Environ. 113 (5), 946–956.

Franklin, J., 2009. Mapping Species Distributions: Spatial Inference and Prediction.Cambridge University Press, Cambridge, UK.

Frazer, G., Wulder, M.A., Niemann, K.O., 2005. Simulation and quantification of thefine-scale spatial pattern and heterogeneity of forest canopy structure: alacunarity-based method designed for analysis of continuous canopy heights.Forest Ecol. Manag. 214, 65–90.

Gardner, R.H., 1999. QRULE: A Computer Program for Landscape HypothesisTesting. Appalachian Laboratory, University of Maryland Center forEnvironmental Science, Frostburg, MD, USA.

Goetz, S., Steinberg, D., Dubayah, R., Blair, B., 2007. Laser remote sensing of canopyhabitat heterogeneity as a predictor of bird species richness in an easterntemperate forest. USA Rem. Sens. Environ. 108, 254–263.

Goetz, S., Steinberg, D., Betts, M.G., 2010. Lidar remote sensing variables predictbreeding habitat of a Neotropical migrant bird. Ecology 91 (6), 1569–1576.

Gould, W.A., Vierling, L., Martinuzzi, S., 2010. Integrating lidar in modeling GAPhabitats. GAP Anal. Proj. Rpt. 17, 33–34.

Hall, S.A., Burke, I.C., Box, D.O., Kaufmann, M.R., Stoker, J.M., 2005. Estimating standstructure using discrete-return lidar: an example from low density, fire proneponderosa pine forests. For. Ecol. Manage. 208, 189–209.

Hawbaker, T.J., Keuler, N.S., Lesek, A.A., Gobakken, T., Contrucci, K., Radeloff, V.C.,2009. Improved estimates of forest vegetation structure and biomass with aLiDAR-optimized sampling design. J. Geophys. Res. 114, 1–11.

Hawbaker, T.J., Gobakken, T., Lesak, A., Tromborg, E., Contrucci, K., Radeloff, V., 2010.Light detection and ranging-based measures of mixed hardwood foreststructure. For. Sci. 56, 313–326.

Hinsley, S.A., Hill, R.A., Gaveau, D.L.A., Bellamy, P.E., 2002. Quantifying woodlandstructure and habitat quality for birds using airborne laser scanning. Funct. Ecol.16, 851–857.

Hooper, R.G., 1988. Longleaf pines used for cavities by red-cockaded woodpeckers. J.Wildlife Manage. 52, 392–398.

110 L.S. Smart et al. / Forest Ecology and Management 281 (2012) 100–110

Hudak, A.T., Crookston, N.L., Evans, J.S., Falkowski, M.J., Smith, A.M.S., Gessler, P.E.,Morgan, P., 2006. Regression modeling and mapping of coniferous forest basalarea and tree density from discrete-return lidar and multi-spectral satellitedata. Canad. J. Rem. Sens. 32, 126–138.

Hyde, P., Dubayah, R., Peterson, B., Blair, J.B., Hofton, M., Hunsaker, C., 2005.Mapping forest structure for wildlife habitat analysis using waveform lidar:validation of montane ecosystems. Rem. Sens. Environ. 96, 427–437.

Hyde, P., Dubayah, R., Walker, W., Blair, J.B., Hofton, M., Hunsaker, C., 2006. Mappingforest structure for wildlife habitat analysis using multi-sensor (LiDAR, SAR/InSAR, ETM+, Quickbird) synergy. Rem. Sens. Environ. 102, 63–73.

Hyyppa, J., Hyyppa, H., Inkinen, M., Engdahl, M., Linko, S., Zhu, Y., 2000. Accuracycomparison of various remote sensing data sources in the retrieval of foreststand attributes. For. Ecol. Manage. 128, 109–120.

Imhoff, M.L., Sisk, T.D., Milne, A., Morgan, G., Orr, T., 1997. Remotely sensedindicators of habitat heterogeneity: use of synthetic aperture radar in mappingvegetation structure and bird habitat. Rem. Sens. Environ. 60, 217–227.

Jackson, J.A., 1971. The evolution, taxonomy, distributions, past populations, andcurrent status of the red-cockaded woodpecker. In: Thompson, R.L. (Ed.), TheEcology and Management of the Red-Cockaded Woodpecker. Bureau of SportFisheries and Wildlife and Tall Timbers Research Station, Tallahassee, Florida,pp. 4–29.

Kelly, J.F., Pletschet, S.M., Leslie Jr., D.M., 1994. Decline of the red-cockadedwoodpecker (Picoides borealis) in southeastern Oklahoma. Am. Midl. Nat. 132,275–283.

Kruger, L.M., Midgely, J.J., Cowling, R.M., 1997. Resprouters vs. reseeders in SouthAfrican forest trees; a model based on forest canopy height. Funct. Ecol. 11,101–105.

