ith
A.
ad, F
Available online 8 May 2011
Keywords:
d a
830
plantation. We examined 198 plantations that varied widely in age (5?47 years), size (1.25?47 ha), and
ent ca
age than surrounding unmanaged forests (Odion et al., 2004;
Thompson et al., 2007; Weatherspoon and Skinner, 1995). This is
likely attributable to higher stem densities and continuous cano-
pies, which are characteristic features of plantations and can in-
crease vulnerability to crown ?re (Kobziar et al., 2009; Stephens
of forest management, including silviculture treatments of various
ages, sizes, and techniques. This included >8300 ha of even-aged
conifer plantations, which were established following clearcut har-
vesting and planted primarily with Douglas-?r (Pseudotsuga menzi-
esii) and to a much lesser degree ponderosa pine (Pinus ponderosa)
and sugar pine (Pinus lambertiana). The region has a long history of
using even-aged silvicultural practices to achieve timber produc-
tion goals, which were a dominant management objective from
the 1950s until the early 1990s, when federal logging was curtailed
with the adoption of the Northwest Forest Plan (Walstad, 1992).
Accordingly, most plantations encountered by the Biscuit Fire
? Corresponding author. Tel.: +1 540 635 6580; fax: +1 540 635 6506.
E-mail addresses: thompsonjr@si.edu (J.R. Thompson), tspies@fs.fed.us (T.A.
Spies), keith.olsen@oregonstate.edu (K.A. Olsen).
1 Tel.: +1 541 750 7354; fax: +1 541 750 7329.
Forest Ecology and Management 262 (2011) 355?360
Contents lists availab
Forest Ecology an
.e l2 Tel.: +1 541 750 7279; fax: +1 541 750 7329.of burn damage left in the aftermath of a large wild?re. Several
studies have documented the persistent in?uence of partial har-
vests and fuel treatments on wild?re effects, (e.g. Finney et al.,
2005; Pollet and Omi, 2002; Prichard et al., 2010; Raymond and
Peterson, 2005). Equally important, though somewhat less repre-
sented in the literature, are studies that quantify the in?uence of
even-aged silvicultural treatments on wild?re effects. Even-age
plantations are a common feature of forested landscapes world-
wide and there are more than 17 million hectares of conifer plan-
tations in the US alone (FAO, 2005). The available evidence
suggests that plantations experience higher levels of canopy dam-
ness of tree bark can increase, all of which may decrease the risk
of ?re damage in some forest types (Agee, 1993; Hanus et al.,
2000). This pattern suggests that the increased risk of canopy dam-
age within plantations could be reduced with the passage of time,
but an extensive search of the literature produced no empirical
data regarding how ?re damage varies with plantation age or
structure.
The 2002 Biscuit Fire burned at mixed-severities across
>200,000 ha of mixed conifer and evergreen hardwood forests in
southwest Oregon and northwest California. The fuel complex
encountered by the Biscuit Fire was strongly affected by a legacyBiscuit Fire
Burn severity
Land use legacy
Klamath Mountains
Random Forest analysis
Geostatistical regression
1. Introduction
The legacy of past forest managem0378-1127/$ - see front matter Published by Elsevier
doi:10.1016/j.foreco.2011.04.001landscape context. The average level of canopy damage within the plantations was 77%. Based on Ran-
dom Forest variable importance values, plantation age was the best predictor of canopy damage. Average
annual precipitation, elevation and topographic position were ranked second, third, and fourth, respec-
tively. A model selection procedure, using geo-statistical regression models and Akaike?s information cri-
terion, corroborated the importance of plantation age relative to the other predictors tested and also
suggested that the in?uence of age varied over time. The top ranked regression model indicated that
the level of canopy damage reached its maximum around age 15 and stayed relatively high until age
25 before declining.
Published by Elsevier B.V.
n in?uence the mosaic
and Moghaddas, 2005; Graham et al., 2004). However, as trees
within plantations mature, self-pruning results in higher crown
base heights, self-thinning can reduce tree density, and the thick-Received in revised form 3 March 2011
Accepted 1 April 2011
tography to examine how the level of canopy damage varied within these plantations in relation to
topography, weather, vegetation-cover, and management history, with an emphasis on the age of theCanopy damage to conifer plantations w
varies with stand age
Jonathan R. Thompson a,?, Thomas A. Spies b,1, Keith
a Smithsonian Institution, Smithsonian Conservation Biology Institute, 1500 Remount Ro
bUSDA Forest Service, Paci?c Northwest Research Station, Corvallis, OR 97331, USA
cDept. of Forest Science, Oregon State University, Corvallis, OR 97331, USA
a r t i c l e i n f o
Article history:
Received 4 December 2010
a b s t r a c t
The 2002 Biscuit Fire burne
land, including more than
journal homepage: wwwB.V.in a large mixed-severity wild?re
Olsen c,2
ront Royal, VA 22630, USA
t mixed-severities encompassing over 200,000 ha of publicly owned forest-
0 ha of conifer plantations. We used pre- and post-?re digital aerial pho-
le at ScienceDirect
d Management
sevier .com/locate / foreco
on the day of burn, climate (productivity) and vegetation-cover.
