Both Cyclone‐induced and Convective Storms Drive Disturbance Patterns in European Primary Beech Forests

Abstract

Wind is the leading disturbance agent in European forests, and the magnitude of wind impacts on forest mortality has increased over recent decades. However, the atmospheric triggers behind severe winds in Western Europe (large‐scale cyclones) differ from those in Southeastern Europe (small‐scale convective instability). This geographic difference in wind drivers alters the spatial scale of resulting disturbances and potentially the sensitivity to climate change. Over the 20th century, the severity and prevalence of cyclone‐induced windstorms have increased while the prevalence of atmospheric instability has decreased and thus, the trajectory of Europe‐wide windthrow remains uncertain. To better predict forest sensitivity and trends of windthrow disturbance we used dendrochronological methods to reconstruct 140 years of disturbance history in beech‐dominated primary forests of Central and Eastern Europe. We compared generalized linear mixed models of these disturbance time series to determine whether large‐scale cyclones or small‐scale convective storms were more responsible for disturbance severity while also accounting for topography and stand character variables likely to influence windthrow susceptibility. More exposed forests, forests with a longer absence of disturbance, and forests lacking recent high severity disturbance showed increased sensitivity to both wind drivers. Large‐scale cyclone‐induced windstorms were the main driver of disturbance severity at both the plot and stand scale (0.1–∼100 ha) whereas convective instability effects were more localized (0.1 ha). Though the prevalence and severity of cyclone‐induced windstorms have increased over the 20 century, primary beech forests did not display an increase in the severity of windthrow observed over the same period.

Full text

1. Introduction
A total of 40% of Europe is covered with forest (182 million ha; Cook,2019), and windthrow is directly
responsible for more than 50% of damage reported in those forested areas each year (18.6 million m3 on
average; Gardiner etal., 2010; Schelhaas, 2008). Furthermore, total wind-induced forest damage has in-
creased since the early 20th century (Schelhaas etal.,2003; Seidl etal.,2017; Usbeck etal.,2010). Shifting
disturbance dynamics and managerial responses to disturbance can alter habitat provisioning (Bengtsson
etal.,2000; Kozák etal.,2018), regulate carbon storage (Burrascano etal.,2013; Carey etal.,2001; Harmon
etal.,1990; Luyssaert etal.,2008; Seedre etal.,2020), and impact the role of forestry in the European econo-
my (Leverkus etal.,2012; Müller etal.,2019). Thus, understanding the drivers of shifting wind disturbance
patterns is needed to facilitate informed decision-making for forest and conservation management, and
additionally quantify the future role of Europe's forests in global biogeochemical cycles.
Abstract Wind is the leading disturbance agent in European forests, and the magnitude of wind
impacts on forest mortality has increased over recent decades. However, the atmospheric triggers
behind severe winds in Western Europe (large-scale cyclones) differ from those in Southeastern Europe
(small-scale convective instability). This geographic difference in wind drivers alters the spatial scale of
resulting disturbances and potentially the sensitivity to climate change. Over the 20th century, the severity
and prevalence of cyclone-induced windstorms have increased while the prevalence of atmospheric
instability has decreased and thus, the trajectory of Europe-wide windthrow remains uncertain. To better
predict forest sensitivity and trends of windthrow disturbance we used dendrochronological methods to
reconstruct 140years of disturbance history in beech-dominated primary forests of Central and Eastern
Europe. We compared generalized linear mixed models of these disturbance time series to determine
whether large-scale cyclones or small-scale convective storms were more responsible for disturbance
severity while also accounting for topography and stand character variables likely to influence windthrow
susceptibility. More exposed forests, forests with a longer absence of disturbance, and forests lacking
recent high severity disturbance showed increased sensitivity to both wind drivers. Large-scale cyclone-
induced windstorms were the main driver of disturbance severity at both the plot and stand scale
(0.1–∼100ha) whereas convective instability effects were more localized (0.1ha). Though the prevalence
and severity of cyclone-induced windstorms have increased over the 20 century, primary beech forests did
not display an increase in the severity of windthrow observed over the same period.
Plain Language Summary Two main atmospheric patterns are driving European windthrow
with large-scale winter storms being more prevalent in Western Europe and summer thunderstorm-
generated winds being more prevalent in Eastern Europe. In central Europe, most forests display a
mixed-severity disturbance regime indicating that both large- and small-scale disturbances are occurring.
However, few studies have been conducted looking at the prevalence of large-scale winter windstorms
and small-scale summer thunderstorms. Here we found evidence for both, but recently large-scale winter
windstorms have had a greater impact.
PETTIT ET AL.
© 2021. The Authors.
This is an open access article under
the terms of the Creative Commons
Attribution-NonCommercial License,
which permits use, distribution and
reproduction in any medium, provided
the original work is properly cited and
is not used for commercial purposes.
Both Cyclone-induced and Convective Storms Drive
Disturbance Patterns in European Primary Beech Forests
J. L. Pettit1,2 , J. M. Pettit1,2, P. Janda1, M. Rydval1 , V. Čada1 , J. S. Schurman1,
T. A. Nagel3, R. Bače1, M. Saulnier1 , J. Hofmeister1, R. Matula1, D. Kozák1, M. Frankovič1,
D. O. Turcu4, M. Mikoláš1, and M. Svoboda1
1Faculty of Forestry and Wood Sciences, Czech University of Life Sciences Prague, Suchdol, Czech Republic,
2Department of Biology, Minot State University, Minot, ND, USA, 3Department of Forestry and Renewable Forest
Resources, Biotechnical Faculty, University of Ljubljana, Ljubljana, Slovenia, 4Forest Research and Management
Institute – Timişoara Branch (ICAS), Voluntari, Ilfov, Romania
Key Points:
• Windstorms and convective
instability drive primary forest
windthrow damage
• Topographic exposure and time and
severity of last disturbance moderate
windthrow
• Windstorms create higher severity
windthrow than convective
instability
Supporting Information:
Supporting Information may be found
in the online version of this article.
Correspondence to:
J. L. Pettit,
pettitjoey@gmail.com
Citation:
Pettit, J. L., Pettit, J. M., Janda, P.,
Rydval, M., Čada, V., Schurman, J. S.,
etal. (2021). Both cyclone-induced and
convective storms drive disturbance
patterns in European primary beech
forests. Journal of Geophysical Research:
Atmospheres, 126, e2020JD033929.
https://doi.org/10.1029/2020JD033929
Received 21 SEP 2020
Accepted 6 JAN 2021
Author Contributions:
Writing – original draft: J. M. Pettit,
P. Janda, M. Rydval, V. Čada, J. S.
Schurman, T. A. Nagel, R. Bače, J.
Hofmeister, R. Matula, D. Kozák, M.
Frankovič, M. Mikoláš, M. Svoboda
Writing – review & editing: M.
Saulnier, D. O. Turcu
10.1029/2020JD033929
RESEARCH ARTICLE
1 of 17