Lefsky, M.A., Cohen, W.B., Parker, G.G., Harding, D.J., 2002. Lidar remote sensing forecosystem studies. BioScience 52, 19–30.

Lennartz, M.R., Geissler, R.F., Harlow, R.C., Long, K.M., Chitwood, K.M., Jackson, J.A.,1983. Status of the red-cockaded woodpecker o federal lands in the south. In:Wood, D.A. (Ed.), Red-cockaded Woodpecker Symposium II Proceedings. FloridaGame and Fresh Water Fishing Commission, Tallahassee, Florida, pp. 7–12.

Ligon, J.D., Stacey, P.B., Conner, R.N., Bock, C.E., Adkisson, C.S., 1986. Report of theAmerican ornithologist’s union committee for the conservation of the red-cockaded woodpecker. Auk 103, 848–855.

Lim, K., Treitz, P., Wulder, M., St-Onge, B., Flood, M., 2003. LiDAR remote sensing offorest structure. Prog. Phys. Geog. 27 (1), 88–106.

Lowry, J. et al., 2007. Mapping moderate-scale land cover over very large geographicareas within a collaborative framework: a case study of the Southwest RegionalGap Analysis Project (SWReGAP). Rem. Sens. Environ. 108, 59–73.

Matlack, G.R., 1994. Vegetation dynamics of the forest edge—trends in space andsuccessional time. Ecology 82, 113–123.

MacArthur, R.H., MacArthur, J.W., 1961. On bird species diversity. Ecology 42, 594–598.

MacArthur, R.H., Macarthur, J.W., Preer, J., 1962. On bird species diversity. Am. Nat.96, 167–174.

Maclean, G.A., Krabill, W.B., 1986. Gross-merchantable timber volume estimationusing an airborne LIDAR system. Canad. For. Res. 12, 7–18.

Maltamo, M., Packalén, P., Yu, X., Eerikäinen, K., Hyyppä, J., Pitkänen, J., 2005.Identifying and quantifying structural characteristics of heterogeneous borealforests using laser scanner data. For. Ecol. Manag. 216, 41–50.

Marinuzzi, S., Vierling, L.A., Gould, W.A., Falkowski, M.J., Evans, J.S., Hudak, A.T.,Vierling, K.T., 2009. Mapping snags and understory shrubs for a LiDAR-basedassessment of wildlife habitat suitability. Rem. Sens. Environ. 113, 2533–2546.

Martens, S.N., Borel, C.C., Gerstl, S.A., 1993. High fidelity computer reconstructionsof tree canopy architecture. Bull. Ecol. Soc. Am. 74, 345.

McGarigal, K., Cushman, S. A., Neel, M. C., Ene, E., 2002. FRAGSTATS: Spatial PatternAnalysis Program for Categorical Maps. Computer software program producedby the authors at the University of Massachusetts, Amherst. <http://www.umass.edu/landeco/research/fragstats/fragstats.html>.

McKerrow, A.J., Williams, S.G., Collazo, J.A., 2006. The North Carolina Gap AnalysisProject: Final Report. North Carolina Cooperative Fish and Wildlife ResearchUnit, North Carolina State University, Raleigh, North Carolina. <http://www.basic.ncsu.edu/ncgap/NCFinal%20Report.pdf>.

Mills, L., Fredrickson, R., Moorhead, B., 1993. Characteristics of old-growth forestsassociated with northern spotted owls in Olympic National Park. J. WildlifeManage. 57, 315–321.

Morsdorf, F., Kotz, B., Meier, E., Itten, K.I., Allgower, B., 2006. Estimation of LAI andfractional cover from small footprint airborne laser scanning data based on gapfraction. Rem. Sens. Environ. 104, 50–61.

Müller, J., Stadler, J., Brandl, R., 2010. Composition versus physiognomy ofvegetation as predictors of bird assemblages. Rem. Sens. Environ. 114, 490–495.

North Carolina Flood Mapping Program, 2008. North Carolina Division ofEmergency Management. <http://www.ncfloodmaps.com/issue_papers.htm>.

Newell, F.L., Rodewald, A., 2011. Role of topography, canopystructure, and floristicsin nest-site selection and nesting success of canopy songbirds. For. Ecol.Manage. 262, 739–749.

Ørka, H.O., Næsset, E., Bollandsås, O.M., 2009. Classifying species of individual treesby intensity and structure features derived from airborne laser scanner data.Rem. Sens. Environ. 113, 1163–1174.

Pascual, C., García-Abril, A., Cohen, W.B., Martin-Fernández, S., 2012. Relationshipbetween lidar-derived forest canopy height and Landsat images. Int. J. Rem.Sens. 30 (5), 1261–1280.

Peet, R.K., Wentworth, T.R., White, P.S., 1998. A flexible, multipurpose method forrecording vegetation composition and structure. Castanea 63, 262–274.