Ideally, ?re effects should be quanti?ed through pre-?re ?eld mea-
andsurements of fuel conditions coupled to post-?re measures of ?re-
effects on above- and below-ground resources. Unfortunately, the
expense of ?eld sampling and the inability to forecast wild?re loca-
tions and measure them in advance of wild?re occurrence limits
the use of this approach. Instead, we interpreted vegetation condi-
tions within pre- and post-?re digital aerial photos. By interpreting
vegetation condition using digital aerial photography, we are able
to attain some of the ecological resolution of ground plots but with
the data collection facility of remote sensing.
Based on our previous work quantifying ?re damage within the
unmanaged portion of the Biscuit Fire, which showed the highest
levels of canopy damage in very young and shrubby vegetation
(Thompson and Spies, 2009), we hypothesized that plantation
age would be negatively correlated with the level of canopy dam-
age but that the effect of plantation age would decrease over time.
Further, we hypothesized that daily ?re weather conditions would
also be an important predictor of canopy damage, with extreme
?re weather conditions overriding all other structural or other
environmental variables.
2. Methods
2.1. Study area
The study was conducted within the perimeter of the 2002 Bis-
cuit Fire, which encompassed approximately 200,000 ha of the
Klamath Mountains in southwest Oregon and northwest California.
The area is primarily managed by the Rogue?Siskiyou National For-
est (RSNF) and is within the mixed evergreen vegetation zone
(Franklin and Dyrness, 1988). While the Biscuit region does include
areas of low-productivity, ultrama?c soils, those regions were ex-
cluded from this study. The plantations we examined are underlain
by igneous, meta-sedimentary, and metamorphic soil parent mate-
rials. Unmanaged forests on these soils are dominated by conifer
species such as Douglas-?r, sugar pine, and white ?r (Abies concol-
or). Dominant evergreen hardwoods include tanoak (Lithocarpus
densi?ora), Paci?c madrone (Arbutus menziesii), and canyon live-
oak (Quercus chrysolepis). Manzanita (Arctostaphylos sp.), and Sa-
dler oak (Quercus sadleriana) are common shrubs. Topography
within the Biscuit Fire is steep and complex; elevations range fromrange in age from approximately 10 to 50 years old. We are aware
of no previous empirical studies to explicitly consider the relation-
ship between plantation age and wild?re damage, however,
Graham (2003) did note that plantations younger than 12 years
experienced higher levels of burn severity (as determined from
Landsat-derived burn mapping) than did older plantations during
the 2002 Hayman Fire in Colorado. Similarly, Thompson et al.
(2007) found high levels of canopy damage within 12- to 15-
year-old management units that were salvage-logged and planted
after the 1987 Silver Fire then burned again in the Biscuit Fire. Fire
modeling also suggests that plantations are more vulnerable than
unmanaged stands from the time they are young saplings
(<5 years) at least until they reach >50 cm diameter at breast
height (Stephens and Moghaddas, 2005).
Our overall objective was to develop a better understanding the
factors that were associated with burn damage within the inten-
sively managed portion of the Biscuit Fire. We used pre- and
post-?re digital aerial photography to examine how the level of
Biscuit Fire canopy damage varied within 198 plantations in rela-
tion to their age, topographical setting, the weather conditions
356 J.R. Thompson et al. / Forest Ecology100 to 1500 m. Mean January temperature is 6 C. Mean July tem-
perature is 16 C. Mean annual precipitation is 270 cm, with great-
er than 90% occurring as a mixture of snow and rain during winterand spring (Daly et al., 2002). A detailed account of the Biscuit
Fire?s effect on vegetation cover within unmanaged areas can be
found in Thompson and Spies (2009).
