Ecological Applications, 18(7), 2008, pp. 1652–1663
? 2008 by the Ecological Society of America
TREE MORTALITY FOLLOWING PARTIAL HARVESTS IS DETERMINED
BY SKIDDING PROXIMITY
H. C. THORPE,1S. C. THOMAS, AND J. P. CASPERSEN
Faculty of Forestry, University of Toronto, 33 Willcocks St., Toronto, Ontario M5S3B3 Canada
habitat and ecological services provided by managed forest stands by better emulating natural
disturbances. The potential for elevated mortality of residual trees following such harvests
remains a critical concern for forest managers, and may present a barrier to more widespread
implementation of the approach. We used a harvest chronosequence combined with
dendrochronological techniques and an individual-based neighborhood analysis to examine
the rate and time course of residual-tree mortality in the first decade following operational
partial ‘‘structural retention’’ harvests in the boreal forest of Ontario, Canada. In the first year
after harvest, residual-tree mortality peaked at 12.6 times the preharvest rate. Subsequently,
mortality declined rapidly and approached preharvest levels within 10 years. Proximity to skid
trails was the most important predictor both of windthrow and standing death, which
contributed roughly equally to total postharvest mortality. Local exposure further increased
windthrow risk, while crowding enhanced the risk of standing mortality. Ten years after
harvest, an average of 10.5% of residual trees had died as a result of elevated postharvest
mortality. Predicted cumulative elevated mortality in the first decade after harvest ranged from
2.4% to 37% of residual trees across the observed gradient of skid trail proximity, indicating
that postharvest mortality will remain at or below acceptable rates only if skidding impacts are
minimized. These results represent an important step toward understanding how elevated
mortality may influence stand dynamics and habitat supply following moderate-severity
disturbances such as partial harvests, insect outbreaks, and windstorms.
Recently developed structural retention harvesting strategies aim to improve
analyses; partial harvest; Picea mariana; structural retention; tree mortality.
alternative silviculture; black spruce; boreal forest; dendrochronology; neighborhood
In recent decades, silvicultural alternatives to clear-
cutting have been developed and implemented in forest
regions worldwide (e.g., Franklin et al. 1997, Harvey et
al. 2002, Lindenmayer and McCarthy 2002, Keeton
2006). These practices generally attempt to emulate
aspects of natural disturbances and retain key structural
characteristics of late-seral forests (e.g., Franklin et al.
1997). Predicated on the assumption that retaining
features such as live trees and coarse woody debris will
permit the maintenance of biodiversity and ecological
function in managed forests, ‘‘structural retention’’
represents the major means through which recent
advances in ecological understanding have been incor-
porated into modern forest management. Although
early results show improved maintenance of species
diversity (e.g., Lance and Phinney 2001, Deans et al.
2005, Dovc ˇ iak et al. 2006) and large growth responses of
residual trees (e.g., Latham and Tappeiner 2002, Bebber
et al. 2004, Thorpe et al. 2007), the general success or
failure of retention treatments has yet to be established.
Evaluating these novel practices requires long-term,
detailed information on stand development following
harvest, particularly on rates of postharvest tree
Disturbance is a fundamental driver of forest ecosys-
tems, exhibiting strong influence on stand structure
(Hanson and Lorimer 2007, McCarthy and Weetman
2007) and species composition (Canham et al. 2001,
Papaik and Canham 2006, Rich et al. 2007). Distur-
bances vary widely in intensity, from catastrophic fires
to gap-phase processes. Although research has tradi-
tionally focused on these extremes (Seymour et al. 2002),
moderate-severity disturbances have been gaining in-
creasing attention (e.g., Hanson and Lorimer 2007).
Such disturbances, which include windstorms and insect
outbreaks, are analogous to partial harvests insofar as
they ‘‘retain’’ a substantial component of residual trees.
These trees are likely to experience stress related to
increased exposure, microclimatic changes (e.g., Hei-
thecker and Halpern 2006), and reduced water avail-
ability (e.g., Liu et al. 2003), all of which may elevate
tree mortality rates. If this is the case, moderate-severity
disturbances might influence forest dynamics over much
longer time scales than is currently assumed. Previous
studies have documented higher mortality risk for trees
located on gap edges compared to forest-interior trees
(Young and Hubbell 1991, Lin et al. 2004), but research
Manuscript received 12 October 2007; revised 7 March 2008;
accepted 19 March 2008. Corresponding Editor: J. A. Antos.
has not generally considered the potential for elevated
tree mortality following intermediate-scale natural
disturbances (but see Kulakowski and Veblen 2003,
Taylor and MacLean 2007).
The matter of elevated tree mortality after partial
harvest, in contrast, has been gaining research attention
over the past decade. Reported rates of postharvest
mortality span the entire possible range, from losses of
,1% to 100% of residual trees (Jo ¨ nsson et al. 2007,
reviewed in Thorpe and Thomas 2007, Bladon et al.
2008). This variation is in large part attributable to
differences in time scales and retention levels considered.
Quantifying mortality risk at the level of the individual
tree across gradients of retention avoids potential
confounding effects of stand-level analyses (Caspersen
2006), and allows for more general conclusions regard-
ing rates and causes of elevated mortality.
The use of neighborhood models to analyze popula-
tion dynamics at the individual-tree level across ranges
of small-scale, spatially explicit factors is becoming
increasingly common (Canham and Uriarte 2006).
Previous studies have employed neighborhood models
to explore tree growth (e.g., Weiner 1984, Canham et al.
2004, Canham et al. 2006), but to our knowledge this is
the first study to employ an individual-based spatial
neighborhood analysis to examine postharvest tree
mortality. Existing individual-based simulation models
such as SORTIE-ND (available online)2assume that
increased resource levels following disturbance always
act to increase tree survivorship, and thus assume
implicitly that elevated postharvest mortality does not
occur. Incorporation of spatially explicit predictors of
postharvest mortality into individual-based simulation
models is thus essential for accurate forecasting of stand
dynamics and development under alternative silvicul-
In this study we employ dendrochronological methods
and a harvest chronosequence to quantify mortality
before, during, and after partial harvests, and address
the following questions: (1) Does partial harvesting
increase the risk of mortality for residual trees? (2) If so,
what is the magnitude and time course of increased
mortality? (3) How is postharvest mortality risk
influenced by the size of individual trees and their
proximity to skid trails, exposure to wind, and/or
The clay belt is a major physiographic region that
covers 125000 km2of northern Ontario and Que ´ bec,
Canada. Created by glacial lake deposits, the region is
characterized by flat topography and poorly drained
organic soils. Forests are dominated by black spruce
(Picea mariana (Mill.) B.S.P.) and understorys are
composed of ericaceous plants such as Rhododendron
groenlandicum (Labrador tea), and bryophytes Pleuro-
zium schreberi and Sphagnum species. The study area,
bounded by 488510–508120N and 798490–808420W, is
near Cochrane, Ontario, in the north-central portion of
the clay belt, and has a mean annual temperature of
0.68C and average annual precipitation of 880 mm,
including nearly 3 m of snow (Environment Canada
Study sites were treated with Harvest with Advance
Regeneration Protection (HARP), a silvicultural system
developed for uneven-aged black spruce stands (Fig. 1).
