Open Journal of Forestry
2012. Vol.2, No.2, 77-87
Published Online April 2012 in SciRes (http://www.SciRP.org/journal/ojf) http://dx.doi.org/10.4236/ojf.2012.22011
Copyright © 2012 SciRes. 77
Predicting Stem Windthrow Probability in a
Northern Hardwood Forest Using a Wind
Intensity Bio-Indicator Approach
Philippe Nolet1,2, Frédérik Doyon1,2, Daniel Bouffard 3
1Insitut des Sciences de la Forêt tempérée, Ripon, Canada
2Université du Québec en Outaouais, Gatineau, Canada
3Insitut Québécois d’Aménagement de la Forêt Feuillue, Ripon, Canada
Email: philippe.nolet@uqo. ca
Received March 2nd, 2012; revised April 1st, 2012; accepted April 8th, 2012
Unlike fire or insect outbreaks, for which a suppression program can be implemented, it is impossible to
prevent a windstorm event or stop it while it is occurring. Reducing stand susceptibility to windstorms
requires a good understanding of the factors affecting this susceptibility. Distinct species- and size-related
differences in stem windthrow susceptibility are difficult to obtain because it is impossible to distinguish
their relative effects from those of wind intensity. Using a damage assessment database (60 20-metre ra-
dius plots) acquired after an exceptional wind storm in Western Quebec in 2007, we developed an ap-
proach in which proportions of windthrown sugar maple poles were used as bio-indicators of wind inten-
sities affecting the plots. We distinguished between single and interactive effects of wind intensity, spe-
cies, stem size, and local basal area on stem windthrow susceptibility. The best logistic regression model
predicting stem windthrow included the wind intensity bio-indicator, species, basal area, and the species
by diameter at breast height (DBH, 1.3 m) interaction. Stem windthrow probability generally increased
with DBH and decreased with basal area. Species wind-firmness was ordered as: yellow birch > sugar
maple = eastern hemlock = American beech > ironwood > basswood = other hardwoods = other soft-
woods. Our method remained an indirect method of measuring wind intensity and its real test would re-
quire a comparison with anemometer measurements during a windstorm. Despite its indirect nature, the
method is both simple and ecologically sound. Hence, it opens the door to conducting similar windthrow
studies in other ecosystems.
Keywords: Windthrow Prediction; Species Wind-Firmness; Tolerant Hardwood Forests; Silviculture
Guidelines
Introduction
Windthrow represents one of the most important catastrophic
natural disturbances in shade-tolerant hardwood stands of North
America (Bormann & Likens, 1979; Canham & Loucks, 1984;
Foster, 1988; Seymour et al., 2002). Mortality from windthrow
in these forests can be comparable to that which results from
single tree mortality (Woods, 2004; Nolet et al., 2007). As with
other natural disturbances, windthrow affects ecosystem ele-
ments, such as stand structural complexity (Han son & Lorimer,
2007; Fukui et al., 2011; Kneeshaw et al., 2011) and soil char-
acteristics (Clinton & Baker, 2000; Simon et al., 2011), which
have key roles in ecosystem processes (Putz et al., 1983). It also
affects forestry operations by reducing timber volume availabil-
ity and by drastically changing harvest schedules, which both
bring economic losses. Unlike fire (Stephens & Ruth, 2005) or
insect outbreaks (e.g. Stedinger, 1984), for which a suppression
program can be implemented to reduce such economic losses, a
windstorm event cannot be prevented nor can it be stopped
while it is occurring. The only action that forest managers can
take is to reduce stand susceptibility to windstorms, which first
requires a good understanding of the factors that affect this
susceptibility. Many studies have addressed this issue, but these
have been mainly conducted for coniferous stands (Huggard et
al., 1999; Ruel, 2000; Ni Dhubhain et al., 2001).
Several factors influence a tree’s probability of being wind-
thrown (Everham & Brokaw, 1996) and such factors operate at
different scales (Ruel, 1995; Boose et al., 2001; Scott &
Mitchell, 2005; Valinger & Fridman, 2011). These range from
the landscape scale (exposure to prevailing storm winds) to the
stand scale (moisture regime, soil thickness, proximity to
openings, stand basal area) and down to the individual stem
(species, stem size, stem health). At the level of the individual
stem, species and size play essential roles in windthrow suscep-
tibility. It has been generally observed that 1) larger stems are
more vulnerable than smaller ones (Peterson, 2007), 2) soft-
wood species are more vulnerable than hardwoods (Foster,
1988; Scott & Mitchell, 2005), and 3) early-successional spe-
cies are more vulnerable than late-successional species (Foster,
1988; Everham & Brokaw, 1996; Rich et al., 2007). In forests,
where many species and diameter classes may be found, such
generalities are not sufficient to develop prevention practices
that are aimed at decreasing stand susceptibility to windstorms.
Disentangl i ng t he E ffe cts of Wind Intensi t y and
Susceptibility Factors
The severity of a windthrow event in a given stand is a
P. NOLET ET AL.
product of the wind intensity that affects the stand and the
various factors that influence stem windthrow susceptibility.
Distinct differences in stem susceptibility to windthrow, which
vary according to species and size, are difficult to obtain be-
cause, for most of the time, it is impossible to distinguish their
relative effects from those of wind intensity and site factors.
For instance, if an eastern hemlock (Tsuga canadensis L. Carr.)
dominated stand is more greatly affected by windthrow during
a storm event than the surrounding hardwood stands, it is very
difficult to determine whether this is a result of 1) more intense
winds at that specific location, 2) this particular species’ sus-
ceptibility, 3) the size of the stems, or 4) simply the site condi-
tions (e.g., soil moisture) at that specific location. To disentan-
gle the effects of wind intensity and susceptibility factors re-
quires a measure of wind intensity that has affected each spe-
cific stand. Given the frequency and unpredictability of such
events, it is almost impossible to obtain direct measures (e.g.,
anemometer readings) of wind intensity during a windstorm.