2.2. Management data
Our analysis focused on 200 even-aged plantations randomly
selected from a RSNF spatial database that described the location
of all signi?cant historical logging and planting, which included a
total of 652 conifer plantations (8300 ha) within the ?re?s perime-
ter. To be eligible for inclusion in this study, each unit must have
been clearcut between 1960 and 1996 and have a record of suc-
cessful conifer planting. Of the 200 selected units, 35 were sal-
vage-harvests completed between 1988 and 1991 following the
1987 Silver Fire (to determine if the salvage units had a unique
in?uence on canopy damage we analyzed our data with and with-
out these plantations.). Two units were later removed because
their positions were inaccurate within the spatial database. Re-
cords were incomplete regarding species composition and volume
removed, site preparation, and planting density. However, discus-
sions with RSNF employees indicated that some live trees were left
after harvests and that planting was overwhelmingly Douglas-?r
with a much lesser component of ponderosa and sugar-pine. Multi-
ple planting dates, all clustered within 1?3 years of harvest, were
often associated with individual management units. We therefore
used the date of harvest as a surrogate for the plantation?s estab-
lishment date, unless there was evidence that original planting
had failed and the site had been reforested at a later date. Harvest
date information was considered reliable by RSNF personnel (pers.
comm. J. Hawkins, Gold Beach Ranger District, RSNF).
2.3. Aerial photo plots
Photo-plots were a grid of 50-by-50 meter cells overlain onto
the variably-shaped harvest unit polygons supplied by the RSNF.
On large harvest units (>6.25 ha), we randomly selected 25 cells
to use as the plot. For management units <6.25 ha but >1.25 ha,
we used all cells as the photo-plot. Management units <1.25 ha
were excluded from this study. The best available pre-Biscuit Fire
photos were digital orthoquads taken as part of the USDA National
Agriculture Imagery Program in August 2000; they were panchro-
matic with a 1 m grain size. The post-Biscuit Fire photos were ta-
ken on September 24, 2002, were true color, and had a 25 cm
grain size. We spatially co-registered the pre- to post-?re photo
plots using approximately 15 ground control points per plot and
used a ?rst-order polynomial transformation for geo-recti?cation.
Starting with the pre-?re photos, a single researcher (Thompson)
estimated the percent cover of live vegetation and bare ground/
grass cover (which were indistinguishable) in each cell in every
plot. Then, using the post-?re photos, the same researcher mea-
sured the percent of the vegetation cover that was scorched or con-
sumed (i.e. canopy damage) by the Biscuit Fire. Cell-level estimates
were then averaged to obtain plot-level values. Our original intent
was to separate canopy consumption from canopy scorch to infer
differences in ?re behavior (i.e. surface ?re versus torching). Unfor-
tunately, however, the vertical and horizontal continuum between
scotch and consumption we witnessed in the photos and in ?eld
assessments revealed that any attempts to make inferences in this
regard would be unreliable. Therefore, we treated scorch and con-
sumption collectively as ??canopy damage.??
At the onset of the research, we developed a catalog of paired
oblique-to-aerial photos for use as a training manual and later
informally ground-truthed a subset of photo-plots, which revealed
Management 262 (2011) 355?360excellent correspondence between post-?re ?eld conditions and
photo measurements. Indeed, the 25 cm resolution of the post-?re
photography permitted a unambiguous interpretation of the ?re?s
over time by adding a polynomial term. Semivariograms of model
residuals from an ordinary least squares (OLS) regression displayed
strong positive spatial autocorrelation to distances >5 km (not
shown). Due to the lack of independence of the residuals and the
shape of the semivariogram we chose to ?t a generalized least
squares (GLS) regression models that included a spherical spatial
correlation structure using the ?nlme? package (Pinheiro et al.,
2009) within the R statistical environment (R Development Core
Team, 2006). GLS regression relies on the distance between sample
locations and the form of the correlation structure to derive a var-
iance?covariance matrix, which is, in turn, used to solve a
weighted OLS regression (Dormann et al., 2007).
analyses.
and Management 262 (2011) 355?360 357effects on tree canopies. Nonetheless, it is important to note that
canopy damage measured from a planer view of the landscape
(i.e. from an aerial photo) is not strictly equivalent to the propor-
tion of the crown volume damaged as measured in the ?eld.