HARP treatments are characterized by alternating clear-
cut strips (5–7 m wide) where harvest equipment travels,
and partial-cut strips (5–9 m wide) in which diameter-
limit cutting (generally ;12 cm) is carried out. Some
larger trees are also retained within partial-cut strips to
meet biodiversity guidelines (OMNR 2001), and har-
vesting takes place in winter to protect the organic soils.
From June to August 2005, we sampled 18 stands
(hereafter ‘‘cutblocks’’) that were harvested between
1995 and 2004 and represented a decade-long harvest
chronosequence (including all years but 2001; HARP
was not carried out in the study area during that year).
We selected sites to ensure that cutblocks from the same
harvest year were spatially interspersed among stands
from other harvest years.
Within each cutblock we established three circular, 20
m radius plots, spaced 300–400 m apart, and measured
diameter, species, and class of all dead and live stems .5
cm diameter at breast height (dbh; ‘‘breast height’’ is 1.3
of northern Ontario, Canada. Photo courtesy of Abitibi-
Bowater Incorporated, Montreal, Que ´ bec, Canada.
Partial harvesting in the black spruce boreal forest
October 20081653TREE MORTALITY AFTER PARTIAL HARVEST
m). The five classes included: (1) live trees; (2) stumps
(measured at stump height [dsh: 0.3 m] and later
transformed to dbh using a regression equation devel-
oped from paired dbh and dsh measurements taken on a
random subset of trees); (3) knockdowns (stems killed
unintentionally by direct harvesting machinery impacts);
(4) snags (stems that died standing); and (5) windthrows,
including uprooted stems and stem snaps (stems that
broke while alive). Stem snaps were distinguished from
fallen snags based on wood structural characteristics: in
contrast to fallen snags, stem snaps generally splintered
at the point of breakage, maintained points of attach-
ment, and retained bark (Senecal et al. 2004). Some trees
that died standing and fell soon after may have been
mistakenly classified as stem snaps. This could have led
to overestimated rates of windthrow vs. standing death
risk, but would not have affected overall mortality rates.
Finally, we mapped the spatial location of each stem
by recording its distance and azimuth from the plot
center using an Impulse 200 Laser rangefinder with
Mapstar Electronic Compass Module II attachment
(Laser Technology Incorporated, Centennial, Colorado,
USA), and collected disk samples from all dead stems
whose outermost rings had not yet decayed. This
included trees that died before and after harvest; data
from trees that died before harvest were used to quantify
background mortality rates.
Disk samples were sanded with increasingly fine grit
until growth rings were clearly visible. We measured tree
ring widths along two radii of each disk using Win-
Dendro v. 2003b (Regent Instruments, Que ´ bec City,
Que ´ bec, Canada) and cross-dated each ring series using
a local master chronology created for another study
(Thorpe et al. 2007) in conjunction with the program
COFECHA (Holmes 1983). An individual’s year of
death was classified as the year following its final year of
diameter growth. We also classified each stump as either
on a skid trail or in a partially cut area by plotting stem
maps, delineating harvesting machinery travel corridors
(hereafter ‘‘skid trails’’), and manually classifying each
To ensure unbiased estimates of postharvest mortality
rates, we considered only the most recent five years
(2001–2005) of mortality dynamics in our analysis. We
were confident in our ability to cross-date all trees that
died in this window. Relative to the harvest date, the
span of this window differed among cutblocks in the
chronosequence; for example, fromþ4 toþ8 years since
harvest in 1997 cutblocks, and from?3 toþ1 years since
harvest in 2004 cutblocks.
To quantify the risk of tree mortality following partial
harvest, we estimated mortality risk before, during, and
after harvest using time since harvest and a set of
individual-level measurements (tree size, skid trail
proximity, exposure, and crowding) as predictor vari-
ables. The following section describes the set of models
we tested, the maximum likelihood methods employed
to estimate their parameters, and the methods used to
select the best model, calculate its parameter confidence
intervals, and assess its fit.
The basic model.—Prior to harvest, trees die at some
background rate b, the preharvest annual mortality risk,
estimated using residual tree and stump data from
partial-cut strips. In the year of harvest, individuals may
be harvested (stumps), or they may be killed uninten-
tionally by harvesting machinery (knockdowns). We
expected that the risk of being cut would increase
nonlinearly with tree size:
sti¼ lð1 ? e?wdbhiÞð1Þ
where stiis the risk of individual i of size dbhibeing cut,
and l and w are estimated parameters. The knockdown
risk, in contrast, is likely to decline exponentially with
where kdiis the risk of individual i of size dbhibeing
knocked down, and h and v are estimated parameters.
We expected that residual-tree mortality risk would peak
immediately after harvest and then decline exponentially
with time since harvest, down to some constant level:
Wiy¼ Ii;we?swtþ x
Siy¼ Ii;se?sstþ 1
where wiyand siyare the respective annual postharvest
risks of windthrow and standing mortality for individual
i in year y, Ii,wand Ii,sare the risks of windthrow and
standing death for individual i immediately after
harvest, and sw, ss, x, and 1 are estimated parameters.
The year y is a calendar year ranging from 2001 to 2005,
and time since harvest, t, is the difference between y and
the year of harvest. The parameters swand ssdetermine
the rate at which windthrow or standing death risk
declines with time since harvest, and x and 1 are the
constant postharvest rates of windthrow and standing
The total risk of mortality for an individual i in year y,
piy, depends on the time since harvest, t:
We used maximum likelihood methods to estimate the
parameters in Eqs. 1–5 using software written specifi-
cally for this study in the C programming language.
Mortality risks before, during, and after harvest were
estimated simultaneously using a simulated annealing
algorithm, based on the Metropolis algorithm (Metrop-
olis et al. 1953, Press et al. 1992), to search for parameter
values that would maximize the log-likelihood (L) of the
b þ stiþ kdi
if t ¼ 0
if t .0:
H. C. THORPE ET AL.1654
Vol. 18, No. 7
observed data set:
½Miylnð1 ? piyÞ þ ð1 ? MiyÞlnðpiyÞ?ð6Þ
where njis the total number of stems in plot j and Miyis
a dummy variable that indicates the status (live: 1; dead:
0) of individual i in year y.
Alternate models.—Risk of windthrow or standing
death following harvest may be influenced by a number
of predictor variables not considered in the basic model.