Journal of Geophysical Research: Atmospheres
Extratropical cyclones are one of two meteorological drivers of intense wind in Europe, and these cyclones
can influence forest areas in excess of hundreds of km2 (Brázdil etal.,2004). The temperature, pressure, and
humidity gradients produced by cyclones, and their associated fronts, can induce winds in excess of 20m/s
across large affected areas (Brázdil etal.,2004). As temperature and related atmospheric humidity values
have risen in Europe (Hartmann etal.,2013), fronts associated with cyclones have increased in strength
(Schemm etal.,2017). These fronts have caused windstorms in Western Europe, including several recent
events (e.g., storms Vivian, Lothar, and Martin) each of which disturbed more than 100 million m3 of debris
(Gardiner etal.,2010). The recent increase in strength of cyclones, usually in winter, has led some research-
ers to suggests that cyclone-induced windstorms are increasing the frequency and severity of windthrow
disturbance events (Schelhaas etal.,2003; Usbeck etal.,2010). However, although intense wind prevalence
has increased (Figure1), an increasing disturbance trend is not apparent in disturbance reconstructions
from primary and old-growth forests of Central or Eastern Europe (Čada etal.,2020; Firm et al., 2009;
Schurman etal.,2018; Zielonka etal.,2010).
Certain disturbance reconstructions from these forests show decreasing trends in disturbance severity over
the 20th century. This could be due in part to reductions of other disturbance and mortality agents (Seidl
et al., 2017). However, with the majority of European disturbance agents increasing in intensity (Seidl
etal.,2017) and the fact that mountain regions of Central and Eastern Europe are subject to significant
increases in the prevalence of intense wind speeds (see Figure3 in Donat et al.,2011 and Figure 1 in
this study), we would expect to see increases in disturbance severity if winds produced by cyclonic storms
are drastically increasing. Thus, large-scale cyclones may not be the sole windthrow driver in Central and
PETTIT ET AL.
10.1029/2020JD033929
2 of 17
Figure 1. Average annual (a) wind speed and (b) convective available potential energy (CAPE) based on 20th century
reanalysis data. Locations of Carpathian Mountain beech plots (the focal forests of this study) are shown as filled dots.
Trends in the prevalence of extreme days per year in Carpathian beech forests plots (open dots) show (c) intense wind
speed has increased and (d) severe CAPE has decreased over the 20th century. Local polynomial regression trend lines
are shown in red.