2.4. Topographic and weather variables
We used a 10-m digital elevation model to calculate the average
elevation, percent slope, Beers? transformed aspect (Beers et al.,
1966), and topographic position for each photo-plot. To capture lo-
cal and broad scale variation in topography, we calculated topo-
graphic position at two scales: ??TP-Fine?? is the difference
between the mean plot elevation and the mean elevation in an
annulus 150?300 m from the plot, while ??TP-Coarse?? uses an
annulus 850?1000 m from the plot. The topographic index values
are in units of meters, but their usefulness is chie?y in a relative
sense (c.f. Jones et al., 2000). For example, within the TP-Fine index,
a value of, say 30 m, re?ects the fact that most of the area immedi-
ately around the focal site (within 150?300 m) is at a higher eleva-
tion. The RSNF provided a map that depicted the daily progression
of the Biscuit Fire, which we used to assign weather data to each
photo-plot based on the day it burned. We assigned the average
temperature, relative humidity, wind speed, and cosine trans-
formed wind direction between 10:00 and 19:00 for each day as
calculated from the Quail Prairie Remote Automated Weather Sta-
tion, located within the ?re perimeter. To capture regional gradi-
ents in productivity associated with moisture availability, we
assigned each photo plot the average local annual precipitation
for the climatological period spanning 1971?2000 to each plot
based on the PRISM model (Daly et al., 2002).
2.5. Data analysis
To rank the predictor variables in terms of the strength of their
relationship to the response, we calculated variable importance
values using the Random Forest (RF) algorithm (Liaw and Wiener,
2002) within the R statistical environment (R Development Core
Team, 2006). While RF is relatively new to forestry and ecological
research, its use is growing and, in simulation and comparative
analyses, it has consistently out-performed other methods for pre-
diction accuracy and ranking variable importance (Cutler et al.,
2007; Lawler et al., 2006; Prasad et al., 2006). The RF algorithm
(as applied to these data) selects 1500 bootstrap samples, each
containing two-thirds of the photo plots. For each sample, it cre-
ates an un-pruned regression tree with modi?cation that, at each
node, it randomly selects only one-third of the predictor variables
and chooses the best partition from among those variables. To as-
sess the predictive power of the model, RF calculates an ensemble
average of all the regression trees, which is used to predict the le-
vel of canopy damage for the plots not included in the bootstrap
sample. The RF model is then used to calculate importance values
for each of the predictor variables by calculating the percent in-
crease in the mean squared error (MSE) in the predicted data when
the values for that predictor are permuted and the others are left
intact.
To further assess potential relationship between canopy dam-
age and the predictor variables (including potential interactions)
we compared a series of regression models using Akaike?s informa-
tion criterion, (AIC; Burnham and Anderson, 2002). We compared
11 different regression models that included the top-ranking pre-
dictors from the RF analysis (i.e. those predictors uses inclusion
the model reduces the MSE by >10%) in addition to a null model
that contained no predictor variables. Due to the relative impor-
J.R. Thompson et al. / Forest Ecologytance of plantation age in the RF model and our hypothesis that
the in?uence of age would vary over time, we also assessed
whether the relationship between age and canopy damage variedN
um
be
r o
f P
la
nt
at
io
ns
1960 1980 2000
0
5
10
15
20
253. Results
Sampled plantations ranged in age from 5 to 47 years (Fig. 1)
and in size from 1.25 to 47 ha. Ninety-seven percent of the planta-
tions (192 of 198) had >1% canopy damage. The average level of
canopy damage within photo plots was 77% (SD = 20.1; Table 1).
The RF model explained 34% of variability in canopy damage and
identi?ed plantation age as the most important predictor variable
(Fig. 2), with older plantations experiencing lower levels of canopy
damage. Average annual precipitation had a generally negative
relationship with canopy damage and was ranked second by the
RF model. Elevation and topographic position both had a positive
relationship with canopy damage and were ranked third and
fourth, respectively. No other predictor variable included within
the RF model reduced the MSE by >10%.
Based on the RF results, we compared 11 different GLS regres-
sion models (Table 2). The top ranked model included plantation
age and a polynomial term that allowed the effect of age on canopy
damage to vary. This model was signi?cant at P < 0.0001 and had a
pseudo-R2 of 0.30 (Fig. 3). Modeled percent canopy damage
reached its maximum (91%) in plantations that were around age
15 and stayed relatively high (above 80%) within plantations that
were between 15 and 25 years old before declining in older planta-
tions. Based on conventions of the information theoretic approach
(whereby models whose AIC statistics are within two units of the
highest ranked model are considered equal (Burnham and Ander-
son, 2002)), no other model ?t the data as well. However, it is
important to note that, while modeled canopy damage does de-
cline after plantations reach age 25, there is considerable variabil-
ity in the data, and some plantations >25 years did experience high
levels of damage. These results were not qualitatively different
when we removed those plantations that were created after
post-?re salvage logging from 1988 to 1990 then reran theYear Plantation Established
Fig. 1. The distribution of sampled plantation ages.
alys
andTable 1
Summary statistics for response and predictor variables used in the Random Forest an
358 J.R. Thompson et al. / Forest Ecology4. Discussion
Given the absence of controlled experiments within large wild-
?res, long-term records of forest management type, location, and
burn).