We tested for effects of tree size, skid trail proximity,
exposure, and crowding on mortality risk immediately
Ii;w¼ qwþ Di;wþ Ki;wþ Ei;w
Ii;s¼ qsþ Di;sþ Ki;sþ Ci;s
where qwand qsare estimated parameters, Di,wand Di,s
are the effects of tree diameter on windthrow and
standing death risk, Ki,wand Ki,sdescribe how skid trail
proximity influences windthrow and standing death risk,
Ei,wis the exposure effect on windthrow risk, and Ci,s
represents how crowding affects standing death risk.
The effects of diameter on windthrow and standing
death risk were estimated as follows:
where dw and ds are estimated (negative or positive)
Proximity to skid trails may increase both windthrow
and standing death risk if skid trails are associated with
where m is the number of knockdowns and stumps
located on skid trails and within k meters of tree i. The
parameters k, jw, and jsare estimated; k quantifies the
scale over which skid trails influence the mortality risk of
We expected that windthrow risk would increase with
declining postharvest neighborhood basal area, a proxy
for local exposure:
where gwis an estimated parameter, and BAi,postis the
postharvest neighborhood basal area of individual i.
BAi,postisthe sumofthecross-sectionalareaofeach ofthe
k neighboring trees, j, located within distance c of tree i:
where c is an estimated parameter that quantifies the
distance over which neighboring trees influence a target
tree’s mortality risk and dbh is in meters. For ease of
interpretation, BAi,postis expressed on a per hectare basis.
Potential neighbors considered in BAi,post calculations
included live and dead residual trees.
In contrast to our hypothesis for windthrow, we
expected standing death risk to increase with posthar-
vest neighborhood basal area, a proxy for local
where /sis an estimated constant.
Model selection.—We tested for the potential effects
of size, skid trail proximity, exposure, and crowding on
postharvest windthrow and standing death risk by fitting
the full model to the data (dw, ds, jw, js, gw, and /s6¼ 0)
and then fitting models that excluded the effects of one
or more variables (dw, ds, jw, js, gw, and/or /s¼ 0). We
used Akaike’s Information Criterion (AICc) to select the
AICc¼ ?2L þ 2K
n ? K ? 1
where L is the log-likelihood of the observed data set
and K is the number of estimated model parameters. The
model with the lowest AICc is considered most
parsimonious (Burnham and Anderson 2002).
Parameter confidence limits and model fit.—To obtain
parameter confidence limits, we randomly sampled
parameter values of the best model to obtain 100000
parameter sets. We calculated the log-likelihood of each
parameter set and its deviance, D, from the maximum
log-likelihood (D ¼ 2(L ? Lmax)). We excluded
parameter sets whose deviance exceeded the critical
value of the v2distribution (a ¼ 0.05, df ¼ 1), and from
the remaining sets of parameters, chose the maximum
and minimum values of each parameter as the 95%
confidence interval (Hilborn and Mangel 1997).
We calculated the predicted probability of mortality
for each individual ( pi), using the best model and its
associated maximum likelihood parameter estimates.
Individuals were grouped into eight classes based on
their predicted probability of mortality (0–0.025, 0.025–
0.05, 0.05–0.15, 0.15–0.25, 0.25–0.4, 0.4–0.6, 0.6–0.8,
0.8–1). Within each class, we calculated the mean
predicted probability of mortality and the observed
proportion of dead individuals. Pairs of observed and
predicted values were plotted to assess model fit.
Sample size and stand structure
We sampled 10361 stems across 54 plots in 18
cutblocks, including 4292 live residual trees, 4683
stumps, 648 knockdowns, and 738 dead trees from
which we collected a disk sample. The outermost rings of
many of the disks had decayed beyond our ability to
obtain an accurate year of death (n ¼ 108). Of the 630
October 20081655TREE MORTALITY AFTER PARTIAL HARVEST
stems we were able to cross-date, 269 died within the
2001–2005 window and were considered in our statistical
analysis. Of these dead trees, 139 had died standing, 56
were uprooted, and 74 were broken (total 130 wind-
thrown). Black spruce made up 92% of the sampled
stems. Balsam fir (Abies balsamea (L.) Mill.) comprised
an additional 7%, while paper birch (Betula papyrifera
Marsh.) and balsam poplar (Populus balsamifera L.)
made up ,1% of stems.
Harvesting reduced the density of stems .5 cm dbh
from 1511 6 87 stems/ha (cutblock average 6 SE) to 725
6 58 stems/ha, and the basal area from 16.35 6 0.62
m2/ha to 4.04 6 0.34 m2/ha (Fig. 2). Stumps comprised
the majority of the basal area, 11.83 6 0.63 m2/ha, and
knockdowns contributed an additional 0.48 6 0.045
m2/ha. Harvesting was concentrated in the larger size
classes: 97% of stems .15 cm dbh had been cut. Residual-
tree and knockdown abundances, in contrast, peaked in
the 5–10 cm size class (Fig. 2A). By the time of sampling,
live trees made up 3.47 6 0.31 m2/ha of the residual basal
area, while 0.34 6 0.04 m2/ha had been windthrown, and
0.23 6 0.05 m2/ha had died standing (Fig. 2D).
Observed mortality response to harvest
Mortality of residual trees peaked at 3.8% in the first
year after harvest, a rate 12.6 times higher than the
preharvest annual rate of 0.28% (Fig. 3). Mortality
declined with time, and was only 0.2% in the 10th year
after harvest (Fig. 3). A decade after harvest, an average
of 13.3% of residual trees had died; 6.3% had been
windthrown and 7.0% had died standing. Compared to
basal area by diameter class. The 54 inventoried 0.13-ha plots included 4922 residual trees (live and dead; 725 6 58 per hectare
[mean 6 SE]), 4683 stumps (690 6 44 per hectare), and 648 knockdowns (95 6 9.8 per hectare) .5 cm dbh. Stand basal area was
composed of 4.04 6 0.27 m2/ha of residual trees, 11.83 6 0.63 m2/ha of stumps, and 0.48 6 0.045 m2/ha of knockdowns.
Number of stems .5 cm dbh (A) before and (B) after harvest, and corresponding (C) preharvest and (D) postharvest
after partial harvest. Dashed vertical line indicates harvest
event. Direct harvest (i.e., stump and knockdown) mortality is
not included. Dotted line indicates estimated preharvest
background mortality rate (0.28% per year).
Mean observed annual mortality (6SE) before and
H. C. THORPE ET AL. 1656
Vol. 18, No. 7
the cumulative decadal mortality rate of 2.8% expected
in the absence of harvest, this represents a 475% increase
Model selection and goodness of fit
The most parsimonious mortality model (DAICc¼ 0)
included the effects of: tree size (Dw) on windthrow risk,
skid trail proximity (Kw, Ks) on windthrow and standing
death risk, exposure (Ew) on windthrow risk, and
crowding (Cs) on standing death risk (Table 1, Model
1). The selected model and associated parameters (Table
2) yielded a good fit to the observed data, with a 0.96
slope between predicted and observed mortality and
correct live/dead classification of 79% of all stems (Table
1, Fig. 4 and 5A, B).