Canham et al. (2001) have developed an innovative approach
that computes the relative importance of wind intensity and
stem characteristics in estimating the probability that a given
stem will be windthrown. The novelty of this approach relies on
three main aspects: 1) the measure of wind intensity is indirect;
2) the differences in wind intensity that are experienced by
plots (or stands) are relative to one another; and 3) this indirect
and relative wind intensity measure is obtained simultaneously
through the computation of species-specific susceptibility to
windthrow using a global pseudo-optimization (GPO) proce-
dure. Having an indirect and relative measure of wind intensity
(the first two aspects of the approach) is a major innovation
since windthrow causality due to wind intensity can be distin-
guished from the effect of species-specific characteristics, to-
gether with testing the interaction between these variables.
However, we believe that the third aspect of the approach (the
GPO procedure) is weakest since it requires the use of a large
number of parameters in the statistical model (64 in Canham et
al., 2001). To avoid spurious effects that could be incurred by
including many parameters in a model, Anderson et al. (2001)
argue that their numbers should be limited and that models with
30-plus parameters often find little support.
We sought a method that could use the best aspects of the
approach developed by Canham et al. (2001) (1 and 2) and
improve its weakest aspect (3). We investigated a damage as-
sessment database (60 20-metre radius plots) that was acquired
after an exceptional wind storm in Western Quebec (Environ-
ment Canada, 2007). We observed sugar maple poles (i.e.,
stems 9.1 - 19.0 cm diameter at breast height, DBH, 1.3 m
above the ground surface) in all pre-storm stands. This allows
us to implement an idea put forth by Wood (1995), which con-
sists of using the proportion of windthrown stems of a definite
species and size as an indicator of wind intensity. By calculat-
ing the proportion of windthrown sugar maple (Acer saccharum
Marsh.) poles within each plot, we obtain a value that can be
interpreted as an indirect and relative measure of wind intensity.
In other words, we believe that sugar maple poles in our dataset
may be used as a bio-indicator of wind intensity that has af-
fected the plots. Therefore, the objectives of this paper are:
to distinguish between the individual and interactive effects
of wind intensity, species, stem size, and local basal area on
stem susceptib ility to windthrow using a wind bi o-indicator
approach, and
to verify how such results may be taken into account in
selection cuts to prevent eventual loss from windthrow in
northern hardwood forests.
Materials and Methods
Study Area
This study was conducted in the Papineau-Labelle Wildlife
Reserve (46˚13'48''W, 75˚09'55''N) of Quebec, between Lakes
Montjoie and Du Sourd, and about 100 km northeast of Can-
ada’s capital, Ottawa. The area is located in the eastern portion
of Lac du Poisson Blanc landscape unit (Robitaille & Saucier,
1998) of the western sugar maple-yellow birch (Betula al-
leghaniensis Britton) bioclimatic region (Saucier et al., 2011).
The landscape contains numerous hills with elevations < 450 m
a.s.l. and averaging 300 m in height. Mean annual temperature
is 3.7˚C, mean annual precipitation is roughly 1100 mm (in-
cluding 250 mm as snow), and the number of degree days
above 0˚C is 2716 (Environment Canada, 2007). Surficial ge-
ology for the study area is characterized by thin to moderately
thin glacial till composed of metamorphic rocks, such as gneiss,
topped by sandy Dystric Brunisols (GPPC, 2010). The forest
canopy is dominated by sugar maple in association with yellow
birch, American beech (Fagus grandifolia Ehrh.), American
basswood (Tilia Americana L.), ironwood (Ostrya virginiana
Mill. K. Koch), eastern hemlock, and balsam fir (Abies bal-
samea L. Mill.). The forest inventory map, which was produced
for the area by the provincial government, indicates that partial
harvesting was conducted in the stands in 2001. Observations
made on site at the time of sampling confirmed this informa-
tion.
Storm Event
The storm that caused the windthrow event discussed here
had occurred on July 17th 2006. On the day of the storm, the
nearest airport (Maniwaki Airport; 46˚16'W, 75˚59'N) regis-
tered peak wind speeds reaching 61 km·hour–1 (Environment
Canada, 2011), while the weather station at Rouyn-Noranda,
Quebec (48˚03'W, 77˚47'N), registered peak winds of 117
km·hour–1 (Environment Canada, 2011). Given the damage
observed in some stands, peak winds in the study area were
likely closer to the winds observed at Rouyn than those re-
ported at Maniwaki.
Sampling Protocol
Sixty 20-metre radius plots were established within the study
area. A stratified random sampling scheme was applied to ob-
tain a range of windthrow severities; 20 plots were established
in each class of windthrow severity (high, medium, and low
based on visual estimation). These classes were only used to
stratify the sampling scheme and were not used in the ensuing
analysis. To minimize spatial autocorrelation (stand composi-
tion, wind severity) or other spatial dependencies among plots,
plots were separated by a minimum distance of 100 m. In addi-
tion, to avoid pseudo-replication and to maintain interspersion
of disturbance levels, we ensured that there was a change in the
disturbance severity classes in between the plots. We sampled
only mesic sites with gentle slopes (<15%) to minimize the
effects of site differences on windthrow probabilities.
Selection and characterization of windthrown stems followed
(Canham et al., 2001; Woods, 2004). All windthrown trees
Copyright © 2012 SciRes.
78
P. NOLET ET AL.
within plots that had a diameter at breast height (DBH, 1.3 m
above the soil surface before windthrow) greater than 9.1 cm
were included in the study. A stem was considered to be wind-
thrown if it rested at a horizontal angle < 45˚, or if the stem was
broken below the base of the crown. For all windthrown stems,
species and DBH were also recorded.
Data Analysis
As previously mentioned, we calculated a bio-indicator of
wind intensity (BWI) for each sampling plot by dividing the
number of windthrown sugar maple poles (DBH class 9.1 -
19.0 cm) by the total number of sugar maple poles that were
present. We used only plots having at least four sugar maple
poles prior windthrow, which required that eight of the initial
60 plots be excluded. The probability of a stem being wind-
thrown was predicted as a function of four potential variables,
i.e., BWI, species, DBH, and stand basal areal (BA), together
with interactions among these variables. Using the model selec-
tion approach (Anderson et al., 2000; Johnson & Omland,
2004), we compared 22 logistic regression models (using the R
package, R Development Core Team, 2011). Nine models had
only single effects, while thirteen models included two-way
interactions among the explanatory variables. The model selec-
tion approach is especially well-suited to complex systems for
which a common model selection procedure could lead inaccu-
rately to the identification of one “best” model without ac-
counting for the large uncertainty in the model selection itself.