Journal of Geophysical Research: Atmospheres
Eastern Europe, or some moderator of wind (e.g. stand character and canopy roughness) may be reducing
the influence of these storms.
The second major meteorological driver of intense wind in Europe is atmospheric instability, which gener-
ates strong winds through microbursts and occasionally, tornadoes. Despite the fact that severe convective
storms are common across Europe (Taszarek etal.,2019), studies of convective storms on European forest
disturbance are limited (Brázdil etal.,2018; Furtuna etal.,2018; Nagel etal.,2006,2017). These convective
storm elements can create higher speed winds than extratropical cyclones but at smaller scales and for
shorter periods. For example, over the course of a few days, cyclones can elevate winds across hundreds of
km2; whereas most convective storms usually only have strong winds that affect less than 1km2 over the
course of ca. 1h (Brázdil etal.,2004). Thus, to help differentiate these drivers, we refer to these phenomena
in this text as “large-scale cyclones” and ‘small-scale convective instability'.
Throughout Europe, the temporal and regional patterns of small-scale convective instability differ from
cyclones (Figure1). Atmospheric conditions associated with strong cyclones (elevated atmospheric hu-
midity and subsequent front strength, Schemm etal.,2017) are on the rise, whereas conditions associated
with strong convective storms (high relative humidity which influences air parcel buoyancy; Del Genio
etal.,2007) are decreasing in prevalence as Northern Hemisphere temperatures rise (Hartmann etal.,2013).
In Southeastern Europe convective instability is more common (Furtuna etal., 2018; Nagel et al., 2017;
Taszarek etal.,2019), whereas cyclones dominate in Western Europe. Despite the lower prevalence of cy-
clone-induced winds in Central, Southern, and Eastern Europe, windthrow is still a major disturbance
agent driving forest dynamics there (Nagel etal.,2017; Sommerfeld etal.,2018; Synek etal.,2020). Distur-
bance reconstructions in old-growth European beech (Fagus sylvatica L.) dominated forests from Slovenia
to Montenegro show that windthrow is the most prevalent disturbance agent but large-scale stand-leveling
windstorms are not the norm (Furtuna etal., 2018; Nagel etal.,2017). Instead, smaller-scale windthrow
events (<10ha) are more common, and most occur during the summer suggesting that small-scale convec-
tive instability is the main driver and not large-scale cyclonic storms.
These two drivers of wind form a cline of disturbance size across Europe with large-scale disturbance being
common in Western Europe (e.g., France and Germany) and smaller-scale disturbance more common in
Southeastern Europe (e.g., Slovenia to Montenegro) coincident with the prevalence of intense windstorms
in the west (Bett etal.,2017) and intense convective instability in the southeast (Brooks etal.,2003; Taszarek
etal.,2019). Based on disturbance reconstructions from the Carpathian Mountains at the border of Central
and Eastern Europe, most disturbances are smaller scale (<10ha; Čada etal.,2020) and thus, if large-scale
cyclones are driving windthrow dynamics in these forests, some moderator of wind is increasing forest
resistance to large-scale disturbance. Otherwise, it may be that small-scale convective instability is at least
partially responsible for the smaller disturbance patch size there, just as it is in Southeastern Europe. It
should be noted, however, that most studies describing the large-scale wind disturbance in Western Europe-
an forests come from commercially managed forests where forest structure and composition have been al-
tered from a natural state, which has the potential to temporarily increase forest vulnerability to windthrow,
possibly biasing disturbance events toward the severe end of the spectrum (Everham & Brokaw,1996; Gar-
diner etal.,2005,2010; Quine & Gardiner,2007; Schelhaas etal.,2003). Thus, studies attempting to discern
the relative influence of large-scale cyclones and small-scale convective instability should analyze forests
that lack or control for evidence of management practices including stand thinning and lengthening of
rotation periods to reduce this possible disturbance severity and scale bias.
Regardless of the relative presence of large-scale cyclones or small-scale convective instability, forest expo-
sure and disturbance history can mediate windthrow severity. Variables influencing forest vulnerability to
windthrow that can be estimated or reconstructed across the 20th century include topographic exposure
(Quine & White,1998; Senf & Seidl,2018), and time series of forest disturbance severity and time since the
last disturbance (Janda etal.,2017). Topographic exposure is a measure of shielding based on a virtual hori-
zon angle at a fixed distance from a point on the map. Points at valley bottoms are more shielded from wind
than points at the peak of a mountain. Thus, forests on ridges and mountain peaks are more vulnerable than
those in valleys (Ruel etal.,2002). When measuring the windthrow risk of a particular stand, forest dynam-
ics can also influence susceptibility to windthrow. For example, a recently disturbed forest with few living
stems remaining will need adequate time for trees to regenerate before becoming susceptible to disturbance
PETTIT ET AL.
10.1029/2020JD033929
3 of 17