Phillips, S.J., Anderson, R.P., Schapire, R.E., 2006. Maximum entropy modeling ofspecies geographic distributions. Ecol. Model. 190, 231–259.

Plotnick, R.E., Gardner, R.H., O’Neill, R.V., 1993. Lacunarity indices as measures oflandscape texture. Land. Ecol. 8 (3), 201–211.

R Development Core Team, 2009. R Project. A Language and Environment forStatistical Computing. <http://www.r-project.org/>.

Rosenberg, D.K., Anthony, R.G., 1992. Characteristics of northern flying squirrelpopulations in second and ol-growth forests in western Oregon. Can. J. Zoo. 70,161–166.

Roth, R.R., 1976. Spatial heterogeneity and bird species diversity. Ecology 57, 773–782.

Rudolph, D.C., Conner, R.N., Shaefer, R.R., 2002. Red-cockaded woodpecker foragingbehavior in relation to midstory vegetation. Wilson Bull. 114, 235–242.

Sexton, J.O., Bax, T., Siqueira, P., Swenson, J.J., Hensley, S., 2009. A comparison oflidar, radar, and field measurements of canopy heights in pine and hardwoodforests of southeastern North America. For. Ecol. Manage. 257, 1136–1147.

Sherrill, K.R., Lefsky, M.A., Bradford, J.B., Ryan, M.G., 2008. Forest structureestimation and pattern exploration from discrete-return lidar in subalpineforests in the central Rockies. Can. J. Forest Res. 38, 2081–2096.

Smart, L., 2009. Characterizing Spatial Pattern and Heterogeneity of Pine Forests inNorth Carolina’s Coastal Plain using LiDAR. Masters project. Nicholas School ofEnvironment, Duke University. <http://dukespace.lib.duke.edu/dspace/bitstream/handle/10161/1027/SmartL_FinalMP.pdf?sequence=1>.

Stoker, J., Harding, D., Parrish, J., 2008. The need for a national lidar dataset.Photogramm. Eng. Rem. S. 74, 1066–1068.

Treuhaft, R.T., Madsen, S.N., Moghaddam, M., van Zyl, J.J., 1996. Vegetationcharacteristics and underlying topography from interferometric radar. Rad.Sci. 31, 1449–1485.

US Fish and Wildlife Service, 2003. Recovery Plan for the Red-CockadedWoodpecker (Picoides borealis): Second Revision. US Fish and Wildlife Service,Atlanta, GA. <http://ecos.fws.gov/docs/recovery_plan/030320_2.pdf>.

US Marine Corps Base Camp Lejeune, 2006. Integrated Natural ResourcesManagement Plan (2007–2011). USMCB Camp Lejeune, North Carolina.<http://www.lejeune.usmc.mil/emd/INRMP%202007/table_of_contents.htm>.

Vierling, K., Vierling, L.A., Gould, W.A., Martinuzzi, S., Clawges, R.M., 2008. Lidar:shedding new light on habitat characterization and modeling. Front. Ecol.Environ. 6, 90–98.

Vogelmann, J.E., Sohl, T.L., Campbell, P.V., Shaw, D.M., 1998. Regional land covercharacterization using Landsat Thematic Mapper data and ancillary datasources. Environmental Monitoring and Assessment 51, 415–428, <http://www.epa.gov/mrlc/nlcd-2001.html>.

Walters, J.R., 1991. Application of ecological principles to the management ofendangered species: the case of the red-cockaded woodpecker. Ann. Rev. Ecol.Syst. 22, 505–523.

Waring, R.H., Way, J.B., Hunt, R., Morrisey, L., Ranson, K.J., Weishampel, J.F., Oren, R.,Franklin, S.E., 1995. Biological toolbox – imaging RADAR for ecosystem studies.Bioscience 45, 715–723.

Weishampel, J.F., Sloan, J.H., Boutet, J.C., Godin, J.R., 1998. Mesoscale changes intextural properties of ‘‘intact’’ Peruvian rainforests (1970s–1980s). Int. J. Rem.Sens. 19, 1007–1014.

Weishampel, J.F., Blair, J.B., Know, R.G., Dubayah, R., Clark, D.B., 2000. Volumetriclidar return pattern from an old-growth tropical rainforest canopy. Int. J. Rem.Sens. 21 (2), 409–415.

Welden, C.W., Hewett, S.W., Hubbell, S.P., Foster, R.B., 1991. Sapling survival,growth, and recruitment: relationship to canopy height in a neotropical forest.Ecology 72, 35–50.

Wood, D.A. (Ed.), 1983. Red-Cockaded Woodpecker Symposium II Proceedings.Florida Game and Fresh Water Fish Commission. 112pp.

Woodcock, C.E., Strahler, A.H., 1987. The factor of scale in remote sensing. Rem.Sens. Environ. 21 (3), 311–332.