Model performance was verified mainly through Akaike’s In-
formation Criterion (AIC), which was corrected for small sam-
ple sizes (AICc). From this value, we calculated the Akaike
weight to rank the models and, for each model, to evaluate its
probability of being the best option. We did not test for any
three-way interactions, or for any interaction term that included
BA. We could not find any ecological reasons to test interac-
tions including BA, which is a local stand-level variable. BA
was used in some models to verify whether or not its addition
would improve model performance compared to the corre-
sponding models that did not incorporate it.
Once the highest performing and most ecologically sound
model was chosen, we tested its goodness-of-fit in three ways.
First, we verified if the model fit equally well across the range
of predicted probabilities. To do so, we computed and plotted
the observed mean proportion of windthrown stems by classes
of stem-predicted windthrow probabilities. Second, we com-
pared observed windthrow proportions in the plots with the
mean predicted windthrow probabilities. Finally, we compared
observed windthrow proportions to mean predicted windthrow
probabilities by species-DBH (10-cm classes) combinations.
To verify how the model can be useful in decreasing stand
susceptibility to windthrow, we compared the possible loss due
to windthrow after applying two hypothetical partial cut treat-
ments. The first treatment was designed to mimic a typical
single-tree selection cut (SC) in which 30% of the basal area is
harvested equally among species and DBH classes. The second
treatment, a wind-firm optimized partial cut (WOP), was de-
signed to harvest the less wind-firm stems of the stand, as pre-
dicted by our model, until 30% basal area removal was attained.
These treatments were applied to the same and single hypo-
thetical stand, represented by the same structure and composi-
tion as that described in Table 1, except that stem number was
multiplied to achieve a basal area of 26 m2·ha–1. This basal area
is representative of the pre-harvest stands in maple-dominated
forests in Québec. After applying these two treatments on this
initial stand, two theoretical residual stands remained. Then,
employing our model, we computed the expected loss of basal
area in these stands that would result from an eventual wind
storm (relative intensity = .5).
Results
Sugar maple (SM) was the most abundant species in the plots,
followed by American beech (AB) and yellow birch (YB) (Ta-
ble 1). Three other species, viz., ironwood (IW), eastern hem-
lock (EH) and, American basswood (BW), were represented by
at least 20 individual trees in the dataset. All species were pre-
sent in all DBH classes, except for IW, which occurred only in
the two smallest DBH classes. Species with a frequency < 20
were grouped with “other hardwoods” (OH) or “other soft-
woods” (OS). The OH group included red maple (Acer rubrum
L.), white or American ash (Fraxinus americana L.), white or
paper birch (Betula papyrifera Marsh.), trembling aspen
(Populus tremuloides Michx.), and black cherry (Prunus serot-
ina Ehrh.). The OS group included balsam fir, eastern white-
cedar (Thuja occidentalis L.), and white spruce (Picea glauca
Moench Voss). Such species compositions and DBH distribu-
tions were representative of stands in the study area.
Comparing models that used only one explanatory variable
(models 1 to 4; Table 2) clearly showed that BWI was the best
variable for predicting stem windthrow probabilities, followed
in decreasing order by species, DBH, and BA. By gradually
increasing model complexity, we observed that adding species
and DBH (model 8) markedly increased model performance
(decreasing AICc) compared to the use of BWI alone (model 1).
At this step of the analyses, adding BA (model 9) decreased
model performance. The use of BWI*DBH interaction de-
creased model performance slightly (models 10 to 13), while
BWI*species slightly increased it (models 14 to 16). The
DBH*species (models 17 to 19) interaction clearly contributed
the most to model performance. Including both BWI*species
and DBH*species interactions led to the best model perform-
ance (model 21)—slightly better than model 19—but also re-
quired a substantial increase in the number of parameters used
in the model. After the incorporation of interactions in the
models (models 10 to 22), models using BA consistently per-
formed better than the corresponding models not using BA (i.e.,
model 13 vs 12; model 16 vs 15; model 19 vs 18; model 21 vs
20).
Overall, and according to the model weights (Wi in Table 2),
two models (models 19 and 21) were superior to the others;
since model 21 did not clearly outperform model 19, we cannot
affirm that the former was definitely the best model. The plot-
ting of windthrow probability against BWI (with DBH and BA
held constant), according to models 19 and 21 (Figures 1 and
2), revealed two important facts. First, the confidence intervals
were much wider for model 21 than for model 19. Second, the
effect of BWI on windthrow probability was positive for most
species and for both models, but it was negative for the OH
group, according to model 21. We argue that it is impossible
that a stem windthrow probability decreases with wind intensity;
therefore, we can no longer consider this model as valid. We
believe that the increased numbers of parameters in model 21,
compared to model 19, led to an artificial and misleading in-
crease in model performance (beter AICc). Due to these issue s,
t
Copyright © 2012 SciRes. 79
P. NOLET ET AL.
Copyright © 2012 SciRes.
80
Table 1.
Stem density (individuals ·ha–1) by species and diameter classes for the whole dataset (52 plots).
DBH class/species AB BW EH IW OH OS SM YB Total
9.1 - 19.0 cm 231 10 1 64 5 19 606 135 1071
19.1 - 1 - 29.0 cm 123 9 12 9 14 3 623 76 869
29.1 - 39.0 cm 57 3 15 8 2 321 20 426
>39.1 cm 20 2 23 2 6 147 5 205
Total 431 24 51 73 29 30 1697 236 2571
AB = American beech; BW = American basswood; EH = eastern hemlock; IW = ironwood; OH = other hardwood species; OS = other softwood species; SM = sugar
maple; YB = yellow birch.
Table 2.
Comparison of vari ous stem windthrow logistic models.
