Hemlock Woolly Adelgid (Adelges tsugae) and Hemlock
(Tsuga spp.) in Western North Carolina: What do the Forest
Inventory and Analysis Data Tell Us?
James T. Vogt, Francis A. Roesch, and Mark J. Brown
Southeastern Naturalist, Volume 15, Issue 4 (2016): 631–645
Full-text pdf (Accessible only to subscribers.To subscribe click here.)
Southeastern Naturalist
631
J.T. Vogt, .A. Roesch, and M.J. Brown
22001166 SOUTHEASTERN NATURALIST 1V5o(4l.) :1653,1 N–6o4. 54
Hemlock Woolly Adelgid (Adelges tsugae) and Hemlock
(Tsuga spp.) in Western North Carolina: What do the Forest
Inventory and Analysis Data Tell Us?
James T. Vogt1,*, Francis A. Roesch2, and Mark J. Brown1
Abstract - Tsuga canadensis (Eastern Hemlock) and T. caroliniana (Carolina Hemlock)
are important components of western North Carolina forests. The invasive Adelges
tsugae (Hemlock Woolly Adelgid [HWA]) was first reported in NC in 1995, and by
2007 the entire range of hemlock in the state was infested. An examination of the Forest
Inventory and Analysis (FIA) program data for FIA Unit 4 (21 mountainous counties
in western North Carolina), looking at remeasured trees for the time period 1999–2013,
demonstrated that diameter net growth of hemlock decreased and mortality increased
with increasing duration of HWA infestation. Hemlock trees in this study had a ~50%
chance of survival after 12 years of confirmed HWA infestation in the county where they
occur, and growth of surviving trees was reduced by ~50% over the same time period.
This study demonstrates the utility of FIA data for examining effects of an introduced,
invasive pest on tree growth and mortality over a relatively small area. Some advantages
and limitations to our approach are discussed.
Introduction
Hemlock forests in eastern North America are highly valued for their ecological
and aesthetic characteristics. Tsuga canadensis (L.) Carrière (Eastern Hemlock) is
considered to be a foundation species (Ellison et al. 2005, Vose et al. 2013), providing
a suite of unique ecological functions related to microclimate, nutrient cycling,
soil ecology, stream ecology, and wildlife (Abella 2014). Also important but less
widely distributed, T. caroliniana Engelm. (Carolina Hemlock) generally occupies
sites along rock outcroppings on mountain bluffs and ridges with dry, nutrient-poor
soils (Jetton et al. 2008) and occasionally is found in cool, moist valleys and ravines
(Rentch et al. 2000). Carolina Hemlock occupies a relatively small geographic area
in southwest Virginia, western North Carolina, extreme northeast Georgia, northwest
South Carolina, and eastern Tennessee. Adelges tsugae (Annand) (Hemlock
Woolly Adelgid [HWA]) feeds on both species and has been in the eastern United
States since the 1950s, when it was detected in Virginia. It has since spread north to
Maine and Vermont and south to Georgia with a rate of spread that varies according
to environmental variables and was determined to average approximately 12–15
km y-1 in one study (Evans and Gregoire 2007). Feeding activity of HWA results in
1United States Department of Agriculture, Forest Service, Southern Research Station, Forest
Inventory and Analysis, 4700 Old Kingston Pike, Knoxville, TN 37919. 2United States
Department of Agriculture, Forest Service, Southern Research Station, Forest Inventory
and Analysis, 200 W.T. Weaver Boulevard, Asheville, NC 28804. *Corresponding author -
jtvogt@fs.fed.us.
Manuscript Editor: Robert Jetton
Southeastern Naturalist
J.T. Vogt, .A. Roesch, and M.J. Brown
2016 Vol. 15, No. 4
632
needle loss, tree decline, and mortality in as few as 4 years (McClure 1991). For
a review of HWA establishment, biology, and control, see Havill et al. (2014) and
references therein.
Hemlock stands in the southern Appalachians are especially vulnerable to decline
and mortality from HWA infestation (Lovett et al. 2006, Nuckolls et al. 2009).
In the mountainous counties of western North Carolina, hemlocks are a familiar
sight in cool coves, on north-facing slopes, and rock outcrops. A relative newcomer
to North Carolina, HWA was first collected in the mid-1990s. By 2007, widespread
hemlock mortality was noted in infested counties, and the most recent published
timber inventory for North Carolina reveals an overall downward trend in live hemlock
volume and an increase in dead hemlock volume in North Carolina from 2007
to 2013 (Brown and Vogt 2015).