Variables Mean
Response variable
Percent crown damage 77.8
Predictor variables
Age (years) 22
Harvest size (ha) 14.1
Vegetation cover (%) 89.8
Bare/grass cover (%) 10.3
Elevation (m) 885
Topographic position (?ne) 3.4
Topographic position (coarse) 26.8
Slope (%) 40
Beer?s aspect 0.2
Average annual precipitation (cm) 320
Temperature on DOB (C) 26.6
Relative humidity on DOB (%) 31
Wind speed on DOB (km/h) 9
Wind direction on DOB (cosine transformed) 0.24
Plantation Size
Wind Speed
Wind direction
Grass/Open Cover
TPI-Coarse
Aspect
Vegetation Cover
Slope
Humidity
Burn Period
Temperature (+)
TPI-Fine (+)
Elevation (+)
Precipitation (-)
Plantation Age (-)
0 20 40 60
% increase in MSE
Fig. 2. Variable importance plots for predictor variables from a Random Forests
model of canopy damage within conifer plantations. Predictor variables are along
the y-axis and the average increase in the mean square error when data for that
variable are permuted and all other are left unchanged is on the x-axis. The
direction of the relationship is given in parentheses for predictor variables whose
Pearson?s correlation were signi?cant at p < 0.05; however, we urge caution in this
interpretation as Random Forest variable importance values are not based on linear
relationships alone.
Table 2
Comparison of geo-statistical regression models based on Akaike information criteria
(AIC; interaction terms implicitly include their associated additive term).
Rank Model Form AIC D AIC xi
1 AGE + AGE2 1713.9 0 0.907
2 AGE + AGE2 + TPI 1719.3 5.4 0.061
3 AGE + AGE2 + ELEV 1721.3 7.4 0.022
4 AGE + AGE2 + PRECIP 1723.8 9.9 0.006
5 AGE + AGE2 + PRECIP + TPI 1726.7 12.8 0.002
6 AGE 1727.6 14.1 0.001
7 AGE + AGE2 + PRECIP + ELEV 1729.2 15.3 0.000
8 AGE  TPI 1738.5 24.6 0.000
9 AGE  ELEV 1748.8 34.9 0.000
10 AGE  PRECIP 1753.9 40.0 0.000
11 NULL MODEL 1760.2 46.3 0.000is of crown damage within conifer plantations in the 2002 Biscuit Fire (DOB = day of
Standard deviation Minimum Maximum
20.1 0 100
8.4 5 42
15.3 1.25 47
10.7 47 100
10.0 0 53
227 265 1346
13.1 25.7 49.4
77.0 176.4 203.3
13.6 12 79
0.5 0.97 0.99
71 171 439
4.9 16.6 35.8
15 10 65.5
Management 262 (2011) 355?360intensity are important for retrospectively assessing wild?re ef-
fects. Unfortunately, with the exception of plantation age and its
status as a ??successful?? reforestation effort, we had no reliable
and consistent records documenting the plantations? speci?c man-
agement history or composition at the time of the ?re. The lack of
site information is important limitation of this study. Indeed, site
preparation has been shown to be an important predictor of plan-
tation canopy damage, for example, where broadcast burned sites
experienced signi?cantly less damage than untreated or piled-and-
burned sites (Weatherspoon and Skinner, 1995). Similarly, in a
fortuitous experiment regarding ?re effects after thinning in 90?
120 year old unmanaged stands within the Biscuit Fire, tree mor-
tality was lowest (5%) on sites that were thinned in 1996 then
broadcast burned in 2001, just 1 year before the ?re, intermediate
in unmanaged sites (53?54%), and highest in sites that were
thinned in 1996 but not broadcast burned (80?100%; Raymond
and Peterson, 2005). Nonetheless, we were able to look back over
40 years of even-age forest management and document a relatively
strong relationship between plantation age and the level of canopy
2.1 4.2 18.2
0.44 0.3 0.75
Fig. 3. Relationship between plantation age and percent canopy damage used to ?t
a generalized least squares regression model with a spatial spherical correlation
structure to accommodate positive spatial autocorrelation. Dashed lines represent
95% con?dence intervals.
damage in relation to several variables describing the vegetation-
anddamage. To our knowledge, no similar empirical information doc-
umenting this relationship exists.