No Interactions Single effects K Log-likelihood AICc Wi
1 none BWI 2 –852.2 1708.4 0%
2 none DBH 2 –908.7 1821.3 0%
3 none SP 8 –886.4 1788.9 0%
4 none BA 2 –915.8 1835.6 0%
5 none BWI + DBH 3 –846.9 1699.8 0%
6 none BWI + SP 9 –836.0 1690.1 0%
7 none DBH + SP 9 –881.7 1781.4 0%
8 none BWI + DBH + SP 10 –826.4 1672.8 0%
9 none BWI + DBH + SP + BA 11 –826.1 1674.2 0%
10 BWI*DBH SP 9 –831.2 1680.5 0%
11 BWI*DBH DBH + SP 10 –831.2 1682.4 0%
12 BWI*DBH BWI + DBH + SP 11 –828.7 1679.5 0 %
13 BWI*DBH BWI + DBH + SP + BA 12 –824.7 1673.5 0%
14 BWI*SP DBH 10 –834.0 1688.0 0%
15 BWI*SP DBH + SP 17 –818.3 1670.9 0%
16 BWI*SP DBH + SP + BA 18 –813.1 1662.4 0%
17 DBH*SP BWI 9 –829.9 1677.9 0%
18 DBH*SP BWI + SP 17 –812.2 1658.6 3%
19 DBH*SP BWI + SP + BA 18 –809.3 1654.8 20%
20 DBH*SP + BWI*SP SP 24 –805.0 1658.4 3%
21 DBH*SP + BWI*SP SP + BA 25 –800.9 1652.3 70%
22 DBH*SP + BWI*SP BA 18 –811.1 1658.5 3%
AICc = corrected Akaike info rmation criteri on; K = number of model pa rameters; Wi = weight of the model compared to othe r models; B WI = bio -indicator of wind i nt en-
sity; DBH = diameter at breast height; SP = species; BA = basal area.
we concluded model 19 to be our best model.
Model 19 adequately fitted the observed data regardless of
how the data were summarized. First, the model provided valid
results for most of its prediction range (Figure 3). For the
classes of predicted windthrow probability ranging from .0 to .6,
the model closely predicted the observed probabilities. For
probability classes ranging from .6 to .9, the model overesti-
mated stem windthrow probability by about 15%, while fitting
the highest class (.9 to 1.0) very well. Second, the predicted
proportion of windthrown stems by plot was strongly associ-
ated with the observed proportion of windthrown stems (R2
= .72) and the relation is close to 1 (1.17, Figure 4). Th ird, an d
most importantly, the model closely predicted the proportion of
windthrown stems that were observed in most species-DBH
combinations (Figure 5). It fitted the data very well for the
most frequently encountered speces, SM, AB, and YB, although i
P. NOLET ET AL.
Figure 1.
Effect of wind intensity on stem windthrow probability according to model 21, with DBH fixed at 30 cm and
BA at 20 m2·ha–1. Grey lines indicate confidence intervals (95%) associated with BWI parameter. AB,
American beech; BW, American basswood; EH, eastern hemlock; IW, ironwood; OH, other hardwoods; OS,
other softwoods; SM, sugar maple; YB, yellow birch.
Figure 2.
Effect of wind intensity on stem windthrow probability according to model 19, with DBH fixed at 30 cm and
BA at 20 m2·ha–1. Grey lines indicate confidence intervals (95%) associated with BWI parameter. AB,
American beech; BW, American basswood; EH, eastern hemlock; IW, ironwood; OH, other hardwoods; OS,
other softwoods; SM, sugar maple; YB, yellow birch.
Figure 3.
Observed proportion of windthrown stems as a function of classes of
predicted stem windthrow probabilities. Bars represent the observed
proportion of windthrown trees as a function of the predicted wind-
throw probability. Values above each bar indicate the number of ob-
servations in that class. The diagonal line (1:1) indicates a perfect fit
between the observed proportion and the expected proba b ility.
the last one was overestimated in the highest DBH class. The
model also fitted less commonly occurring species such as BW,
OS, IW, and EH (except for the lowest DBH class) quite well.
The main problem in terms of goodness-of-fit was observed
with the OH species group for which the model overestimated
(up to 30%) the windthrow probability in some DBH classes,
while underestimating (up to 40%) it in others.
Generally, DBH had a positive effect on the probability of a
tree being windthrown (Figure 6). This relation, however, did
not appear to hold for YB and BW. For BW, this counter-intui-
tive result could be attributed to low numbers of observations in
the various DBH classes. The same interpretation cannot be
invoked with respect to YB. Moreover, both upper and lower
confidence limits also showed the same trend. For SM, the
DBH effect on windthrow probability was weak, while it had
slightly more noticeable effects on AB and EH. The DBH ef-
fect was most obvious for the OH and OS species groups, al-
though confidence intervals were wide due to the low number
of observations.
Copyright © 2012 SciRes. 81
P. NOLET ET AL.
Figure 4.
Observed proportion of windthrown stems in each plot as a function of
the mean predicted stem windthrown probability in each plot. The
diagonal dotted line indicates a 1:1 correspondence between observed
and expected.
Susceptibility to windthrow clearly differed among species
(Figure 7). YB was definitively the least vulnerable to wind-
throw, except at lower DBH for which its susceptibility is
comparable to those of SM, AB, IW and EH. The differences in
windthrow susceptibility among these latter species were not
clearly noticeable. BW, OH and OS were definitely most sus-
ceptible to windthrow, even though DBH did not affect their
susceptibility eq ually.
Model 19 predicted a negative effect of BA on susceptibility
to windthrow. For a SM tree with a DBH of 30 cm, the prob-
ability of being windthrown decreases from 50% at a BA of 10
m2·ha–1 to 36% at a BA of 25 m2·ha–1 under a BWI of .5 (Fig-
ure 8). At the stand level, however, it suggests that the loss due
to windthrow of two hypothetical pure SM stands (with only 30
cm DBH trees) with basal areas of 10 m2·ha–1 and 25 m2·ha–1
would be 5 m2·ha–1 and 9 m2·ha–1, respectively. This would
mean that the negative effect of BA on stem windthrow suscep-
tibility was not strong enough to counteract the fact that, when
a windstorm strikes, the greater the number of trees there in a
stand, the greater the trees that are windthrown.