The US Department of Agriculture, Forest Service, Forest Inventory and
Analysis (FIA) program has been collecting data on the extent and condition of
forested land since its inception in 1929 (Smith 2002). For several decades, FIA
conducted periodic inventories of states on a rotating basis, with up to 18 years
between consecutive surveys (Gillespie 1999). States began to conduct annual
inventories in 1999, sampling 10 to 20 percent of the survey plots within a state
each year (O’Connell et al. 2015). Under the new annual inventory, statewide
data are available in cycles of 5, 7, or 10 years, depending on the proportion
of plots sampled annually in a particular state. The systematic spatio-temporal
design of 1 plot for every 2403 ha of land (Reams et al. 2005), over the cycle, allows
users to define a spatio-temporal population of interest and obtain a sample
of plot observations for analysis (Roesch 2007, Smith 2002). Re-measured plot
designs have long been used to track the characteristics of individual trees over
time (Martin 1982).
A number of studies have attempted to quantify HWA’s influence on hemlock
populations at various scales. Trotter et al. (2013) examined hemlock across 21
eastern US states, using FIA data, and concluded that both live and dead basal
area of hemlock increased over the 20-year period prior to 2007. While dramatic
reductions in hemlock abundance due to HWA were not evident at this broad scale,
the authors noted that stands do appear to be accumulating dead hemlock and that
hemlock density may be starting to decrease in longer-infested states (e.g., Connecticut).
Morin and Liebhold (2015) took a regional look at both hemlock and
Fagus grandifolia Ehrh. (American Beech), a species suffering from widespread
decline due to beech bark disease, across the 22 eastern states where they overlap.
They found that annual net growth of hemlock decreased with increasing duration
of HWA infestation, while annual mortality of hemlock increased. At the broad
scale of their study, the decline in net growth of hemlock was not apparent until the
duration of HWA infestation surpassed 15 years. Both hemlock and beech exhibited
compensatory growth as a result of declines in the other species. Several authors
have evaluated changes in hemlock populations in response to HWA at a smaller
scale such as stand-level, where mortality may vary widely but can exceed 80–95%
(Orwig and Foster 1998, Small et al. 2005).
Southeastern Naturalist
633
J.T. Vogt, .A. Roesch, and M.J. Brown
2016 Vol. 15, No. 4
Some studies have shed light on how site characteristics may influence tree
health and mortality due to HWA infestation. For example, Orwig et al. (2002)
determined that hemlock stands on more xeric slopes in New England declined
more rapidly than other stands, but that the intensity of decline and mortality are
ultimately determined by duration of infestation. Some studies have found little
or no relationship between site characteristics (e.g., slope, aspect, moisture) and
HWA-related decline and mortality (e.g., Rentch et al. 2009), whereas others
have demonstrated weak associations between HWA impacts and various landscape
or site factors (Royle and Lathrop 2000, Young and Morton 2002). Kantola
et al. (2014, 2016) observed the highest density of dead hemlocks in riparian areas,
on steep hillsides, and at higher elevations. Several authors have noted that
less-dominant, suppressed trees succumb to HWA more rapidly than trees with
dominant crown positions (Eschtruth et al. 2006, Onken 1995, Orwig and Foster
1998, Orwig et al. 2002).
The current study was undertaken to characterize trends in the hemlock resource
in western North Carolina, and to assess the ability to discern effects of HWA on
hemlock diameter growth and mortality using FIA data—including site variables—
at a relatively limited, multi-county scale.
Field-site Description
Our field site consisted of the 21 mountainous counties of western North
Carolina that comprise FIA unit 4 in the state (Fig. 1). An FIA unit is intended
to comprise a large enough collection of counties for robust estimation of measures
of interest such as basal area, trees per hectare, etc. The boundaries to
the north, south, and west are the boundaries of the adjoining states, while the
eastern boundary corresponds roughly to the transition from mountains (central
Appalachian broadleaf forest, coniferous forest, and meadow) to piedmont. The
vast majority of hemlocks in North Carolina grow within the study area, where
HWA was first confirmed present in 2000. Two national forests comprise a large
portion of Unit 4: the Pisgah in the central counties and the Nantahala in the
southern counties. Approximately half of the Great Smoky Mountains National
Park resides in the study area as well.
Figure 1. Forest Inventory and Analysis Units in North Carolina.
Southeastern Naturalist
J.T. Vogt, .A. Roesch, and M.J. Brown
2016 Vol. 15, No. 4
634
Methods
We used 2 data sources. Initial year of confirmed HWA presence in each county
(HWAc) was obtained from North Carolina Forest Service (NCFS), Forest Health
Branch. Data were a result of follow-up visits by NCFS personnel and “windshield
surveys”; they were not collected systematically (K. Oten, Department of Agriculture
and Consumer Services, North Carolina Forest Service, Goldsboro, N, pers.
comm.). The second data source was the USDA Forest Service’s FIA database
(FIADB; O’Connell et al. 2015). For all of the models below, we used an initial filter
for the FIA data to include hemlock trees (FIA species codes 261 and 262) within
our study area (FIA Unit 4) that were observed by the annual inventory design for
at least the second time between 2003 and 2013.