Of the 15 predictor variables we examined, we found that level
of canopy damage in plantations was most strongly related to its
age. Indeed, the variable describing the age of plantations reduced
the error in the model by more than twice as much as any of the
other predictors (Fig. 2). The shape of the best-?tting geo-statisti-
cal regression model suggested that the level of canopy damage
reached its maximum around age 15 and stayed relatively high un-
til age 25 before declining (Fig. 3). This pattern is consistent with
what is known about the ?re ecology of Douglas-?r forests,
whether planted or naturally regenerated (Starker, 1934; Agee,
1993). The ?re resistance of Douglas-?r increases with age due to
a continually thickening layer of protective bark and due to
increasing height-to-crown, which is associated with reduced like-
lihood of torching or crown ?res (Scott and Reinhardt, 2001;
Graham et al., 2004). In effect, as Douglas-?r matures it transitions
from an ??avoider?? (a species that is vulnerable to low intensity
?res) to a ??resister?? (a species that has adaptations that increase
the probability of survival during low intensity ?res; Agee, 1993,
pp. 206 and 285). Empirical growth curves show that by the time
a Douglas-?r plantation in southwest Oregon is 25 years old an
average tree is typically between 8 and 16 m tall and have crown
base heights >3 m off the ground, depending on site class and stem
density (Hann and Scrivani, 1987; Hanus et al., 2000). The combi-
nation of bark thickness and a suf?ciently high crown-base-height
is the likely explanation for, on average, decreasing canopy damage
in the older plantations. However, given the data available to us in
this study it is impossible to know exactly which ?re resistance
strategy (or combination of strategies) was responsible for the pat-
tern of decreasing canopy damage with plantation age.
Our previous research in the unmanaged portion of the Biscuit
suggested that weather conditions on the day of the burn were
an important correlate of canopy damage (Thompson and Spies,
2009). Therefore, it was surprising that the predictor variables
describing daily weather conditions were comparatively unimpor-
tant (Fig. 2 and Table 2). There are at least two possible explana-
tions for the difference. In the study of unmanaged forests, we
did not have explicit information on the age or structure of the
stands, but most were mature mixed-conifer >50 years old. It is
possible that ?re weather is a better predictor of canopy damage
in older stands where extremes in wind and fuel moisture are nec-
essary to transition from a surface ?re to torching the canopy or
running as a crown ?re (Van Wagner, 1977). Another possible rea-
son for the difference relates to differences in the sampling extent
and intensity. The study of unmanaged forest had many more plots
that encompassed a much larger area and burned over a longer
period of time spanning a greater range of variability in weather
conditions. Average annual precipitation was ranked as the second
most important predictor of canopy damage within the plantation
and had a generally negative correlation. In this landscape, precip-
itation in correlated with productivity (Coops and Waring, 2001).
Given the much stronger relationship between canopy damage
and plantation age, the weak negative relationship with precipita-
tion may suggest that greater moisture and productivity acceler-
ated stand development and, in turn, decreased the age at which
?re resistance is reached.
Given the ubiquity of plantations within ?re-prone landscapes,
it is perhaps surprising that so little research has been done regard-
ing ?re behavior in even-aged conifer plantations. The existing
empirical research suggests that when compared to more hetero-
geneous unmanaged forests, plantations are associated with ele-
vated ?re damage (Weatherspoon and Skinner, 1995; Odion
J.R. Thompson et al. / Forest Ecologyet al., 2004). While the intent of this study was not to compare
unmanaged stands to plantations, it is worth noting that, in a sep-
arate examination of Biscuit Fire effects (Thompson and Spies,cover, topographic setting, weather conditions on the day a site
burned, and the time since the plantations were established (i.e.
the plantations? age). We found that age of a plantation was the
best predictor of the level canopy damage and that the other vari-
ables were comparatively poor predictors. The best ?tting geosta-
tistical regression model indicated that the level of canopy damage
reached its maximum around age 15 and stayed relatively high un-
til age 25 before declining. Based on a previous analysis of unman-
aged vegetation (Thompson and Spies, 2009), we had hypothesized
that daily weather conditions would be important predictor vari-
ables within the models. However, the data did not support this
hypothesis. Our ?ndings, while observational and thus not general-
izable, offer managers and forest scientists a rare empirical per-
spective on patterns ?re damage within even-age conifer
plantations, which are a common landscape feature throughout
the western North America. At least in this case, the data suggest
that young plantations were vulnerable to canopy damage regard-
less of their environmental setting.
Acknowledgements
This research was funded by the Joint Fire Science Program. We
thank Duck Creek Inc., for technical help and W. Cohen, T. Atzet,
and R. Miller for helpful reviews.