The model may be used to rank the individual stems of a
stand that are the most likely to be windthrown after a wind-
storm event. The loss due to an eventual windstorm in a stand
that was managed using a partial cut designed to harvest the
less wind-firm stems (WOP) would be about 1 m2·ha–1 lower
than in a similar stand that had been managed using a typical
Figure 5.
Observed (Obs) proportions of windthrown ste ms in species-DBH combinations compared to the mean predicted (Pred) stem windthrow probabilities.
AB, American beech; BW, American basswood; EH, eastern hemlock; IW, ironwood; OH, other hardwoods; OS, other softwoods; SM, sugar maple;
YB, yellow birch.
Copyright © 2012 SciRes.
82
P. NOLET ET AL.
Figure 6.
Effect of DBH on stem windthrow probability according to model 19 with BWI fixed at .5 and basal area at
20 m2·ha–1. Grey lines indicate confidence intervals (95%) associated with DBH parameters. AB, American
beech; BW, American basswood; EH, eastern hemlock; IW, ironwood; OH, other hardwoods; OS, other
softwoods; SM, sugar maple; YB, yellow birch.
Figure 7.
Comparisons among species of the effect of DBH on stem windthrow
probability according to model 19 with BWI fixed at .5 and BA at 20
m2·ha–1. The curves, which are the same than those shown in Figure 6,
are presented without the confidence intervals to facilitate the com-
parison among species. AB, American beech; BW, American basswood;
EH, eastern hemlock; IW, ironwood; OH, other hardwoods; OS, other
softwoods; SM, sugar maple; YB, yellow birch.
tree-selection cut (Figure 9). This result was valid only for the
example we used (see methodology). Results could vary de-
pending on the initial stand characteristics and storm intensity.
Discussion
Wood (1995) first proposed the use of species-DBH combi-
nations as surrogates for wind intensity. Rich et al. (2007)
partly applied the idea a few years later building a surrogate
that implied many species having a similar DBH distribution.
To our knowledge, the present study was the first to implement
Wood’s idea in its exact form. To be used as a surrogate for a
specific disturbance agent, a bio-indicator should exhibit bal-
anced sensitivity to the disturbance agent. A bio-indicator that
Figure 8.
Effect of BA on stem windthrow probability according to model 19,
with BWI fixed at 0.5 and DBH at 30 cm. Grey lines indicate confi-
dence intervals (95%) associated to BA parameter.
is too sensitive will fail to distinguish differences at higher
values within the spectrum of disturbance intensities, while a
lack of sensitivity will lead to a similar problem within the
lower end of that same spectrum. Since the proportion of SM
poles that were windthrown varied from .0 to .7 in our plots, it
is reasonable to believe that SM poles exhibited sufficient sen-
sitivity to windthrow to be used as a bio-indicator. Even though
BWI (bio-indicator of wind intensity) was the most important
variable in our model, we cannot verify how it was related to
the actual wind intensity experienced by the plots. In other
words, BWI seems to be related to wind intensity, but we can-
not demonstrate it. The mathematical approach developed by
(Canham et al., 2001) has the same limitations. In addition,
using the approach developed by the aforementioned authors
would have required the estimation of 85 parameters (60 for the
Copyright © 2012 SciRes. 83
P. NOLET ET AL.
Figure 9.
Comparison of predicted loss due to windthrow after a typical tree
selection partial cut (TSP) and after a wind-firm-optimized partial cut
(WOP). The initial stand composition and structure correspond to the
average of all plots in the da taset, standardized at 26 m2·ha–1.
plots, 24 for the species, and 1 for BA). Since we already ob-
served that the use of 26 parameters led to over-fitting of
model 21, it is likely that using 85 would have led to a simila r
problem. Anderson et al. (2000) have urged ecologists not to
model the data; rather, they should attempt to isolate informa-
tion in the data from the noise. The wind bio-indicator ap-
proach that was developed in this study is in concordance
with this advice.
According to our results, we could order species from most
to least wind-firm as follows: yellow birch (YB) > sugar maple
(SM) = eastern hemlock (EH) = American beech (AB) > iron-
wood (IW) > American basswood (BW) = other hardwoods
(OH) = other softwoods (OS). Likewise, we have summarized
results from several other northern hardwood forests (Table 3)
to facilitate comparisons with our results. Other studies have
observed that yellow birch (YB) was the least, or among the
least, vulnerable species (Canham et al., 2001; Woods, 2004;
Hanson & Lorimer, 2007) (Table 3). Peterson (2007), who
studied many sites, found that the ranking of YB susceptibility
was inconsistent when compared with other species. Sugar
maple (SM) was the second most resistant species according to
our study. Incidentally, Peterson (2007) observed the same
classifying order for SM in his three study sites reported in
Table 3. Canham et al. (2001) reported SM and YB as the most
resistant species to windstorm, as in concordance with our re-
sults. American beech (AB) exhibited susceptibility very simi-
lar to SM (Woods, 2004; Person, 2007; this study). Only Can-
ham et al. (2001) observed a much higher susceptibility for AB
than for SM and this was probably due to the beech bark dis-
ease (Papaik et al., 2005). Eastern hemlock (EH) shows the
least consistent susceptibility ranking among the studies. While
its susceptibility was similar to that of SM in our study, hem-
lock appeared to be less vulnerable (than SM) according to
Woods (2004), and more vulnerable according to Canham et al.
(2001) and Hanson & Lorimer (2007). In the studies reported in
Table 3, Peterson (2007) found that EH had a susceptibility
similar to that of SM in a first site, lower susceptibility in a
second site, and higher susceptibility in a third site. It was
noteworthy that EH was always more wind-firm than other
softwood species. When present, OS was always the most vul-
nerable species, according to the rankings in Table 3. In our
study, the OH species group was as vulnerable as OS. Canham
et al. (2001) and Peterson (2007) also observed that this species
group was usually highly vulnerable to windstorm. However,
Peterson (2007) also observed a very high wind-firmness for
species with strong wood, such as hickories (Carya species)
and white oak (Quercus alba L.).