Diameter-growth models
For diameter-growth regression models, we required that the trees were alive for
the entire interval, had a re-measurement interval of at least 2 years, were greater
than 12.7 cm (5 inches) diameter at breast height (1.37 m [4.5 feet] above ground;
dbh) at time 2 and greater than 2.54 cm (1 inch) dbh at time 1, and had dbh at time
2 greater than dbh at time 1. English units were used in our analyses.
To evaluate the effects of HWA presence in the county on hemlock tree growth
we performed a regression analysis. Our dependent variable of interest was the
average annual diameter growth between tree measurements.
For each tree i, the number of years after the year of confirmed presence of HWA
in the county, denoted as HWAC, that the time 2 observation of tree i occurred (Ti2)
was calculated using equation [1]:
Yi = Ti2 - HWAC, if positive; or 0, if otherwise [1]
Initially our independent variables of interest were Yi (defined above), and the following
variables, derived from FIADB: the dbh of each tree i at observation time 1
Table 1. FIA definitions of the variable Crown Class (CC), which describes relative crown position
in a stand.
Code Description
1 Open growth – trees with crowns that have received full light from above and from all sides
throughout all or most of their life, particularly during early development.
2 Dominant – trees with crowns extending above the general level of the canopy and receiving
full light from above and partly from the sides; larger than the average trees in the stand, and
with crowns well developed, but possibly somewhat crowded on the sides.
3 Co-dominant – trees with crowns forming part of the general level of the crown cover and
receiving full light from above, but comparatively little from the side. Usually with medium
crowns more or less crowded on the sides.
4 Intermediate – trees shorter than those in the preceding two classes, with crowns either below
or extending in to the canopy formed by the dominant and co-dominant trees, receiving little
direct light from above, and none from the sides; usually with small crowns very crowded on
the sides.
5 Overtopped – trees with crowns entirely below the general canopy level and receiving no
direct light either from above or the sides.
Southeastern Naturalist
635
J.T. Vogt, .A. Roesch, and M.J. Brown
2016 Vol. 15, No. 4
(di1), the crown class of tree i at observation time 1 (CCi1; Table 1), the percent slope
of the ground surrounding tree i (Slopei), the elevation of the ground surrounding
tree i (Elevi), the site class surrounding tree i (Sitei; Table 2), and the direction (in
degrees) of the slope surrounding tree i converted to the number of degrees from
due north (Aspecti). Crown class is an indication of the crown’s position relative
to other trees in the stand, and site class is the estimated or predicted productivity
of the site in terms of its capacity to grow crops of industrial wood. For missing
elevations in 16 of 796 observations, the mean elevation of all other trees was used.
Mortality models
For the mortality models, in addition to the initial filter for the FIA data, we also
required that the trees had a remeasurement interval of at least 1 year, were greater
than 2.54 cm (1 inch) in dbh at time 1, and the time 2 (mortality) observation occurred
at least one year after the confirmed presence of HWA in the county.
After performing some standard exploratory data analyses, we performed a conditional
logistic regression analysis. Our dependent variable of interest was annual
mortality. Four factors that seem likely to influence the longevity of hemlock trees
in the presence of HWA are the tree’s size, the length of time HWA has been present,
the tree’s crown position relative to the surrounding trees, and the quality of the
site in which the tree is growing. Consequently, our initial independent variables of
interest were the diameter of each tree i at 4.5 feet above the ground at observation
time 1 (di1), Yi (defined above), crown class of tree i at observation time 1, assigned
a value from1 to 3 defined using [2]
PCCi = 1, if CCi ≤ 3; 2 if CCi = 4; or 3 if CCi =5 [2]
the percent slope of the ground surrounding tree i , assigned a value from 1 to 3
defined using [3]
SlopeCi = 1, if Slopei ≤ 33; 2 if 33 < Slopei ≤ 50; or 3 if Slopei > 50, [3]
and site class surrounding tree i, assigned a value of 1 or 2 defined using [4]
SiteCi = 1, if SITECLCD > 4; or 2, if SITECLCD ≤ 4 [4]
The vigor of trees can have some relationship to the size of the tree, once other factors
such as crown class have been considered. We defined 3 size classes of trees
using tree dbh in [5]:
dbhCi = 1, if dbhi1 ≤ 6; 2 if 6 < dbhi1 < 10; and 3 if dbhi1 ≥ 10 [5]
Table 2. The linear conversion of FIA site class code to Site.