Alternate model support.—The model that excluded all
size effects had some support compared to the best
model, with a DAICcvalue of 3.0 (Table 1, Model 3
[Burnham and Anderson 2002]). Only two other models
had DAICc values below 5: the full model (Table 1,
Model 2, DAICc¼ 2.0) and the model that included Ds
but not Dw(Table 1, Model 4, DAICc¼4.5). While these
models had some support, they did not provide any
additional predictive power to their simpler forms
(Model 1 (ds¼ 0) and Model 3 (dw¼ 0, ds¼ 0),
respectively). Models that excluded effects of skid trail
proximity, exposure, and/or crowding had DAICcvalues
ranging from 9.8 to 42, and had essentially no support.
Postharvest mortality.—Predicted risk of windthrow
peaked at 2.7% in the first year after harvest for residual
trees of average size, skid trail proximity, and exposure.
This risk declined with time, and had neared the
estimated postharvest risk constant (x: 0.12% per yr;
Table 2) 7–8 years after harvest. Windthrow risk
increased with residual-tree size, and ranged from 2.2%
to 3.3% in the first year after harvest for trees from 5 to
12.4 cm dbh, a size range representing 95% of residual
trees (Fig. 5A). For trees of average skid trail proximity
and crowding, the risk of standing death peaked at 1.3%
in the first year after harvest (Fig. 5B). This risk declined
more slowly than for windthrow and was not related to
Skid trail proximity (Kw, Ks) was the most important
predictor of postharvest mortality, both via windthrow
and standing death. Skid trail proximity values ranged
from 0 to 26 skid trail stumps and knockdowns (mean:
TABLE 1. AICccomparisons of alternate mortality models.
Effects included in
windthrow risk, Iw
Effects included in
standing death risk, Is
Dwþ Kwþ Ew
Dwþ Kwþ Ew
Dsþ Ksþ Cs
Dsþ Ksþ Cs
Notes: Dwand Dsare the size effects on windthrow (w) and standing death risk (s) (Eqs. 9, 10), Kwand Ksare skid trail proximity
effects (Eqs. 11, 12), Ewis the exposure effect (Eq. 13), and Csis the crowding effect (Eq. 15). DAICcis the difference between the
minimum AICc, associated with the best model, and the AICcof any alternate model (Burnham and Anderson 2002). Models that
did not include effects of skid trail proximity, exposure, and crowding had essentially no support (DAICcvalues . 9.8) and are not
shown. Concordance was calculated as the sum of the number of dead trees whose predicted probability of mortality was .0.5 and
the number of live trees whose predicted probability of mortality was ?0.5, divided by the total number of trees. Goodness of fit is
the slope of the line between predicted probability and observed proportion of mortality, binned into eight mortality classes (0–
0.025, 0.025–0.05, 0.05–0.15, 0.15–0.25, 0.25–0.4, 0.4–0.6, 0.6–0.8, 0.8–1).
1, Model 1).
Maximum likelihood parameter estimates (and 95% confidence limits) of the most parsimonious mortality model (Table
Parameter MLEParameterMLEAssociated terms
Elevated postharvest mortality
Skid trail proximity and neighborhood radii (m)
7.35 (7.07, 8.11)
Pre- and postharvest background mortality
0.0028 (0.0019, 0.0034)
0.0012 (0.0018, 0.008)
4.88 (4.64, 6.12)
0.0093 (0.0072, 0.0098)
0.0028 (0.0013, 0.0046)
0.0066 (0.0046, 0.0085)
0.0021 (0.0014, 0.027)
0.00012 (0.000, 0.0041)
0.629 (0.542, 0.773)
size effect (Dw)
skid trail effects (Kw, Ks)
exposure (Ew) and crowding (Cs) effects
0.0014 (0.002, 0.0008)
0.0005 (0.0004, 0.0006)
0.000 (0.000, 0.0002)
0.189 (0.156, 0.260)
1.50 (1.38, 1.80)
0.00004 (0.0000, 0.001)
0.287 (0.237, 0.425)
0.183 (0.175, 0.228)
harvest (stump) risk (st)
knockdown risk (kd)
October 20081657TREE MORTALITY AFTER PARTIAL HARVEST
7.0) located within 7.35 m of a residual tree (the skidding
proximity radius k; Table 2). For residual trees of
average size and postharvest neighborhood basal area,
the observed range of skid trail proximities was
associated with predicted windthrow rates of 0.2–9.3%
and standing death rates of 0.4–3.5% in the first year
after harvest (Fig. 5C, D). Although skid trail proximity
had a larger immediate influence on windthrow risk, the
relatively slow time course of standing death resulted in
similar cumulative impacts of skidding proximity on
windthrow and standing death risk over the first decade
Windthrow risk also increased with local exposure,
Ew, although this effect explained less of the observed
variation in windthrow than did skid trail proximity.
Postharvest neighborhood basal area (BApost), calculat-
ed within 1.5 m of each residual tree (the neighborhood
radius c; Table 2), was negatively related to exposure,
and ranged from 79 to 0 m2/ha (mean: 9.8 m2/ha). This
range was associated with predicted windthrow rates
from 0% to 3.8% in the first year after harvest for trees
of average size and skid trail proximity (Fig. 5E).
Standing death risk increased with local crowding, Cs:
trees in crowded neighborhoods experienced higher
postharvest standing mortality risk than those with
few or no neighbors. At average skid trail proximity and
peak neighborhood crowding (BApost¼79 m2/ha), 4.4%
of residual trees were expected to die standing in the first
year after harvest. More modest standing death rates of
0.8–2.4% in the first year after harvest were expected for
BApostvalues between 0 and 35 m2/ha, a range that
represented the neighborhoods of 95% of residual trees
Background mortality.—The estimated annual rate of
mortality prior to harvest was 0.28%, a rate considerably
higher than the sum of the estimated constant posthar-
vest windthrow (x ¼ 0.12%) and snag recruitment (1 ¼
0.004%) rates. This indicates that partial harvests reduce
background rates of mortality, at least over the first
decade after harvest.
Harvest intensity and mortality risk.—Cumulative
elevated mortality, the combined rates of windthrow
and standing death over the first decade after harvest,
depended primarily on skidding proximity (Fig. 6).
Across the observed range of skid trail proximities,
estimated cumulative mortality ranged from 2.4% to
37%, and represented mortality rates elevated up to 12
times above background (Fig. 6A). Predicted mortality
across the range of skid trail proximity was split roughly
equally between snags and windthrow. In contrast, total
cumulative mortality predictions varied only slightly
across local retention levels (BApost), from 11.2% to
11.8%, a pattern attributed to opposite trends in the
causes of standing death vs. windthrow. At high values
of BApost, standing death risk was highest, while
windthrow risk peaked at low values of BApost(Fig. 6B).