Overall, despite the variety of ecosystems that were studied,
the variety of the windstorms that had affected them, and the
variety of methods that were used to study them, some general
conclusions may be drawn. First, among the most frequently
occurring species in the northern hardwoods, YB seemed to be
the most wind-firm, followed by SM, EH, and AB. Second,
softwood species were the less wind-firm most of the time,
except for EH. Clearly, a species’ successional status was not a
good indicator of wind-firmness. The higher wind-firmness of
YB and SM compared to that of EH and AB was such an ex-
ample, since the latter are recognized as late-successional spe-
cies when compared to the former (Doyon et al., 1998). Also,
Table 3.
Comparison of species wind- fir mness relative ranking according to various studies in no rt h e as t er n ha r d wo o d f o re s t s o f N o rt h America.
Species wind-firmness relative ranking1 according to
Species2 This study Canham et al.
(2001) Hanson &
Lorimer (2007)Peterson (2007)
Tionesa’94 Peterson (2007)
TexHill Peterson (2007)
Gould Woods (2004)
AB 2 3 - 2 - - 2
BW 4 - - - - - -
EH 2 2 2 1 3 2 1
IW 3 - - - - - -
OH 4 2 and 4 - 2 1 and 3 3 -
OS 4 4 - - - 4 -
SM 2 1 1 2 2 2 2
YB 1 1 1 3 - 1 1
1Species wind-firmness rank was not always explicitly specified by the authors , so we attribu ted a ran king using figures and tab les provid ed i n th e articles. Althoug h some
species rankin gs could be es tablished easily, others required some subjectivity. A ranking of 1 meant that the species was considered the most wind-firm for the study or
site; 2AB = American beech; BW= American basswood; EH = eastern hemlock; IW = ironwood; OH = other hardwoods; OS = other softwoods; SM = sugar maple; YB =
ellow birch. y
Copyright © 2012 SciRes.
84
P. NOLET ET AL.
Peterson (2007) observed that, hickories and sugar maple, con-
sidered early- and late-successional species, respectively, pre-
sent similar traits (deep rooting and strong wood) that result in
high windfirmness. As outlined by (Nolet et al., 2008), it is
more appropriate to use species-specific traits to explain eco-
logical phenomena (in this case, species wind-firmness) than
broad and subjective categories such as early- or late-succes-
sional status.
As observed in most studies (see Everham & Brokaw, 1996),
stem windthrow probability increased with DBH for most spe-
cies included in our study. In a study using a methodology
similar to that applied in this study, but driven in a completely
different ecosystem, (Rich et al., 2007) observed a positive
relationship between wind intensity and the nine species that
they studied. In our study, the strength of the relationships was
positive and relatively weak for SM, AB, and EH, while it was
positive and strong for OH and OS. Canham et al. (2001) also
observed that the DBH effect on windthrow probability was
stronger for less wind-firm species. However, one of the most
surprising results of the present study was certainly the negative
effect of DBH on windthrow probability for YB. We believe
that YB was a highly wind-firm species across the whole range
of DBH that was studied. Further, we observed a higher wind-
throw probability for small stems only because they were more
likely to be caught in a domino effect due to a large stem falling.
This domino effect, which is often observed in the field (al-
though very difficult to identify and quantify precisely), af-
fected not only YB but all of the species. For less wind-firm
species, however, the domino effect would not lead to a nega-
tive relationship between windthrow probability and DBH.
Although species and DBH had effects on stem windthrow
probabilities, their interaction with wind intensity was not per-
ceivable in our dataset. Therefore, according to our model, the
relative rank of species wind-firmness was not influenced by
wind intensity. This result contradicted Canham et al. (2001),
whose model relied on an interaction between wind intensity,
DBH, and species. In our case, the absence of any interaction
between wind intensity and species on one hand, and wind
intensity and DBH on the other hand, decreased the complexity
of the model and simplified its potential use for silvicultural
prescriptions at the site level or for planning strategies at the
landscape level (e.g., Papaik & Canham, 2006).
Everham and Brokaw (1996) reported many studies that have
observed more windthrow damage in thinned stands than in
unmanaged stands. This phenomenon may be generally attrib-
uted to the wind’s ability to penetrate more thoroughly into
stands that have been partially harvested, thus increasing the
force exerted on trees (Peltola et al., 1999). Everham and Bro-
kaw (1996) also reported studies that have observed less effect
or no effect at all of thinning on stand windthrow damage. It
appeared that, when a positive effect of thinning on wind-
firmness is observed, it is probably because the weakest trees
had already been harvested during the thinning treatment. The
methodology used in this study presents the advantage of not
relying on a managed vs unmanaged comparison since it di-
rectly assessed the effect of surrounding densities on stem
windthrow probability. Our results indicated that tree wind-
throw probability was slightly higher for stems surrounded by
lower BA (i.e., about 1% increase in windthrow probability for
a BA reduction of 1 m2·ha–1). Given that this negative effect of
BA on stem windthrow probability is weak, denser stands are
expected to exper i e n c e more a bsolute damage even though their
relative damage is lower. However, the strength of the rela-
tionship (BA vs wind-firmness) may have been influenced by
the time that had elapsed between the last partial cut and the
windstorm event (5 years). Foster (1988), among others,
pointed out the importance of elapsed time since thinning when
he found increased damage only in recently thinned stands.
Consequently the BA-wind-firmness relationship could have
been stronger or weaker if time elapsed since the partial cut had
been less or longe r.
There have been many studies published (see Table 3) on
windthrow in northern hardwood forests, but very few, if any,
have discussed how their findings could be used to decrease
damage attributable to windthrow. In other parts of the world,
however, it is a highly developed research topic (see Savill,
1983), especially in softwood stands. Observing greater and
more frequent loss in softwood stands may have influenced the
research focus. With the example shown in Figure 9, we dem-
onstrated that, when compared to a typical selection cut, a par-
tial cut that is designed to harvest less wind-firm species could
decrease by about 1 m2·ha–1 the loss that is due to windthrow.