References
Agee, J.K., 1993. Fire Ecology of Paci?c Northwest Forests. Island Press, Washington,2009), the average level of canopy damage within unmanaged for-
est with variable stand histories was lower than is was the within
the plantations measured herein (65% in the previous study of
unmanaged forests versus 78% in the present study). Given the
relationship between plantation age and canopy damage, the dif-
ference in canopy damage between unmanaged and unmanaged
stands may have more to do with the fact that most of the unman-
aged stands were >50 years old, than with their origin as ??unman-
aged?? (i.e. naturally regenerated).
With the clear proviso that our study was observational and
only describes Douglas-?r plantations burned within the Biscuit
Fire, for the sake of context it is also worth noting that ?re model-
ing studies in other regions that have examined ?re behavior and
tree mortality within a range of silvicultural treatment types and
ages have found a similar trends of decreasing ?re damage with
increasing age (e.g. Kobziar et al., 2009; Stephens and Moghaddas,
2005). For example, in Sierra Nevadan ponderosa pine plantations,
high rates of mortality were predicted for untreated conifer planta-
tions (when compared to young growth reserves (80?100 years))
across all diameter classes up to 50 cm DBH, regardless of weather
conditions (Stephens and Moghaddas, 2005). In the Biscuit Fire,
young (<15 years) Douglas-?r stands tended experience high levels
of canopy damage whether they were plantations or naturally
regenerated stands. This was demonstrated through a separate
examination of the areas that burned at high severity in the 1987
Silver Fire and were subsequently re-burned by the Biscuit Fire
(Thompson and Spies, 2010).
5. Conclusion
In this paper, we utilized pre- and post-?re digital aerial pho-
tography to assess ?re-related canopy damage within Douglas-?r
plantations in southwest Oregon. We used parametric and non-
parametric modeling approaches to examine the level of canopy
Management 262 (2011) 355?360 359DC.
Beers, T.W., Dress, P.E., Wensel, L.C., 1966. Aspect transformation in site
productivity research. Journal of Forestry 64, 691?692.
Burnham, K.P., Anderson, D.R., 2002. Model Selection and Multimodel Inference: A
Practical Information-theoretic Approach, second ed. Springer, New York.
Cutler, D.R., Edwards, T.C., Beard, K.H., Cutler, A., Hess, K.T., Gibson, J., Lawler, J.J.,
2007. Random forests for classi?cation in ecology. Ecology 88 (11), 2783?2792.
Coops, N.C., Waring, R.H., 2001. Estimating forest productivity in the eastern
Siskiyou Mountains of southwestern Oregon using a satellite driven process
model, 3-PGS. Canadian Journal of Forest Research 31, 143?154.
Daly, C., Gibson, W.P., Taylor, G.H., Johnson, G.L., Pasteris, P., 2002. A knowledge-
based approach to the statistical mapping of climate. Climate Research 22, 99?
113.
Dormann, C.F., McPherson, J.M., Ara?jo, M.B., Bivand, R., Bolliger, J., Carl, G., Davies,
R.G., Hirzel, A., Jetz, W., Kissling, W.D., K?hn, I., Ohlem?ller, R., Peres-Neto, P.R.,
Reineking, B., Schr?der, B., Schurr, F.M., Wilson, R., 2007. Methods to account for
spatial autocorrelation in the analysis of species distributional data: a review.
Ecography 30 (5), 609?628.
FAO, 2005. Global Forest Resources Assessment 2005: Progress towards Sustainable
Forest Management.
Finney, M.A., McHugh, C., Grenfell, I.C., 2005. Stand- and landscape-level effects of
prescribed burning on two Arizona wild?res. Canadian Journal of Forest
Research 35, 1714?1722.
Franklin, J., Dyrness, C., 1988. Natural Vegetation of Oregon and Washington. OSU
Press, Corvallis.
Graham, R.L., 2003. Hayman Fire Case Study. USFS General Technical Report RMRS-
GTR-114.
Graham, R.L., McCaffrey, S., Jain, T.B., 2004. Science Basis for Changing Forest
Structure to Modify Wild?re Behavior and Severity. USFS General Technical
Report RMRS-GTR-120.
Hann, D.W., Scrivani, J.A., 1987. Dominant-height-growth and Site-index Equations
for Douglas-?r and Ponderosa Pine in Southwestern Orgeon. Oregon State
University, Forest Research Laboratory (Research Bulletin 59).
Hanus, M., Hann, D.W., Marshall, D., 2000. Predicting Height to Crown Base for
Undamaged and Damaged Trees in Southwest Oregon. Oregon State University,
Forest Research Laboratory (RC-29).