Journal of Geophysical Research: Atmospheres
again. Thus, variables like time since the last disturbance and the severity of the last disturbance are likely
to mediate future susceptibility to wind disturbance (Schurman etal.,2018). Also, as we discuss above, lon-
gitude and its association with temperature and humidity patterns can influence windthrow risk through
the relative presence of large-scale cyclone-induced windstorms (more common in Western Europe) versus
small-scale convective instability (more common in Southeastern Europe (Brázdil etal.,2004). Finally, the
presence of a temporal trend in disturbance severity datasets and the differing trends for the two main
drivers of European wind (Figures1c and1d) imply that storm severity or forest vulnerability are changing
through time (Schelhaas etal.,2003; Usbeck etal.,2010). Quantifying the effects of these wind moderat-
ing variables will inform forest susceptibility models of windthrow disturbance and elucidate atmospheric
drivers of wind.
Our objective in this study was to use a large network of primary forest plots, where the main disturbance
agent is wind, to determine the relative influence of wind drivers at two spatial scales and examine trends in
wind-induced forest disturbance over the 20th century. We checked for evidence of large-scale cyclone-in-
duced windstorms and small-scale convective instability in European beech forests at the plot (ca. 0.1ha)
and forest stand scales (ca. 100ha) while also controlling for windthrow susceptibility variables acting at
local and continental scales. By addressing these objectives, we can inform hypotheses presented in the
literature on the reason for recent increases in European windthrow disturbance which include an increase
in intense wind frequency (which we account for) and increases in forest management and cover (which
our methods exclude).
2. Materials and Methods
2.1. Study Plots
We assessed the historical disturbance of 20 beech-dominated primary mixed forests stands within the
Carpathian Mountains of Slovakia and Romania. The presence of primary forests was determined through
forest inventories in Slovakia (Kozák etal.,2018; Mikoláš etal.,2019; Sabatini etal.,2018; also see http://
remoteforests.org) and Romania (Kozák etal.,2018; Sabatini etal.,2018) and detailed descriptions of these
primary forest inventories can be found in the study by Mikoláš etal.,(2019). Primary forest stands occurred
in four geographic clusters which we refer to as landscapes (West Slovakia, East Slovakia, North Romania,
and South Romania) covering 42°–50° latitude and 14°–25° longitude, with plots ranging in elevation from
615 to 1,324m a.s.l (Figure2a). To obtain historical disturbance data from stands, we collected tree cores
from 280 circular plots randomly positioned within primary forests. We used ArcGIS 10.7 to randomly place
the 280 plots in non-overlapping pairs oriented along topography contours with plot pair centers positioned
80m apart (Figure2b). Only beech-dominated mixed forest plots were included in this study to ensure
that windthrow was the predominant disturbance agent (Nagel etal.,2006) and because beech-dominated
mixed forest is the most abundant forest type across temperate Europe, increasing comparability of these
results to other windthrow studies in Europe. The most common tree species within stands in order of
abundance were Fagus sylvatica (71%), Abies alba (15%), Picea abies (6%), and Acer pseudoplatanus (4%).
2.2. Historical Disturbance Chronology Calculation
Annual records of percent canopy area removed were created based on tree-ring data for each plot and
stand. At each plot, trees were selected for coring based on a hierarchy of size classes in a nested circle
design. All trees ≥6cm diameter at breast heigh (DBH) were cored up to 8m from the plot center. Also, a
quarter of canopy and subcanopy trees 10–20cm DBH, and all trees ≥20cm DBH were cored up to 17.84m
from the plot center. These cores were dried prior to mounting and sanding using consecutively finer grit
sandpaper (up to 1000 ANSI grit). Cores were visually crossdated and ring widths were measured using a
Lintab measuring machine and TSAP-Win software. Crossdating was verified with Cofecha and CDendro
software (Holmes,1983; Larsson,2003).
Within crossdated tree-ring series, we used two types of disturbance-indicating growth patterns to re-
construct disturbance events: (1) Rapid early growth of trees established under an open canopy and (2)
abrupt increases in growth, called releases (Altman etal.,2018). Open canopy established trees were
identified as individuals that exceeded a threshold value of mean growth from 5 to 15 years of growth
PETTIT ET AL.
10.1029/2020JD033929
4 of 17