Site class code Estimated productivity Site
1 >15.7 m3/ha (>225 cubic feet/acre/year) 250
2 11.6–15.7 m3/ha (165–225 cubic feet/acre/year) 195
3 8.4–11.5 m3/ha (120–164 cubic feet/acre/year) 142
4 6.0–8.3 m3/ha (85–119 cubic feet/acre/year) 102
5 3.5–5.9 m3/ha (50–84 cubic feet/acre/year) 68
6 1.4–3.4 m3/ha (20–49 cubic feet/acre/year) 35
7 0–1.3 m3/ha (0–19 cubic feet/acre/year) 10
Southeastern Naturalist
J.T. Vogt, .A. Roesch, and M.J. Brown
2016 Vol. 15, No. 4
636
The variables above were collected into column vectors, with a row for each tree.
The vectors have the same names sans the subscript and are symbolized in bold
italics. For each model, the vector DeadTr is the response variable, also with a
row for each tree, in which a value for a tree is equal to 1 if the tree has died and 0
otherwise. In R (R Development Core Team 2008), we used glm in the stats package
with family equal to binomial(link = logit). The available explanatory variables
might suggest the following model [6] for mortality:
DeadTr ~ Y + strata(SiteC, SlopeC,PCC,dbhC), [6]
However, the available data were not sufficient to support such a large model. Additionally,
we note that SiteC and SlopeC are both attempts to estimate the assumed
quality of the land. If they are successful attempts, then they should be somewhat
redundant and it makes sense to eliminate one of those variables first. To this end,
we note that SiteC is a further categorization of a subjectively estimated quantity,
while SlopeC is a categorization of a directly measured variable, with an assumption
that land quality decreases with increased slope; another subjective judgement.
Because SlopeC is derived from an actual measurement, it has an intuitive appeal
that may be lacking in SiteC. On the other hand, foresters are quite adept at judging
the growing capacity of land and SiteC, having only 2 possible values, is a very
broad summarization of the exercise of this judgement. This leaves us with competing
reduced models [7] and [8]:
Mort1: DeadTr ~ Y + strata(SlopeC,dbhC,PCC), and [7]
Mort2: DeadTr ~ Y + strata(SiteC,dbhC,PCC). [8]
Mort1 results in 27 strata, while Mort2 results in 18 strata. The regression analysis
for Mort1 resulted in an insignificant stratum, whereas that did not occur for
Mort2. This suggests that we should either favor Mort2 or further collapse one of
the stratification variables. So we also considered [9]
Mort3: DeadTr ~ Y + strata(SlopeCC,PCC,dbhC), [9]
in which the individual constituents of SlopeCC are defined using [10]:
SlopeCCi = 1, if Slopei ≤ 50; or 2 if Slopei > 50, [10]
Finally, we considered the reduced models [11] and [12]:
Mort4: DeadTr ~ Y + strata(dbhC), and [11]
Mort5: DeadTr ~ Y + strata(SlopeC). [12]
An estimation matrix for each model was calculated from the resulting regression
coefficients. We symbolize the coefficient for year Y by b1 and for each stratum S
by bS. The estimation matrices each contain 13 columns, 1 for each year following
HWA confirmation 1 to 13, respectively. The number of rows for each model correspond
to the number of strata in the model. The estimate for each cell i (within
Year column C = 1 to 13, and stratum row S) was calculated using equation [13]:
Esti = bS + b1YC [13]
Southeastern Naturalist
637
J.T. Vogt, .A. Roesch, and M.J. Brown
2016 Vol. 15, No. 4
The logistic regression estimate, Esti , is the natural log of the odds of death, that is
the log of the probability of dying divided by the probability of living. Therefore
eEsti is the odds of death [14]:
eEsti = p(Deathi) / p(Survivali) = [1 - p(Survivali)] / p(Survivali)
eEsti + 1 = 1 / p(Survivali)
p(Survivali) = 1 / (1 + eEsti) [14]
Note that the stratum coefficient functions as the intercept for the stratum.
A survival matrix (Surv) of the same dimension was then obtained for each
model, calculating each cell using [15]
Survi = 1 / (1 + eEsti) [15]
Results
Diameter-growth models
We investigated a series of log transform models, based on reductions of a full
model. Here we’ll discuss the full model and one reduced model. The full model
[16] was:
ln (δ + 1) = b1ln(di1) + b2Yi + b3CCi1 + b4Slopei + b5Elevi +
b6Sitei + b7Aspecti [16]
Where n = regression parameters n = 1 to 7, ln = the natural logarithm (base e), and
δi = the annual change in diameter of tree i between times 1 and 2.