Tree mortality plays a fundamental role in forest
ecosystems, shaping their structure and composition,
and contributing to the availability of light, nutrients,
and habitat (Franklin et al. 1987). While models often
assume that mortality risk remains constant at a given
resource level (e.g., Tilman 2004), tree mortality is a
highly variable process, often occurring in pulsed events,
through disturbances such as windstorms and insect
outbreaks. Studies have generally assumed that moder-
ate-severity disturbances are discrete events, and have
examined only a ‘‘snapshot’’ of postdisturbance forest
structure and composition (e.g., Canham et al. 2001,
Rich et al. 2007). However, if exposed residual trees
experience increased risk of mortality, disturbances
could alter canopy-tree dynamics over longer time scales
than has generally been supposed. Little research has
examined this potential phenomenon. In one study,
expected increases in beetle-induced tree mortality
following a windstorm were not found (Kulakowski
and Veblen 2003), while another documented highly
elevated windthrow rates for residual trees that survived
spruce budworm outbreaks (Taylor and MacLean
2007). Further research is needed to assess the generality
and magnitude of postdisturbance tree mortality in
Like residual trees in naturally disturbed forests, trees
in partially cut stands are expected to be at increased
risk of mortality after harvest. Our results support this
hypothesis, as we documented a cumulative elevated tree
mortality rate of 10.5% in the first decade after harvest,
with windthrow and standing death contributing rough-
ly equally. Windthrow rates peaked immediately fol-
lowing harvest and declined rapidly thereafter, while the
(Table 1, Model 1). The observed proportion of dead trees is
plotted vs. mean values of predicted probability of mortality for
each of the following eight predicted probability classes (range
with number of trees in square brackets: 0–0.025 , 0.025–
0.05 , 0.05–0.15 , 0.15–0.25 , 0.25–0.4 , 0.4–
0.6 , 0.6–0.8 , and 0.8–1 ). The line represents a
1:1 relationship between predicted probability and observed
proportion of mortality.
Goodness of fit of the maximum likelihood model
H. C. THORPE ET AL.1658
Vol. 18, No. 7
rate of standing death displayed a more moderate peak
followed by a slower postharvest decline. A recent study
showed high rates of windthrow soon after harvest (0–5
years) followed by a period (5–18 years postharvest)
during which standing death was the most important
mode of mortality (Jo ¨ nsson et al. 2007). Although much
of the previous research on postharvest mortality
considers windthrow only (e.g., Huggard et al. 1999,
Ruel et al. 2001, Scott and Mitchell 2005), our results
and those of Jo ¨ nsson et al. (2007) indicate that
neglecting standing mortality will lead to substantially
underestimated total losses of residual trees.
With few exceptions, prior studies of postharvest
mortality have given snapshots rather than integrated
mortality estimates, and have not rigorously distin-
guished postharvest effects from background mortality
rates. Nevertheless, the mortality rate documented here
is intermediate among those previously reported. Fol-
lowing selection harvests, low rates of mortality (;1–
5%) were found in temperate (Wiser et al. 2005,
Caspersen 2006) and tropical (Sist and Nguyen-The
predicted windthrow risk at 5 and 12.4 cm dbh, (B) mean observed percentage of stems dead standing (6SE) and predicted standing
death risk; impact of skid trail proximity on risk of (C) windthrow and (D) standing death, where m is the number of skid trail
stumps and knockdowns within 7.35 m of residual trees; and effect of postharvest basal area (BApost, within 1.5 m) on risk of (E)
windthrow and (F) standing death. Skid trail proximity is set at its mean value (m¼7.0) for panels A, B, E, and F; BApostis set at its
mean value (9.8 m2/ha) for panels A–D; and dbh is set at its mean value (7.9 cm) for panels C and E.
Predicted risk of mortality following partial harvest. (A) Mean observed percentage of stems windthrown (6SE) and
October 20081659 TREE MORTALITY AFTER PARTIAL HARVEST
2002) forests. In studies from western North America,
similarly low mortality rates have been reported in
.40% retention treatments (Coates 1997, Beese and
Bryant 1999, Huggard et al. 1999, Maguire et al. 2006),
while high rates of windthrow (25–50%) have been
found in treatments that retain ,15% (Beese and Bryant
1999, Scott and Mitchell 2005). In the boreal forest, 75%
retention harvests did not lead to elevated mortality in
one study (Ruel et al. 2003), but windthrow rates
reached 15–100% in retention patches in others (Hautala
and Vanha-Majamaa 2007, Jo ¨ nsson et al. 2007). A
recent study of 10% retention cuts in boreal mixedwoods
documented cumulative five-year mortality elevated 7.3–
35% above background levels, depending on species
(aspen . poplar . birch . white spruce [Bladon et al.
2008]). Previous work on black spruce has reported
losses similar to the present study (;9–20% [Ruel 1989,
Ruel et al. 2001]).
The harvest treatment we considered removed nearly
all trees .12 cm diameter. The small range of tree sizes
retained may have influenced the rate of postharvest
mortality observed. However, no general pattern be-
tween mortality risk and tree size has been established,
and it is not clear whether we would have found higher
or lower mortality rates had the treatment retained more
large trees. A previous study found reduced postharvest
mortality risk in both small and large trees (Huggard et
al. 1999), while another documented decreasing post-
harvest mortality with increased tree size (Caspersen
2006). Studies in natural forests generally report
increasing windthrow risk with tree size (e.g., Canham
et al. 2001, Rich et al. 2007).
We estimated a preharvest, or background, annual
mortality rate of 0.28%. This is on the low end of rates
reported in the literature, which vary dramatically, both
interannually and among forest types (e.g., Hennon and
McClellan 2003, Senecal et al. 2004, Maguire et al. 2006,
Jo ¨ nsson et al. 2007). We are unaware of any published
data on tree mortality rates for uneven-aged lowland
black spruce stands, and thus cannot determine whether
the low rate we observed is typical for these sites.
However, had background mortality been 0.5% per year
(a commonly reported rate), the rate of elevated
mortality would still have been substantial (8.3% vs.
10.5% on a cumulative basis).