This quantity should be viewed as an order of magnitude for
mid-intensity storms; in a very high intensity storm (where all
the trees fall) or in a very low intensity storm, there would be
no difference in damage between typical selection cuts and
wind-optimized partial cuts. In Québec, northern hardwood
stands generally grow at a rate of about 0.25 m2·year–1 after a
commercial selection cut (Forget et al., 2007). This means that
preventing a loss of 1 m2·year–1 results in a savings of four
years of growth and, in turn, corresponds to about one-sixth of
harvest rotation in these stands. Thus, we believe that using a
wind-optimized partial cut may be a valuable strategy, espe-
cially in parts of the landscape that are more prone to wind-
storm damage. The fact that species wind-firmness rankings do
not appear to be influenced by wind intensity, as shown by our
results, and that they do not appear to be influenced by site
factors, as shown by Peterson (2007), should encourage forest
managers to adopt such strategies, despite the stochasticity of
windstorm effects.
Conclusion
The bio-indicator method used in this paper for estimating
wind intensity was an effective way of isolating the relative
effects of species, size, and density on stem windthrow prob-
ability, and of verifying the possible interaction among these
variables. It remains an indirect method of measuring wind
intensity and its real test would require a comparison with
anemometer measurements during a windstorm. Unfortunately,
such validation is unlikely because it would require knowing in
advance where a windstorm would strike to install a network of
anemometers. Despite its indirect nature, the method presented
is both simple and ecologically sounded. Because of these
characteristics, we believe it could be used in other circum-
stances and other ecosystems. For instance, many jurisdictions
have implemented permanent plot inventories to describe the
forests under their governance. If the occurrence of windthrow
is tallied during these inventories, plots that experienced them,
and which contain a certain number of the same specific spe-
cies-DBH combinations, could be used to conduct studies
similar to the one presented in this paper. Hence, the method
opens the door to the conducting of many regional windthrow
studies.
Copyright © 2012 SciRes. 85
P. NOLET ET AL.
Finally, even though we are confident of the model devel-
oped in this study, we have recognized that it is incomplete.
From our field experiences, we have identified at least two
variables that could improve the model performance; both
variables were related to the domino effect discussed earlier.
First, we think that the spatial distribution of trees within stands
is important. For example, a small SM has a greater chance of
being thrown down by another tree if it is close to a few big
stems of a low wind-firm species, than if it is close to small
stems of a high wind-firm species. Second, while it is obvious
that some trees are not thrown down directly by the wind, but
indirectly toppled by the fall of other trees, this information was
impossible to gather with certainty in the field. Field techniques
that would facilitate precise pre-storm tree mapping and stem
fall chronologies during the storm should be developed. The
information that is obtained could be incorporated into a statis-
tical model to increase its performance. The effects of wind-
storms on forest stands would probably then appear less sto-
chastic.
Acknowledgements
We are grateful to R. Pouliot, J. Poirier and S. Delagrange
for assistance in the field and to T. L. Bidot and W. Parsons for
English editing. We also thank C.D. Canham for enlightening
discussions during the preparation of the manuscript. Financial
support was provided by the Programme de mise en valeur des
ressources du milieu forestier (Ministère des Ressources
naturelles et de la Faune du Québec).
REFERENCES
Anderson, D. R., Burnham, K. P., & Thompson, W. L. (2000). Null
hypothesis testing: Problems, prevalence, and an alternative. Journal
of Wildlife Management, 64, 913-923. doi:10.2307/3803199
Anderson, D. R., Link, W. A., Johnson, D. H., & Burnham, K. P.
(2001). Suggestions for presenting the results of data analyses.
Journal of Wildlife Management, 65, 373-378. doi:10.2307/3803088
Boose, E. R., Chamberlin, K. E., & Foster, D. R. (2001). Landscape
and regional impacts of hurricanes in New England. Ecological
Monographs, 71, 27-48.
doi:10.1890/0012-9615(2001)071[0027:LARIOH]2.0.CO;2
Bormann, F. H., & Likens, G. E. (1979). Catastrophic disturbance and
the steady state in northern hardwood forests. American Scientist, 67,
660-669.
Canham, C. D., & Loucks, O. L. (1984). Catastrophic windthrow in the
presettlement forests of Wisc onsin. Ecology, 65, 803-809.
doi:10.2307/1938053
Canham, C. D., Papaik, M. J., & Latty, E. F. (2001). Interspecific varia-
tion in susceptibility to windthrow as a function of tree size and
storm severity for northern temperate tree species. Canadian Journal
of Forest Research, 31, 1-10. doi:10.1139/x00-124
Clinton, B. D., & Baker, C. R. (2000). Catastrophic windthrow in the
southern Appalachians: Characteristics of pits and mounds and initial
vegetation responses. Forest Ecology and Management, 126, 51-60.
doi:10.1016/S0378-1127(99)00082-1
Doyon, F., Bouchard, A., & Gagnon, D. (1998). Tree productivity and
successional status in Québec northern hardwoods. Ecoscience, 5,
222-231.
Environment Canada (2 011 ). Climate data online.
http://www.climate.weatheroffice.ec.gc.ca/climateData/canada_e.html
Everham, E. M. I., & Brokaw, N. V. L. (1996). Forest damage and
recovery from catastrophic wind. Botanical Review, 62, 113-185.
doi:10.1007/BF02857920
Forget, E., Nolet, P., Doyon, F., Delagrange, S., & Jardon, Y. (2007).
Ten-year response of northern hardwood stands to commercial selec-
tion cutting in southern Quebec, Canada. Forest Ecology and Man-
agement, 242, 764-775. doi:10.1016/j.foreco.2007.02.010
Foster, D. R. (1988). Species and stand response to catastrophic wind in
central New England, USA. Journal of Ecology, 76, 135-151
doi:10.2307/2260458.
Fukui, D., Hirao, T., Murakami, M., & Hirakawa, H. (2011). Effects of
treefall gaps created by windthrow on bat assemblages in a temperate
forest. Forest Eco logy and Management, 261, 1546-1552.
doi:10.1016/j.foreco.2011.02.001
Hanson, J. J., & Lorimer, C. G. (2007). Forest structure and light re-
gimes following moder ate wind storms: Implications for multi-cohort
management. Ecological Applications, 17, 1325-1340.
doi:10.1890/06-1067.1
Huggard, D. J., Klenner, W., & Vyse, A. (1999). Windthrow following
four harvest treatments in an Engelmann spruce—Subalpine fir forest
in southern interior British Columbia, Canada. Canadian Journal of
Forest Research, 29, 1547-1556. doi:10.1139/x99-135
Johnson, J. B., & Omland , K. S. (2004). Model selection in ecology and
evolution. Trends in Ecology & Evolution, 19, 101-108.
doi.org/10.1016/j.tree.2003.10.013
Kneeshaw, D. D., Harv ey, B. D., Reyes, G. P., Caro n, M. N ., & Barlo w,
S. (2011). Spruce budworm, windthrow and partial cutting: Do dif-
ferent partial disturbances produce different forest structures? Forest
Ecology and Management, 262, 482-490.
doi:10.1016/j.foreco.2011.04.014
GPPC. Le groupe de travail sur les Pédo-paysages du Canada (2010).