The reader should note that, in this model, we have used the simplifying assumption
that we can treat the 2 somewhat subjective, categorical variables (Sitei
and CCi) as continuous variables. In the case of CCi , the results suggest that this
was not an egregious assumption. We did not find the FIA site variable to be helpful
in any of the models. In the case of the full model above, Elev, Site, and Aspect
did not explain a significant amount of variation in annualized growth. A reduced
model was fit with a linear regression in R using the linear model function lm in
the stats package. Years since HWA confirmation and crown class (CC) were significant,
exerting negative and positive effects on predicted growth, respectively.
Slope exerted a negative effect on growth in the model [17]:
ln (δ + 1) = 0.0519 (± 0.0040) * ln(d1) - 0.0065 (± 0.0007) * Y +
0.0117 (± 0.0018) * CC1 - 0.0004 (± 0.0001) * Slope [17]
(F4, 788 = 423.5, P < 0.001, adj. r2 = 0.6809).
We developed growth curves to visualize and interpret our results. In Figure 2,
predicted dbh growth is plotted against dbh for a range of years since HWA confirmation,
holding crown class constant at 2 and slope at 50. In this case, growth
of surviving trees was reduced by ~50% with 12 years of HWA being reported in
the area (county). In Figure 3, we show the results when crown class is set at 2 and
5, and slope is set at 20 and 60. The predicted negative effect of slope was not as
pronounced for trees of higher crown class (less-dominant trees).
Southeastern Naturalist
J.T. Vogt, .A. Roesch, and M.J. Brown
2016 Vol. 15, No. 4
638
Mortality models
An examination of the proportion of hemlock trees and proportion of basal area
of hemlock trees that were alive at observation time 1 and died before observation
time 2 plotted against the year of observation time 2 (Fig. 4) reveals a dramatic
increase in mortality starting in 2007. Other exploratory analyses suggested that
counties with the earliest confirmation of HWA presence also had the earliest increases
in hemlock mortality (data not shown).
All of the mortality models suggested a ~50% chance of survival 10 to 12 years
after HWA confirmation in a county, with some variation in predicted survival due
to the other independent variables. Two of the simpler models are particularly instructive.
Model Mort4, which took into account tree size, demonstrated slightly
higher predicted survival for larger trees and an ~50% survival rate 12 years after
HWA confirmation (Table 3, Fig. 5). Model Mort5 predicted higher survival rates
Figure 2. Predicted annual diameter growth curves from the reduced model for a range of Y
values from 0 to 12 years since HWA confirmation and by initial diameter, for trees in the
dominant crown class with the percent slope fixed at 50.
Southeastern Naturalist
639
J.T. Vogt, .A. Roesch, and M.J. Brown
2016 Vol. 15, No. 4
for trees growing on less-severe slopes, and survival rates over time similar to
Mort4 (Table 4, Fig. 6).
Figure 3. Predicted annual diameter growth curves from the reduced model for a range of
Y values from 0 to 12 years since HWA confirmation by initial diameter. Predicted values
for trees in: the dominant crown class with the percent slope fixed at 60 (upper left), the
overtopped crown class with the percent slope fixed at 60 (upper right), the dominant crown
class with the percent slope fixed at 20 (lower left), the overtopped crown class with the
percent slope fixed at 20 (lower right).
Table 3. Coefficients and ANOVA for mortality model Mort4. Dispersion parameter for binomial
family taken to be 1. Deviance residuals: min = -1.4073, 1Q = -0.6319, median = -0.3680, 3Q =
-0.1951, and max = 2.8992. Null deviance = 1631.67 on 1177 degrees of freedom. Residual deviance
= 914.71 on 1173 degrees of freeom. AIC = 922.71. Number of Fisher scoring iterations = 5.
Variable Estimate Std. error Significance
Y 0.38279 0.03122 .001
dbhC = 1 -4.45038 0.31734 .001
dbhC = 2 -4.57219 0.31110 .001
dbhC = 3 -4.71780 0.32592 .001
Southeastern Naturalist
J.T. Vogt, .A. Roesch, and M.J. Brown
2016 Vol. 15, No. 4
640
Discussion
The number of years since HWA confirmation had a strong, highly significant negative
effect on annualized diameter growth for hemlock in this study. We found the
Table 4. Coefficients and ANOVA for mortality model Mort5. Dispersion parameter for binomial family
taken to be 1. Deviance residuals: min = -1.5188, 1Q = -0.6091, median = -0.3482, 3Q = -0.1762,
and max = 2.9610. Null deviance = 1631.67 on 1177 degrees of freedom. Residual deviance = 899.54
on 1173 degrees of freeom. AIC = 907.54. Number of Fisher scoring iterations = 5.