After harvest, estimated background mortality was
only 0.12%, suggesting that once the postharvest
mortality pulse passes, residual trees experience a period
of relatively low background, or suppression-related,
mortality. The only previous study to consider this also
found lower background mortality after harvest (Cas-
persen 2006). If such reduced rates persist, they could
partially offset mortality losses from the years immedi-
ately following harvest. Long-term data and very
intensive sampling would be required to clarify this
Predictors of postharvest mortality
Skid trail proximity was the most important determi-
nant of windthrow and standing mortality risk. The
most parsimonious predictor of skid trail effects was the
number of skid trail stumps and knockdowns located
within a radius of 7.35 m (k) of an individual residual
tree, a distance roughly equivalent to the skid trail
width. We were surprised at both the predominance of
skid trail effects and their large spatial dimension, given
that operations are aimed at reducing logging impacts
by harvesting during winter. Proximity to skid trails is
likely associated with harvest-related damage such as
bark abrasion, crown damage, and/or root compaction,
all of which could lead to increased risk of mortality. We
decade after partial harvest across a range of (A) skid trail
proximities and (B) postharvest neighborhood basal areas,
representing effects of crowding (high BApost) and exposure
(low BApost). Total elevated mortality is the sum of predicted
windthrow and standing death. Arrows indicate the mean
observed harvest intensity. Estimates were obtained using the
best model (Table 1, Model 1) and its associated parameters
(Table 2) to calculate mortality from t ¼ 1 to t ¼ 10 (omitting
constant postharvest mortality x and 1). In (A), BApostis set at
9.8 m2/ha (mean observed BApost); in (B), skid trail proximity m
is set at 7.0 (mean observed m); and dbh is set at 7.9 cm (mean
observed dbh) for A and B. Note that there was ample
orthogonal variation between skid trail proximity and posthar-
vest basal area; pairs of m and BApostvalues considered are
realistic (see Discussion: Predictors of postharvest mortality).
Predicted cumulative elevated mortality in the first
H. C. THORPE ET AL.1660
Vol. 18, No. 7
suspect that despite winter harvesting, compaction is a
principal cause of postharvest mortality in this system,
although its importance is likely to vary with weather
conditions such as snow and frost depth.
Windthrow risk increased with declining postharvest
neighborhood basal area, a finding we expected since
residual trees become exposed as nearby basal area is
removed. Previous studies have documented increased
windthrow with harvesting intensity at the treatment
level (e.g., Beese and Bryant 1999, Maguire et al. 2006,
Jo ¨ nsson et al. 2007). Our preliminary analyses indicated
no evidence for a nonlinear trend in windthrow risk with
basal area change, and we thus cannot identify a
threshold rate of retention below which windthrow
mortality remains negligible. The harvest method
considered in this study, however, involves low rates of
retention relative to other systems. Although many
residual trees experienced no change in local basal area
within their neighborhood radius of 1.5 m, it is unlikely
that any residual tree avoided all harvest-related stress.
Postharvest neighborhood basal area also influenced
standing death risk, with crowded residual trees (those
with high BApostvalues) experiencing higher risk. We
conducted preliminary analyses that allowed preharvest
basal area to influence standing death risk, but BApost
was a much better predictor. Since trees that are
crowded after harvest would necessarily have been
crowded before, this result suggests that previously
suppressed trees, released at harvest, can respond to
improved growing conditions and avoid mortality. A
similar result was found in northern hardwood forests
(Caspersen 2006). The effect of crowding on postharvest
standing death risk also indicates that suppressed trees
experience greater mortality risk after harvest than prior
to it. One possible mechanism for this is wind-sway-
induced cavitation of xylem vessels and subsequent
effects on water transport (Liu et al. 2003, Bladon et al.
2007). This effect is expected to be more pronounced
among suppressed individuals in dense stands, since they
typically show high height-to-diameter ratios and low
biomechanical stability (e.g., Weiner and Thomas 1992).
The estimated radius of influence for neighborhood
effects (c) underlines the small scale at which neighbor-
hood interactions operate in these black spruce stands.
This neighborhood radius of 1.5 m explains the
discrepancy between neighborhood- and stand-level
basal areas reported. At the stand level, harvesting
reduced stand basal area by 75%, from ;16 to 4 m2/ha
(Fig. 2), while the average postharvest neighborhood
basal area (BApost) was 9.8 m2/ha. Clear-cut skid trails
occupy a large proportion of the harvest area (40–50%)
and represent the largest basal area reductions. These
areas are largely excluded from BApostcalculations since
only a small proportion of the clear-cut trails is located
within 1.5 m of a residual tree.
Contrary to our expectations, postharvest neighbor-
hood basal area and skid trail proximity were not
strongly correlated (r ¼ 0.10; P , 0.0001). While these
two predictor variables cannot be entirely independent,
the low correlation coefficient between them indicates
ample orthogonal variation. Residual trees with high
skid trail proximities (m ? 15) were observed across the
entire range of postharvest neighborhood basal areas, 0–
79 m2/ha, and residual trees that lost all neighbors at
harvest (i.e., every tree within 1.5 m was cut) were
associated with skid trail proximity values ranging from
0 to 24. Thus, a number of residual trees experienced
logging damage without concomitant increases in local
exposure, and many trees that were exposed by harvest
did not suffer from skid trail impacts.
We documented increasing windthrow risk with size
of residual trees, a finding that supports prior research
on windthrow generally (e.g., Canham et al. 2001) and in
black spruce, a species whose probability of windthrow
increases particularly steeply with dbh (Smith et al.
1987, Rich et al. 2007). We found only a weak pattern
between tree size and windthrow risk; the model that
included size effects was only slightly better than the
model without (Table 1, Model 1 vs. 3). This is likely a
result of the small size range of residual trees retained in
the harvest treatment. Contrary to our expectations, we
did not find reduced risk of standing death with
increasing tree size. Again, we might have detected this
effect had a larger size range of trees been retained.
Implications for forest ecosystem management
The retention of live residual trees at harvest is aimed
at better emulating natural forest disturbance: individual
trees or forest patches commonly persist following
wildfire (Franklin et al. 2002), and large proportions
of canopy trees survive moderate-severity disturbances
such as windstorms (Hanson and Lorimer 2007). Like
residual trees in naturally disturbed forests, retained live
trees in recently harvested sites are expected to provide
habitat, contribute to hydrological and nutrient cycling,
and reduce soil erosion (Franklin et al. 1997). Clearly,
high rates of postharvest mortality will lead to reduced
capacity of harvested sites to provide these ecological
functions. Mortality rates exceeding 10% of residual
trees have been considered an ‘‘operational failure’’ in
some regions (Coates 1997). In the first decade following
partial harvests at our study site, an average of 13.3% of
residual trees died, exceeding this suggested threshold.
Thus both ecological and operations objectives of
structural retention harvesting may be seriously com-
promised in this system unless postharvest mortality is
Our analysis examined the influence of two spatial
factors on postharvest mortality: neighborhood basal
area and skid trail proximity. Neighborhood basal area,
analogous to the rate of residual-tree retention on a local
(1.5 m) scale, was not an important predictor of total
postharvest mortality, since windthrow and standing
death risk compensated for each other, peaking at
exposed and crowded conditions, respectively (Fig. 6B).
This result suggests that retention rates per se may not
October 20081661 TREE MORTALITY AFTER PARTIAL HARVEST
be as important a driver of postharvest mortality as has
been assumed, at least over small spatial scales. In
contrast, skid trail proximity had a very strong influence
on predicted total mortality. Where skidding impacts
were low, postharvest mortality was negligible; at peak
skidding proximity, predicted total mortality exceeded
35% of residual trees (Fig. 6A). Our results indicate that,
at the mean observed postharvest basal area, decreasing
skid trail coverage by half (from the mean m¼7.0 to m¼
3.5) would reduce cumulative mortality in the first
decade after harvest by 40%. Skid trail coverage could
be reduced by widening partial-cut strips and using
smaller harvesting machinery (see Groot 2002). Further
research is required to determine the extent to which
skidding proximity influences postharvest mortality in
other forest ecosystems.