Journal of Geophysical Research: Atmospheres
(Fraver & White,2005a). These threshold values were calculated based on logistic regressions compar-
ing empirical data from collected plot seedlings growing in the open canopy or closed canopy conditions
(Janda etal.,2017). Separate regressions were performed for common tree species present (Fagus syl-
vatica, Abies alba, Picea abies, Acer pseudoplatanus, and a group of all other species present pooled in
an additional group) in each landscape (i.e., five species groups * 4 landscapes=20 species: landscape
critical values). Releases, our second disturbance indicating growth pattern, were identified using the
absolute increase method (Fraver & White,2005b; Trotsiuk etal., 2014). This method compares the
mean growth in the 10 years before a focal year and the 10 subsequent years including the focal year
at all possible positions along an individual tree growth series. If the absolute increase in growth is
greater than a threshold value, then a release event is recorded at the focal year. We limit recorded re-
leases to one every 20years and no tree may record a release above a DBH where they are considered to
have attained canopy status. Because the average growth of the species within beech-dominated mixed
forests was different, we calculated absolute increase threshold values for each species group in each
landscape separately and the canopy position DBH cutoff for each species individually. Thus, release
events were defined as years when the growth comparisons indicate a growth increase exceeding 1.25
the standard deviation of increases in the landscape-species combination (Fraver & White,2005b; Trot-
siuk etal.,2014).
Both of these disturbance-indicating growth patterns were then transformed to a measure of canopy area
removed using methods of Lorimer and Frelich(1989) and Schurman etal.,(2018). We used power function
models that estimated canopy area from a disturbance indicating the tree's current DBH (Lorimer & Fre-
lich,1989). Power functions were specific to each tree species group in each landscape.
PETTIT ET AL.
10.1029/2020JD033929
5 of 17
Figure 2. (a) Spatial arrangement of beech dominated primary mixed forest plots in the Carpathian Mountains. Plot colors depict the landscape groupings
used for calculating local disturbance threshold values. Inset map shows Europe with a red bounding box representing the study region. (b) Example
arrangement of plots within one stand showing scale. Notice that the stand footprint is approximately 1km2 (100ha), the maximum size of disturbance usually
recorded from small-scale convective instability. (c) Plot level disturbance reconstruction. Bars represent the canopy area removed estimated from individual
tree disturbance indicating growth patterns and the kernel density smoothing line used to alleviate temporal uncertainty of event timing. (d) Stand level
disturbance reconstruction. Bars represent canopy area removed, estimated from individual tree disturbance indicating growth patterns, and the kernel density
smoothing line calculated at the stand scale.