Variable Estimate Std. error Significance
Y 0.39408 0.03195 0.001
SlopeC = 1 -5.15938 0.34531 0.001
SlopeC = 2 -4.55141 0.31086 0.001
SlopeC = 3 -4.34889 0.31618 0.001
Figure 4. The proportion of hemlock trees and the proportion of basal area of hemlock trees
that were alive at observation time 1 and died before observation time 2, by year of observation
time 2, in FIA Survey Unit 4 in western North Carolina.
Southeastern Naturalist
641
J.T. Vogt, .A. Roesch, and M.J. Brown
2016 Vol. 15, No. 4
strength of this relationship somewhat surprising given the nature of information acquisition
for the presence of the adelgid and the scale at which that information could
be used. That scale resulted in a single value for each county. As such, the presence of
adelgid and county are confounded, and movement of the adelgid within counties is
not accounted for. The area in question has a high proportion of federally owned and
managed land that is not differentially managed by county, and to the best of our
knowledge, the county governments do not differentially manage land in any way
that might contribute to the effects that we have observed. Therefore we think this
confounding is unimportant and give it no weight. Unfortunately, windshield surveys
and other non-systematic means of detecting the presence of HWA don’t yield estimates
of sampling error. Dead and dying hemlock trees stand out in stark contrast
against healthy trees as foliage becomes grayish-green, alerting personnel to the possible
presence of the adelgid, which can then be confirmed by observing the insect.
Figure 5. Predicted survival of hemlock based on model Mort4, by tree size class, in the
years following confirmation of HWA presence in FIA Survey Unit 4 in western North
Carolina.
Southeastern Naturalist
J.T. Vogt, .A. Roesch, and M.J. Brown
2016 Vol. 15, No. 4
642
By the time HWA was detected in a county, it was likely there for some period of time
spanning one to several years, thus increasing the chance of a measurable signal with
respect to hemlock growth and mortality.
Other researchers have found that hemlocks in dominant and co-dominant
crown-class positions survive HWA longer than intermediate and overtopped trees
(Eschtruth et al. 2006, Onken 1995, Orwig and Foster 1998). Trees in lower canopy
positions receive less sunlight and probably have less-extensive root systems,
higher root competition, lower stem capacitance, and a lower carbohydrate reserve.
Trees that have slower radial growth prior to infestation may also be more
susceptible to severe infestation and die sooner (Davis et al. 2007). At first blush,
this may seem incongruous with our results for diameter growth owing to the
positive (but very small) parameter estimate for crown class, which was highly
significant. Due to the reverse ordering of crown classes, (that is 1 = open grown,
2 = dominant, etc.) it may seem that the coefficient should be negative. However,
Figure 6. Predicted survival of hemlock based on Model Mort5, by slope category, in the
years following confirmation of HWA presence in FIA Survey Unit 4 in western North
Carolina.
Southeastern Naturalist
643
J.T. Vogt, .A. Roesch, and M.J. Brown
2016 Vol. 15, No. 4
the growth models are based on the survivor trees, those surviving after significant
HWA infestations area-wide, and it is quite likely that the very weakest trees
have already died, so the growth models may be representing a (temporarily)
healthier population of trees.
Figure 4 shows that, for our data, hemlock mortality has increased dramatically
starting with observations made in 2007. Figure 4 plots the proportion of hemlock
trees and the proportion of basal area of hemlock trees that were alive at observation
time 1 and died before observation time 2, by year of observation time 2, in our
study area. The fact that the 2 curves closely coincide suggests that the increased
mortality is occurring across all size classes. In 2013, FIA started to directly observe
and record the presence and effect of HWA on hemlock trees, which will
allow for drawing of much stronger associations in the future.
Our findings that net growth decreased and mortality increased with duration
of infestation agreed with those of Morin and Liebhold (2015). Of the site factors
considered, only slope had a significant effect on net growth (Fig. 3) and mortality
(Fig. 6). Trees growing on poor sites (often related to slope) may be more susceptible
to HWA (David et al. 2007) as well as trees on steep slopes (Kantola et al.
2014, 2016). Other site factors did not exert significant effects on hemlock growth
and mortality in this study.
Forest Inventory and Analysis data are not specifically designed to evaluate impacts
of invasive pests such as HWA at a landscape scale, as pointed out by Kantola
et al. (2014); however, because the annualized inventory results in repeated observations
of individual trees over an extended period of time, we contend that it has
tremendous potential for examining individual tree characteristics (e.g., growth,
health, and mortality) over time as they are related to HWA and other pest infestations
and plot-level variables. The data are also useful for generating resource
estimates at the FIA-unit level and above to examine trends over time, and to provide
information at the state level that is timely and relevant. In the current study,
we were able to elucidate impacts of HWA on net growth and mortality of hemlock
over a relatively limited area and relatively short span of time, using standard measurements
taken on FIA plots. As plots are re-measured over time, entomologists
and others with an interest in invasive species, individual tree health, and forest
composition will have an increasingly useful, larger database to use to examine
trends and test hypotheses.