Jones, K.B., Heggem, D.T., Wade, T.G., Neale, A.C., Ebert, D.W., Nash, M.S., Mehaffey,
M.H., Hermann, K.A., Selle, A.R., Augustine, S., Goodman, I.A., Pedersen, J.,
Odion, D., Frost, E., Strittholt, J., Jiang, H., Dellasala, D., Moritz, M., 2004. Patterns of
?re severity and forest conditions in the western Klamath Mountains,
California. Conservation Biology 18 (4), 927?936.
Pinheiro J, Bates D, DebRoy S, Sarkar D, and R Development Core Team. 2010. nlme:
Linear and Nonlinear Mixed Effects Models. R package version 3.1-97. Available
from: <http://CRAN.R-project.org/package=nlme>.
Pollet, J., Omi, P.N., 2002. Effect of thinning and prescribed burning on crown ?re
severity in ponderosa pine forests. International Journal of Wildland Fire 11 (1),
1?10.
Prasad, A.M., Iverson, L.R., Liaw, A., 2006. Newer classi?cation and regression tree
techniques: bagging and random forests for ecologic prediction. Ecosystems 9,
181?199.
Prichard, S.J., Peterson, D.L., Jacobson, K., 2010. Fuel treatments reduce the severity
of wild?re effects in dry mixed conifer forest, Washington, USA. Canadian
Journal of Forest Research 40, 1615?1626.
R Development Core Team, 2006. R: A Language and Environment for Statistical
Computing. R Foundation for Statistical Computing, Vienna, Austria.
Raymond, C.L., Peterson, D.L., 2005. Fuel treatments alter the effects of wild?re in a
mixed-evergreen forest, Oregon, USA. Canadian Journal of Forest Research 35
(12), 2981?2995.
Scott, J.H., Reinhardt, E.D., 2001. Assessing Crown Fire Potential by Linking Models
of Surface and Crown Fire Behavior. USDA Forest Service Rocky Mountain
Research Station Research Paper RMRS-RP-29. Fort Collins, CO.
Starker, T.J., 1934. Fire resistance in the forest. Journal of Forestry 32, 462?467.
Stephens, S.L., Moghaddas, J.J., 2005. Silvicultural and reserve impacts on potential
?re behavior and forest conservation: twenty-?ve years of experience from
Sierra Nevada mixed conifer forests. Biological Conservation 125, 369?379.
Thompson, J.R., Spies, T.A., 2009. Vegetation and weather explain variation in crown
damage within a large mixed severity wild?re. Forest Ecology and Management
258, 1684?1694.
Thompson, J.R., Spies, T.A., 2010. Factors associated with crown damage following
recurring mixed-severity wild?res and post-?re management. Landscape
Ecology 25 (5), 775?789.
Thompson, J.R., Spies, T.A., Ganio, L.M., 2007. Reburn severity in managed and
unmanaged vegetation in a large wild?re. Proceedings of the National Academy
of Sciences 104 (25), 10743?10748.
360 J.R. Thompson et al. / Forest Ecology and Management 262 (2011) 355?360Bolgrien, D., Viger, J.M., Chiang, D., Lin, C.J., Zhong, Y., Baker, J., Van Remortel,
R.D., 2000. Assessing landscape conditions relative to water resources in the
western United States: a strategic approach. Environmental Monitoring and
Assessment 64, 227?245.
Kobziar, L.N., McBride, J.R., Stephens, S.L., 2009. The ef?cacy of ?re and fuels
reduction treatments in a Siera Nevada Pine plantation. International Journal of
Wildland Fire 18, 791?801.
Lawler, J.J., White, D., Neilson, R.P., Blaustein, A.R., 2006. Predicting climate induced
range shifts: model differences and model reliability. Global Change Biology 12,
1568?1584.
Liaw, A., Wiener, M., 2002. Classi?cation and regression by random forest. R News 2,
18?22.Van Wagner, C.E., 1977. Conditions for the start and spread of crown ?re. Canadian
Journal of Forest Research 7, 23?34.
Walstad, J.D., 1992. History of the development, use, and management, of forest
resources. In: Hobbs, S.D., Tesch, S.D., Owston, P.W., Stewart, R.E., Tappeiner,
J.C., Wells, G. (Eds.), Reforestation Practices in Southwestern Oregon and
Northern California. Oregon State University Press, Corvallis, OR.
Weatherspoon, C.P., Skinner, C.N., 1995. An assessment of factors associated with
damage to tree crowns from the 1987 wild?res in northern California. Forest
Science 41 (3), 430?451.