Our finding of large mortality responses to harvest
that depend both on time and proximity to the
disturbance represents an important step toward devel-
oping accurate predictions of stand dynamics in
structural retention silviculture. Residual trees also
show large growth responses to harvest (e.g., Latham
and Tappeiner 2002, Bebber et al. 2004, Thorpe et al.
2007), and thus spatially explicit, individual-based
simulation models that incorporate dynamic growth
and mortality responses of residual trees are required to
explore the implications of partial harvest scenarios.
More generally, these models are needed to evaluate the
potential of novel silvicultural approaches to address
biodiversity and timber production concerns over the
long term. Such models will also be useful for
understanding natural forest dynamics, since mortality
patterns analogous to those following structural reten-
tion harvests have been documented after intermediate-
scale disturbances such as insect outbreaks in natural
forests (Taylor and McLean 2007). However, skid trail
impacts were the most important predictor of posthar-
vest mortality in this study. Skid trails have no natural
analog, and therefore mortality rates following struc-
tural retention harvests may be elevated well above
those found after natural disturbance. Thus, while
partial harvesting has been motivated by attempts to
emulate natural disturbances, our results suggest that
stand dynamic patterns and concomitant ecological
processes may differ markedly in harvested compared
to naturally disturbed stands.
We are grateful to Beatriz Lucas, Kate Abbott, Marc Faiella,
and Adam Martin for their assistance in the field, and to Elise
Benczkowski for her dendrochronology work. We thank Bob
Bolton, Jennifer Tallman, and Rod Gemmel of AbitibiBowater
for supplying us with maps, harvesting records, and logistical
support, and two anonymous reviewers for their thoughtful
comments and criticisms. Funding was provided by the
Sustainable Forest Management Network, the Natural Sciences
and Engineering Research Council of Canada, and the Lake
Abitibi Model Forest. This study was conducted in the
Bebber, D. P., S. C. Thomas, W. G. Cole, and D. Balsillie.
2004. Diameter increment in mature eastern white pine Pinus
strobus L. following partial harvest of old-growth stands in
Ontario, Canada. Trees 18:29–34.
Beese, W. J., and A. A. Bryant. 1999. Effect of alternative
silvicultural systems on vegetation and bird communities in
coastal montane forests of British Columbia, Canada. Forest
Ecology and Management 115:231–242.
Bladon, K. D., V. J. Lieffers, U. Silins, S. M. Lanha ¨ usser, and
P. V. Blenis. 2008. Elevated mortality of residual trees
following structural retention harvesting in boreal mixed-
woods. Forestry Chronicle 84:70–75.
Bladon, K. D., U. Silins, S. M. Landha ¨ usser, C. Messier, and
V. J. Lieffers. 2007. Carbon isotope discrimination and water
stress in trembling aspen following variable retention
harvesting. Tree Physiology 27:1065–1071.
Burnham, K. P., and D. R. Anderson. 2002. Model selection
and multimodel inference: a practical information-theoretic
approach. Second edition. Springer-Verlag, New York, New
Canham, C. D., P. T. LePage, and K. D. Coates. 2004. A
neighborhood analysis of canopy tree competition: effects of
shading versus crowding. Canadian Journal of Forest
Canham, C. D., M. J. Papaik, and E. F. Latty. 2001.
Interspecific variation in susceptibility to windthrow as a
function of tree size and storm severity for northern
temperate tree species. Canadian Journal of Forest Research
Canham, C. D., M. J. Papaik, M. Uriarte, W. H. McWilliams,
J. C. Jenkins, and M. J. Twery. 2006. Neighborhood analyses
of canopy tree competition along environmental gradients in
New England forests. Ecological Applications 16:540–554.
Canham, C. D., and M. Uriarte. 2006. Analysis of neighbor-
hood dynamics of forest ecosystems using likelihood methods
and modeling. Ecological Applications 16:62–73.
Caspersen, J. P. 2006. Elevated mortality of residual trees
following single-tree felling in northern hardwood forests.
Canadian Journal of Forest Research 36:1255–1265.
Coates, K. D. 1997. Windthrow damage 2 years after partial
cutting at the Date Creek silvicultural systems study in the
Interior Cedar-Hemlock forests of northwestern British
Columbia. Canadian Journal of Forest Research 27:1695–
Deans, A. M., J. R. Malcolm, S. M. Smith, and M. I. Bellocq.
2005. Edge effects and the responses of aerial insect
assemblages to structural-retention harvesting in Canadian
boreal peatland forests. Forest Ecology and Management
Dovc ˇ iak, M., C. B. Halpern, J. R. Saracco, S. A. Evans, and
D. A. Liguori. 2006. Persistence of ground-layer bryophytes
in a structural retention experiment: initial effects of level and
pattern of overstory retention. Canadian Journal of Forest
Environment Canada. 2002. Canadian climate normals 1971–
Franklin, J. F., D. R. Berg, D. A. Thornburg, and J. C.
Tappeiner. 1997. Alternative silvicultural approaches to
timber harvesting: variable retention harvest systems. Pages
111–139 in K. A. Kohm and J. F. Franklin, editors. Creating
a forestry for the 21st century: the science of ecosystem
management. Island Press, Washington, D.C., USA.
Franklin, J. F., H. H. Shugart, and M. E. Harmon. 1987. Tree
death as an ecological process. BioScience 37:550–556.
Franklin, J. F., T. A. Spies, R. Van Pelt, A. B. Carey, D. A.
Thornburgh, D. R. Berg, D. B. Lindenmayer, M. E.
Harmon, W. S. Keeton, D. C. Shaw, K. Bible, and J. Q.
Chen. 2002. Disturbances and structural development of
H. C. THORPE ET AL.1662
Vol. 18, No. 7
natural forest ecosystems with silvicultural implications, Download full-text
using Douglas-fir forests as an example. Forest Ecology
and Management 155:399–423.
Groot, A. 2002. Is uneven-aged silviculture applicable to
peatland black spruce (Picea mariana) in Ontario, Canada?
Hanson, J. J., and C. G. Lorimer. 2007. Forest structure and
light regimes following moderate wind storms: implications
for multi-cohort management. Ecological Applications 17:
Harvey, B. D., A. Leduc, S. Gauthier, and Y. Bergeron. 2002.
Stand-landscape integration in natural disturbance-based
management of the southern boreal forest. Forest Ecology
and Management 155:369–385.
Hautala, H., and I. Vanha-Majamaa. 2007. Immediate tree
uprooting after retention-felling in a coniferous boreal forest
in Fennoscandia. Canadian Journal of Forest Research 36:
Heithecker, T. D., and C. B. Halpern. 2006. Variation in
microclimate associated with dispersed-retention harvests in
coniferous forests of western Washington. Forest Ecology
and Management 226:60–71.