Acknowledgments
We thank Stan Zarnoch, Frank Koch, and 2 anonymous reviewers for their comments
on the manuscript. We gratefully acknowledge the efforts of our state partners with
North Carolina Forest Service, and the FIA field staff working on the ground to ensure
quality of data.
Literature Cited
Abella, S.R. 2014. Impacts and management of Hemlock Woolly Adelgid in national parks
of the eastern United States. Southeastern Naturalist 13:16–45.
Southeastern Naturalist
J.T. Vogt, .A. Roesch, and M.J. Brown
2016 Vol. 15, No. 4
644
Brown, M.J., and J.T. Vogt. 2015. North Carolina’s Forests, 2013. US Department of Agriculture,
Forest Service, Southern Research Station Resource, Asheville, NC. Bulletin
SRS-205. 92 pp.
Davis, D.D., M.S. Fromm, and M.D. Davis. 2007. Impact of the Hemlock Woolly Adelgid
on radial growth of Eastern Hemlock in Pennsylvania. Pp. 157–162, In D.S. Buckley
and W.K. Clatterbuck (Eds.). e-GTR SRS 101 Proceedings of the 15th Central Hardwood
Forest Conference. US Department of Agriculture, Forest Service, Southern Research
Station, Asheville, NC. 770 pp.
Ellison, A.M., M.S. Bank, B.D. Clinton, E.A. Colburn, K. Elliott, C.R. Ford, D.R. Foster,
B.D. Kloeppel, J.D. Knoepp, G.M. Lovett, J. Mohan, D.A. Orwig, N.L. Rodenhouse,
W.V. Sobczak, K.A. Stinson, J.K. Stone, C.M. Swan, J. Thompson, B. Von Holle, and
J.R. Webster. 2005. Loss of foundation species: Consequences for the structure and dynamics
of forested ecosystems. Frontiers in Ecology and the Environment 3:479–486.
Eschtruth, A.K., N.L. Cleavitt, J.J. Battles, R.A. Evans, and T.J. Fahey. 2006. Vegetation
dynamics in declining Eastern Hemlock stands: 9 years of forest response to Hemlock
Woolly Adelgid infestation. Canadian Journal of Forest Research 36(6):1435–1450.
Evans, A.M., and T.G. Gregoire. 2007. A geographically variable model of Hemlock Woolly
Adelgid spread. Biological Invasions 9:369.
Gillespie, A.J. 1999. Rationale for a national annual forest inventory program. Journal of
Forestry 97:16–20.
Havill, N.P., L.C. Vieira, and S.M. Salom. 2014. Biology and control of Hemlock Woolly
Adelgid. Technology Transfer FHTET-2014-05. Washington, DC. 32 pp.
Jetton, R.M., W.S. Dvorak, and W.A. Whittier. 2008. Ecological and genetic factors that
define the natural distribution of Carolina Hemlock in the southeastern United States
and their role in ex situ conservation. Forest Ecology and Management 255:3212–3221.
Kantola, T., P. Lyytikäinen-Saarenmaa, R.N. Coulson, S. Strauch, M.D. Tchakerian, M.
Holopainen, H. Saarenmaa, and D.A. Streett. 2014. Spatial distribution of Hemlock
Woolly Adelgid-induced hemlock mortality in the Southern Appalachians. Open Journal
of Forestry 4:492.
Kantola, T., P. Lyytikäinen-Saarenmaa, R.N. Coulson, M. Holopainen, M.D. Tchakerian,
and D.A. Streett. 2016. Development of monitoring methods for Hemlock Woolly
Adelgid-induced tree mortality within a Southern Appalachian landscape with inhibited
access. iForest-Biogeosciences and Forestry 9:178–186.
Lovett, G.M., C.D. Canham, M.A. Arthur, K.C. Weathers, and R.D. Fitzhugh. 2006. Forest
ecosystem responses to exotic pests and pathogens in eastern North America. Bioscience
56:395–405.
Martin, G.L. 1982. Notes: A method for estimating ingrowth on permanent horizontal
sample points. Forest Science 28:110-114.
McClure, M.S. 1991. Density-dependent feeback and population cycles in Adelges tsugae
(Homoptera: Adelgidae) on Tsuga canadensis. Environmental Entomology 20:258–264.
Morin, R.S., and A.M. Liebhold. 2015. Invasions by two non-native insects alter regional
forest species composition and successional trajectories. Forest Ecology and Management
341:67–74.