Pédo-paysage du Canada version 3.2. Agriculture et Agroalimentaire
Canada (carte numérique et base de données à l'échelle de 1/1
million).
Ni Dhubhain, A., Walshe, J., Bulfin, M., Keane, M., & Mills, P. (2001).
The initial development of a windthrow risk model for Sitka spruce
in Ireland. Forestry, 74, 161-170. doi:10.1093/forestry/74.2.161
Nolet, P., Delagrange, S., Bouffard, D., Doyon, F., & Forget, E. (2008).
The successional status of sugar maple (Acer saccharum), revisited.
Annals of Forest Science, 65, 208. doi:10.1051/forest:2007091
Nolet, P., Hartmann, H., Bouffard, D., & Doyon, F. (2007). Predicted
and observed sugar maple mortality in relation to site quality indica-
tors. Northern Journal of Applied Forestry, 24, 258-264.
Papaik, M. J., & Canham, C. D. (2006). Species resistance and com-
munity response to wind disturbance regimes in northern temperate
forests. Journal of E cology, 94, 1011-1026.
doi:10.1111/j.1365-2745.2006.01153.x
Papaik, M. J., Canham, C. D., Latty, E. F., & Woods, K. D. (2005).
Effects of an introduced pathogen on resistance to natural distur-
bance: Beech bark disease and windthrow. Canadian Journal of
Forest Research, 35, 1832-1843. doi:10.1139/x05-116
Peltola, H., ki, S., Nen, H., & Ikonen, V. P. (1999). A mechanistic
model for assessing the risk of wind and snow damage to single trees
and stands of Scots pine, Norway spruce, and birch. Canadian Jour-
nal of Forest Research, 29, 647-661. doi:10.1139/x99-029
Peterson, C. J. (2007). Consistent influence of tree diameter and species
on damage in nine eastern North America tornado blowdowns. For-
est Ecology and Management, 250, 96-108.
doi:10.1016/j.foreco.2007.03.013
Putz, F. E., Coley, P. D., Lu, K., Montalvo, A., & Aiello, A. (1983).
Uprooting and snapping of trees: Structural determinants and eco-
logical consequences. Canadian Journal of Forest Research, 13,
1011-1020. doi:10.1139/x83-133
R Development Core Team (2004). R Development Core Team, R: A
language and environment for statistical computing. R Foundation
for Statistical Computing, Vienna.
Rich, R. L., Frelich, L. E., & Reich, P. B. (2007). Wind-throw mortality
in the southern boreal forest: Effects of species, diameter and stand
age. Journal of Ecology, 95, 1261-1273.
doi:10.1111/j.1365-2745.2007.01301.x
Robitaille, A., & Saucier, J.-P. (1998). Paysages régionaux du Québec
méridional. Les Publications du Québec, Québec.
Ruel, J. C. (1995). Understanding windthrow: Silvicultural implications.
The Forestry Chronicle, 71, 435-445.
Ruel, J. C. (2000). Factors influencing windthrow in balsam fir forests:
Copyright © 2012 SciRes.
86
P. NOLET ET AL.
Copyright © 2012 SciRes. 87
From landscape studies to individual tree studies. Forest Ecology
and Management, 135, 169-178.
doi:10.1016/S0378-1127(00)00308-X
Saucier, J.-P., Grondin, P., Robitaille, A., & Bergeron, J. F. (2011).
Carte des zones de végétation et des domaines bioclimatiques du
Québec. Ministère des Ressources naturelles, de la Faune et des
Parcs, Direction des inventaires forestiers.
Savill, P. S. (1983). Silviculture in windy climates. Forestry Abstracts,
44, 473-488.
Scott, R. E., & Mitchell, S. J. (2005). Empirical modelling of wind-
throw risk in partially harvested stands using tree, neighbourhood,
and stand attributes. Forest Ecology and Management, 218, 193-209.
doi:10.1016/j.foreco.2005.07.012
Seymour, R. S., White, A. S., & DeMaynadier, P. G. (2002). Natural
disturbance regimes in northeastern North America—Evaluating sil-
vicultural systems using natural scales and frequencies. Forest Ecol-
ogy and Management, 155, 357-367.
doi:10.1016/S0378-1127(01)00572-2
Simon, A., Gratzer, G., & Sieghardt, M., (2011). The influence of
windthrow microsites on tree regeneration and establishment in an
old growth mountain forest. Forest Ecology and Management, 262,
1289-1297. doi:10.1016/j.foreco.2011.06.028
Stedinger, J. R. (1984). A Spruce budworm-forest model and its impli-
cations for suppression programs. Forest Science, 30, 597-615.
Stephens, S. L., & Ruth, L. W. (2005). Federal forest-fire policy in the
United States. Ecological Application s, 15, 532-542.
doi:10.1890/04-0545
Valinger, E., & Fridman, J. (2011). Factors affecting the probability of
windthrow at stand level as a result of Gudrun winter storm in south-
ern Sweden. Forest Ecology and Management, 262, 398-403.
doi:10.1016/j.foreco.2011.04.004
Wood, C. J. (1995). Understanding wind forces on trees. In M. P.
Coutts, & J. Grace (Eds.), Wind and trees (pp. 133-164). Ca mbridge:
Cambridge University Pres s.
Woods, K. D. (2004). Intermediate disturbance in a late-successional
hemlock-northern hardwood forest. Journal of Ecology, 92, 464-476.
doi:10.1111/j.0022-0477.2004.00881.x.