Journal of Geophysical Research: Atmospheres
To better characterize the plot and stand level timing and severity of all tree level disturbance events of
both disturbance types, we created a raw chronology of pooled canopy area removed before fitting a kernel
density estimation (KDE) function to temporally smooth the disturbance chronology. Yearly values of tree
level canopy area removed were pooled at the plot (Figure2c) and subsequently stand level (Figure2d) to
create raw disturbance chronologies. Then, a kernel density function with a 30-year window was fit to plot
and stand chronologies of raw canopy area removed to create two spatial hierarchies of disturbance chro-
nologies (Trotsiuk etal.,2014).
Peaks in the plot and stand kernel density disturbance chronologies were used for identifying the timing and
severity of disturbance events and for calculating time series describing the time since the last disturbance
and last disturbance severity for each plot and stand (Schurman etal.,2018). We used three criteria for de-
termining a plot peak: (1) The kernel density disturbance chronology had to be increasing for at least the 5
previous years, (2) it had to exceed 10% canopy area removed, and (3) subsequent peaks had to be separated
by at least 10years. Because disturbance chronologies include data from the young growth of trees and
because trees that could record recent disturbance may not have reached our criteria diameter classes at the
time of sampling (median age of all trees in plots at 6cm DBH, the minimum sampling size, was 31years),
we truncated the historical disturbance chronologies so that 1989 represented the most recent year.
2.3. Windthrow Susceptibility Variables
Because past disturbance influences the susceptibility of a stand to future disturbance (Schurman
etal., 2018), we calculated the time series of time since the last disturbance and the severity of that last
disturbance for plots and stands based on kernel density function peaks (FigureS1). Beginning in the year
1600, values of time since the last disturbance and severity of the last disturbance were set to zero, and time
since the last disturbance increased by one every year without disturbance. After detecting the first distur-
bance peak greater than 10% canopy area removed in a kernel density disturbance chronology, the “severity
of last disturbance” variable was set to the peak severity of the disturbance and the “time since last distur-
bance” was reset to zero. After the first disturbance peak, the last disturbance severity and time since the
last disturbance values only changed if a disturbance peak with a magnitude greater than 10% of the canopy
area of the plot or stand occurred more than 10years after the last disturbance peak. We shifted values of
time since last disturbance and severity of last disturbance to 15years after the peak disturbance year (half
the length of the KDE function window; FigureS1), because kernel density peaks occur at the temporal
center of raw disturbance events. Thus, a 15-years shift is necessary to ensure that evidence of previous dis-
turbance, not disturbance during the focal period, was used to predict disturbance severity of the focal year.
The reason for incorporating time since the last disturbance and severity of the last disturbance is twofold.
First, those variables have been shown to influence the susceptibility of forests to future disturbance. Sec-
ond, because disturbance changes the structure of the forest, these variables can be interpreted as a rough
proxy for forest structural complexity. Forests with lower severity disturbance and disturbance that hap-
pened further in the past are more likely to display higher structural complexity (Janda etal.,2017; Meigs
etal.,2017). Thus, we are examining the direct effect of past disturbance on future disturbance as well as
approximating an indirect effect of structure on disturbance susceptibility.
Beyond time since the last disturbance and severity of the last disturbance, wind-induced disturbance in
forests is also likely moderated by factors including topographic exposure, longitude, and time period. Land-
scapes that are more exposed are more likely to experience higher severity disturbance with elevated wind
levels, thus, to account for topographic exposure, we calculated a distance-limited “topex” value for every
plot. Distance limited topex is −1 * the average of the virtual horizon angles up to a distance of 1km from
each plot center at the eight cardinal and intercardinal directions (Quine & White,1998; Ruel etal.,2002;
Schmidt etal.,2010). We calculated topographic exposure values in ArcMap 10.7 based on digital elevation
model layers retrieved from the USGS Earth Explorer (https://earthexplorer.usgs.gov/). Using this method,
more positive values occur on mountain tops where trees are more exposed to wind whereas lower val-
ues occur in valleys where trees are more shielded from the wind. Longitude moderates the wind-induced
disturbance because of the strength of various drivers of wind change as the climate becomes more con-
tinental. Areas closer to the west coast experience higher average wind speeds and more intense cyclones
and more continental areas experience more convective instability (Siedlecki,2009). Finally, there is some
PETTIT ET AL.
10.1029/2020JD033929
6 of 17

Journal of Geophysical Research: Atmospheres
evidence that the effect of wind speed and convection on forest disturbance may vary across the 20th cen-
tury (Gardiner etal.,2010; Schelhaas etal.,2003; Usbeck etal.,2010) possibly due to global change. Thus,
we also account for the influence of “time period” (i.e., year) as an explanatory variable in order to examine
temporal changes in susceptibility to wind.
2.4. Historical Meteorology Data
We use the 20th Century Reanalysis version 3 data set as a source of estimated atmospheric and wind
conditions from 1850 to 1989 (Compo etal.,2006,2011). At the time of writing, this reanalysis product
represents the most temporally extensive data set with the highest resolution which models wind speed
and convective instability. This reanalysis has been tested for accuracy (Slivinski etal.,2019) and has been
compared to other reanalysis datasets (Bett etal.,2017). Values for surface u-wind (zonal or east-west), sur-
face v-wind (meridional or north-south), and convective available potential energy (CAPE), a measure of
small-scale convective instability, were extracted at each plot and stand centroid for all 3-h time steps from
January 1, 1850 to December 31, 1989. Measures of vector wind were calculated from u-wind and v-wind
values at every time step as these are estimated from the temperature and pressure gradients produced by
large-scale patterns including extratropical cyclone windstorms (Leckebusch & Ulbrich,2004). Because
CAPE and wind speed values within the 20th century reanalysis data set are estimates from an ensemble
of models (Compo etal.,2006) and because trees in each area are likely adapted to the prevailing average
wind and storminess (Quine & Gardiner,2007), we create an annual time series of the number of time
steps where wind speed or CAPE were greater than two standard deviations above the mean for that area.
This method is analogous to using the 98th percentile as a threshold as Leckebusch etal.(2008) and Donat
etal.(2011) did when examining the influence of extreme storms on regions of Europe. Also, the number of
extreme wind speed and CAPE periods have previously been shown to correspond to storminess (Anyomi
etal.,2016; Leckebusch etal.,2008; Usbeck etal.,2010).
2.5. Models
Some uncertainty in the timing of disturbance events exists in the reconstructed disturbance dataset be-
cause there are lags in tree growth and recruitment after disturbance. We chose to address this uncertainty
by analyzing the historic dataset in 5-year periods. Our dependent variable, proportion canopy area re-
moved, and independent variables, time since last disturbance (≥10% canopy area removed), the severity
of last disturbance (≥10% canopy area removed), number of high windspeed 3-h intervals, and number of
high CAPE 3-h intervals were averaged into a time series of 5-year periods. Values for distance-limited top-
ographic exposure and longitude (additional explanatory variables) were static through time.
To address our primary goal of describing the influence of wind speed and storminess on primary mixed
forest disturbance, we fit seven univariate and four multivariate Bayesian regression models (Table1) to
the historical disturbance data set and compared models using leave-one-out cross validation which ranks
models based on predictive ability using an information theory approach. Because many plots had 5-year
periods without observed disturbance and because certain susceptibility criteria must be met for a plot to be
vulnerable to wind disturbance (e.g., just after a large disturbance there may not be live trees in the plot and
thus the plot will not be able to record disturbance regardless of wind speeds) we chose to use a two-part
zero-inflated beta model structure. This structure allows us to model and account for variables that increase
the probability of observing disturbance as well as modeling variables influencing disturbance severity. In
the disturbance presence portion of the model, we use a beta distribution model because we are modeling
disturbance severity as the proportion of plot or stand canopy area disturbed, which is bound between zero
and one. The model formula for all models tested can be summarized with the following equations:



i ii
Y ZIBeta p ;
(1)


 
i p pi
px
logit
(2)
log |


ii
x stand



1
(3)
PETTIT ET AL.
10.1029/2020JD033929
7 of 17

Journal of Geophysical Research: Atmospheres



p0,1logit
(4)



student t 3, 0,10‐
(5)



i
x student t 3, 0,10‐
(6)
Equation1 represents the full two-part mixture model where Yi is the modeled disturbance severity of
period i, pi is the probability that a period will exhibit zero disturbance, and ϕi is the mean of the beta
distribution model which estimates disturbance severity given certain forest and atmospheric conditions.
Equation2 and3 represent the respective forms of the zero-inflated and beta portions of the model where
the terms αp and αϕ are the zero-inflated and beta intercepts, the βpxi and βϕxi terms represent a general-form
parameter estimate for the independent variables in the models, and (1|stand) represents the random fac-
tor of stand used to account for our nested design. Equation4,5, and6 are prior distributions used for the
intercept (α) and parameter estimates (β) for the zero-inflated and beta portions of the model (Equation2
and3). All variable 5-year period averages were scaled to z-scores to increase computational efficiency and
so that a flat prior with a half Student's t distribution with 3 degrees of freedom and a scale parameter of
10 could be used in Bayesian regression models. Our four multivariate models (table1) can be described as
(1) the susceptibility variables model, where we account only for factors influencing forest susceptibility to
the wind (see windthrow susceptibility variables section above) but no wind or convective-instability term
is present. This model can be interpreted as a test of the influence of wind on forest disturbance. If this
is the best model, neither CAPE nor wind speed significantly influences forest disturbance. Multivariate
model 2 is the wind speed+susceptibility model, where the wind speed and all two-way interactions with
wind speed are added to the zero-inflation and beta portion of the model. This model tests for the influence
of extratropical cyclones. Multivariate model 3 is the convective instability+susceptibility model which
matches the form of the wind speed+susceptibility model with CAPE substituted for wind speed and it
tests for the influence of convective instability on forest disturbance. Multivariate model 4 is the global mod-
el which includes all terms from the wind speed model and convective instability model but also includes
PETTIT ET AL.
10.1029/2020JD033929
8 of 17
Model Form
Time since last dist. only α+β time since dist.
Severity of last dist. only α+β dist. severity
Topographic exposure only α+β topographic exposure
Time period only α+β time period
Lng. only α+β lng.
Cyclone wind only α+β WSPD
Convective inst. only α+β CAPE
Susceptibility vars. α+β time since dist.+β dist. severity+β topographic exposure+β
lng.+β time period
Cyclone wind+susceptibility α+(β time since dist.+β dist. severity+β topographic exposure+β
lng.+β time period) * β WSPD
Convective inst.+sus. α+(β time since dist.+β dist. severity+β topographic exposure+β
lng.+β time period) * β CAPE
Global α+(β time since dist.+β dist. severity+β topographic exposure+β
lng.+β time period) * (β WSPD+β CAPE)+β WSPD:β CAPE
Note. WSPD represents the number 3-h periods with intense cyclone-induced wind speeds and CAPE represents the
number of 3-h periods with intense convective instability.
Table 1
Fixed Effect Structure of both Zero-inflated and Beta Portions of Models Predicting disturbance Severity in Beech
Dominated Primary Mixed Forests of Central and Eastern Europe