Nuckolls, A.E., N. Wurzburger, C.R. Ford, R.L. Hendrick, J.M. Vost, and B.D. Kloeppel.
2009. Hemlock declines rapidly with Hemlock Woolly Adelgid infestation: Impacts on
the carbon cycle of southern Appalachian forests. Ecosystems 12:179–190.
O’Connell, B.M., E.B. LaPoint, J.A. Turner, T.R. Ridley, S. Pugh, A.M. Wilson, K.L. Waddell,
and B.L. Conkling. 2015. The forest inventory and analysis database: Database
description and user Guide for phase 2 (version 6.0.2). US Department of Agriculture,
Forest Service, Washington, DC.
Southeastern Naturalist
645
J.T. Vogt, .A. Roesch, and M.J. Brown
2016 Vol. 15, No. 4
Onken, B.P. 1995. Long-term impact assessment of Eastern Hemlock forests. Proceedings
of the first Hemlock Woolly Adelgid review, Charlottesville, VA, 96-10. Pp.
58–63, In S.M. Salome, T.C. Tigner, and R.C. Reardon (Eds.). FHTET 96-10, US Department
of Agriculture, Forest Service, Forest health Technology Enterprise Team,
Morgantown, WV.
Orwig, D.A., and D.R. Foster. 1998. Forest response to the introduced Hemlock Woolly
Adelgid in southern New England, USA. Journal of the Torrey Botanical Society
125:60–73.
Orwig, D.A., D.R. Foster, and D.L. Mausel. 2002. Landscape patterns of hemlock decline
in New England due to the introduced Hemlock Woolly Adelgid. Journal of Biogeography
29:1475–1487.
R Development Core Team. 2008. R: A language and environment for statistical computing.
R Foundation for Statistical Computing, Vienna, Austria. Available online at http://
www.R-project.org. Accessed 21 March 2016.
Reams, G.A., W.D. Smith, M.H. Hansen, W.A. Bechtold, F.A. Roesch, and G.G. Moisen.
2005. The Forest Inventory and Analysis sampling frame. Pp. 21–36, In W.A. Bechtold
and P.L. Patterson (Eds.). The Enhanced Forest Inventory and Analysis Program: National
sampling design and estimation procedures. General Technical Report SRS-80.
US Department of Agriculture, Forest Service, Southern Research Station, Ahseville,
NC. 85 pp.
Rentch, J.S., H.S. Adams, R.B. Coxe, and S.L. Stephenson. 2000. An ecological study of a
Carolina Hemlock (Tsuga caroliniana) community in southwestern Virginia. Castanea
65:1–8.
Rentch, J., M.A. Fajvan, R.A. Evans, and B. Onken. 2009. Using dendrochronology to
model Hemlock Woolly Adelgid effects on Eastern Hemlock growth and vulnerability.
Biological Invasions 11:551–563.
Roesch, F.A. 2007. Compatible estimators of the components of change for a rotating-panel
forest-inventory design. Forest Science 53:50–61.
Royle, D., and R. Lathrop. 2000. The effects of site factors on the rate of hemlock decline:
A case study in New Jersey. P. 3, In K.A. McManus, K.S. Shields, and D.R. Souto (Eds.).
Proceedings, Symposium on Sustainable Management of Hemlock Ecosystems in Eastern
North America. General Technical Report NE-267. US Department of Agriculture,
Forest Service, Northeastern Research Station, Newtown Square, PA. 109 pp.
Small, M.J., C.J. Small, and G.D. Dreyer. 2005. Changes in a hemlock-dominated forest
following woolly adelgid infestation in southern New England 1. The Journal of the
Torrey Botanical Society 132:458–470.
Smith, W.B. 2002. Forest inventory and analysis: A national inventory and monitoring program.
Environmental Pollution 116:S233–S242.
Trotter, R.T., III, R.S. Morin, S.N. Oswalt, and A. Liebhold. 2013. Changes in the regional
abundance of hemlock associated with the invasion of Hemlock Woolly Adelgid (Adelges
tsugae Annand). Biological Invasions 15:2667–2679.
Vose, J.M., D.N. Wear, A.E. Mayfield, and C.D. Nelson. 2013. Hemlock Woolly Adelgid in
the southern Appalachians: Control strategies, ecological impacts, and potential management
responses. Forest Ecology and Management 291:209–219.
Young, J.A., and D.D. Morton. 2002. Modeling landscape-level impacts of HWA in
Shenandoah National Park. Pp. 73–85, In R.C. Reardon, B.P. Onken, and L. Lashomb
(Eds). Proceedings, Hemlock Woolly Adelgid in the Eastern United States Symposium.
New Jersey Agricultural Experiment Station, New Brunswick, NJ.