Hennon, P. E., and M. H. McClellan. 2003. Tree mortality and
forest structure in the temperate rain forests of southeast
Alaska. Canadian Journal of Forest Research 33:1621–1634.
Hilborn, R., and M. Mangel. 1997. The ecological detective:
confronting models with data. Princeton University Press,
Princeton, New Jersey, USA.
Holmes, R. L. 1983. Computer-assisted quality control in tree-
ring dating and measurement. Tree-ring Bulletin 43:69–78.
Huggard, D. J., W. Klenner, and A. Vyse. 1999. Windthrow
following four harvest treatments in an Englemann spruce–
subalpine fir forest in southern interior British Columbia,
Canada. Canadian Journal of Forest Research 29:1547–1556.
Jo ¨ nsson, M. T., S. Fraver, B. G. Jonsson, M. Dynesius, M.
Rydga ˚ rd, and P.-A. Esseen. 2007. Eighteen years of tree
mortality and structural change in an experimentally
fragmented Norway spruce forest. Forest Ecology and
Keeton, W. S. 2006. Managing for late-successional/old-growth
characteristics in northern hardwood-conifer forests. Forest
Ecology and Management 235:129–142.
Kulakowski, D., and T. T. Veblen. 2003. Subalpine forest
development following a blowdown in the Mount Zirkel
Wilderness, Colorado. Journal of Vegetation Science 14:653–
Lance, A. N., and M. Phinney. 2001. Bird responses to partial
retention timber harvesting in central interior British
Columbia. Forest Ecology and Management 142:267–280.
Latham, P., and J. Tappeiner. 2002. Response of old-growth
conifers to reduction in stand density in western Oregon
forests. Tree Physiology 22:137–146.
Lin, Y., M. L. Hulting, and C. K. Augspurger. 2004. Causes of
spatial patterns of dead trees in forest fragments in Illinois.
Plant Ecology 170:15–27.
Lindenmayer, D., and M. A. McCarthy. 2002. Congruence
between natural and human forest disturbance: a case study
from Australian montane ash forests. Forest Ecology and
Liu, X., U. Silins, V. J. Lieffers, and R. Man. 2003. Stem
hydraulic properties and growth in lodgepole pine stands
following thinning and sway treatment. Canadian Journal of
Forest Research 33:1295–1303.
Maguire, D. A., D. B. Mainwaring, and C. B. Halpern. 2006.
Stand dynamics after variable-retention harvesting in mature
Douglas-fir forests of Western North America. Allgemeine
Forst und Jagdzeitung 177:120–129.
McCarthy, J. W., and G. Weetman. 2007. Stand structure and
development of an insect-mediated boreal forest landscape.
Forest Ecology and Management 241:101–114.
Metropolis, N., A. W. Rosenbluth, M. N. Rosenbluth, A. H.
Teller, and E. Teller. 1953. Equation of state calculations by
fast computing machines. Journal of Chemical Physics 21:
OMNR. 2001. Forest management guide for natural distur-
bance pattern emulation, Version 3.1. Ontario Ministry of
Natural Resources, Queen’s Printer for Ontario, Toronto,
Papaik, M. J., and C. D. Canham. 2006. Multi-model analysis
of tree competition along environmental gradients in
southern New England forests. Ecological Applications 16:
Press, W. H., S. A. Teukolsky, W. T. Vetterling, and B. P.
Flannery. 1992. Numerical recipes in C: the art of scientific
computing. Cambridge University Press, Cambridge, UK.
Rich, R. L., L. E. Frelich, and P. B. Reich. 2007. Wind-throw
mortality in the southern boreal forest: effects of species,
diameter and stand age. Journal of Ecology 95:1261–1273.
Ruel, J.-C. 1989. Mortality in black spruce leave strips
following strip cutting in 3 regions of Quebec. Forestry
Ruel, J.-C., D. Pin, and K. Cooper. 2001. Windthrow in
riparian buffer strips: effect of wind exposure, thinning and
strip width. Forest Ecology and Management 143:105–113.
Ruel, J.-C., P. Raymond, and M. Pineau. 2003. Windthrow
after shelterwood cutting in balsam fir stands. Northern
Journal of Applied Forestry 20:5–13.
Scott, R. E., and S. J. Mitchell. 2005. Empirical modelling of
windthrow risk in partially harvested stands using tree,
neighbourhood, and stand attributes. Forest Ecology and
Senecal, D., D. Kneeshaw, and C. Messier. 2004. Temporal,
spatial, and structural patterns of adult trembling aspen and
white spruce mortality in Quebec’s boreal forest. Canadian
Journal of Forest Research 34:396–404.
Seymour, R. S., A. S. White, and P. G. deMaynadier. 2002.
Natural disturbance regimes in northeastern North America:
evaluating silvicultural systems using natural scales and
frequencies. Forest Ecology and Management 155:357–367.
Sist, P., and N. Nguyen-The. 2002. Logging damage and the
subsequent dynamics of a dipterocarp forest in East
Kalimantan (1990–1996). Forest Ecology and Management
Smith, V. G., M. Watts, and D. F. James. 1987. Mechanical
stability of black spruce in the Clay Belt region of northern
Ontario. Canadian Journal of Forest Research 17:1080–1091.
Taylor, S. L., and D. A. MacLean. 2007. Spatiotemporal
patterns of mortality in declining balsam fir and spruce
stands. Forest Ecology and Management 253:188–201.
Thorpe, H. C., and S. C. Thomas. 2007. Partial harvesting in
the Canadian boreal: success will depend on stand dynamic
responses. Forestry Chronicle 83:319–325.
Thorpe, H. C., S. C. Thomas, and J. P. Caspersen. 2007.
Residual-tree growth responses to partial stand harvest in the
black spruce (Picea mariana) boreal forest. Canadian Journal
of Forest Research 37:1563–1571.
Tilman, D. 2004. Niche tradeoffs, neutrality, and community
structure: a stochastic theory of resource competition,
invasion, and community assembly. Proceedings of the
National Academy of Sciences (USA) 101:10854–10861.
Weiner, J. 1984. Neighborhood interference amongst Pinus
rigida individuals. Journal of Ecology 72:183–195.
Weiner, J., and S. C. Thomas. 1992. Competition and allometry
in three species of annual plants. Ecology 73:648–656.
Wiser, S. K., R. B. Allen, U. Benecke, G. Baker, and D. Peltzer.
2005. Tree growth and mortality after small-group harvesting
in New Zealand old-growth Nothofagus forests. Canadian
Journal of Forest Research 35:2323–2331.
Young, T. P., and S. P. Hubbell. 1991. Crown asymmetry,
treefalls, and repeat disturbance of broad-leaved forest gaps.
October 20081663 TREE MORTALITY AFTER PARTIAL HARVEST