Journal of Geophysical Research: Atmospheres
the two-way interaction between wind speed and CAPE. These same model forms were fit to both the plot
level and stand level datasets.
3. Results
3.1. Disturbance Reconstruction
All plots within mixed primary forests showed evidence of disturbance based on our reconstruction meth-
ods. When tree level disturbance events were aggregated to the stand level, reconstructions showed evi-
dence of disturbances covering larger areas but the maximum severity observed, measured as canopy area
removed, was 60% lower at the coarser stand level than at the plot level (see variability in Figure3 and Fig-
ureS2). The 5-year interval with the most severe disturbance event on the plot level was one in which 99%
of the canopy area was removed. This occurred in a plot within the Vihorlat stand in Slovakia during the
1885–1890 period. At the stand level, the largest reconstructed disturbance only removed 39% of the canopy
area and occurred in the Belia stand of Romania from 1895 to 1900 (FigureS2).
PETTIT ET AL.
10.1029/2020JD033929
9 of 17
Figure 3. Raw disturbance data observed in beech dominated mixed primary forest plots (blue dots) and stands (red dots) in comparison to hypothesized wind
disturbance moderating forest and location traits. Beta regression and logit β values are given for zero-inflated beta mixture models. The model predicted values
are displayed as a line with a 95% credible interval.

Journal of Geophysical Research: Atmospheres
Even though the range of maximum disturbance severity differed between plots and stands, many other
aspects of the plot and stand disturbance regimes were very similar. Plots had an average of 2.03±0.91 (1
SD) disturbance peaks from 1850 to 1989 where stands averaged 2.21±1.03. This implies a disturbance
return interval of 68years on the plot level and 62years on the stand level. Plot level disturbance averaged
over all plots from 1850 to 1989 was 7.5%±12.4% canopy area removed, and plot disturbance was below
10% canopy area removed for 75.7%±0.1% of the analyzed period. Stand level disturbance severity averaged
over all stands from 1850 to 1989 was 7.5%±6.2% canopy area removed and stand disturbance was below
10% canopy area removed for 75.0%±0.1 of the analyzed period.
3.2. Wind Disturbance Models
In our model comparison of wind drivers (e.g., large-scale extratropical cyclones and small-scale convec-
tive instability) and susceptibility variables, multivariate models that included both wind drivers and their
interactions with susceptibility variables outperformed models without wind as well as single predictor
models (Table2). Single predictor models (e.g., wind speed only, time since disturbance only, etc.) were
outperformed in every instance by multivariate models when predicting patterns of disturbance in forests of
Central and Eastern Europe. At the plot level, the best model overall was the global model which included
both cyclonic and convective wind drivers, the interactions between them, and the interactions between
drivers and susceptibility variables. A measure of model quality, the expected log predictive density (ELPD)
of the best model, estimated using leave-one-out cross-validation was >2.9 times better than the next best
model which only contained susceptibility variables and wind speed. However, when aggregating evidence
of disturbance to the stand level, two models show almost equal evidence toward best-describing patterns in
disturbance. These models are the global model and the wind speed and susceptibility model. Both models
include wind speed driven by large-scale cyclone-induced windstorms and two-way interactions between
wind speed and susceptibility variables, but the global model also contained convective instability and inter-
actions. Because the global model explains more variance, has a lower leave-one-out information criterion
score, shows a slightly better ELPD, and can be used to interpretall interactions, we focus discussion on
this model.
Examining the influence of susceptibility variables on historical disturbance severity, independent of chang-
es in wind driver variables, we saw that topographic exposure, the severity of the last disturbance, time
since last disturbance, and time interval were strong predictors of the presence and severity of plot level
disturbance (Figure3). Longitude, independent of wind variables, was not a good predictor of disturbance
PETTIT ET AL.
10.1029/2020JD033929
10 of 17
Plot level Stand level
Model ELPD ± SE R2Model ELPD ±SE R2
Global 0.0 ±0.0 0.05 Global 0.0 ± 0.0 0.27
Windspeed and susceptibility −25.3 ±8.6 0.04 Windspeed and susceptibility −1.8 ± 4.9 0.25
Convection and susceptibility −88.7 ±14.3 0.03 Convection and susceptibility −29.2 ± 10.0 0.15
Susceptibility vars. −112.6 ±16.4 0.02 Susceptibility vars. −46.1 ± 12.0 0.08
Times since last dist. only −161.2 ±19.7 0.01 Severity of last dist. only −46.8 ± 12.3 0.07
Severity of last dist. only −232.1 ±22.9 0.02 Time period only −49.0 ± 12.9 0.06
Time period only −273.1 ±25.6 0.01 Times since last dist. only −52.5 ± 13.5 0.05
Wind speed only −284.2 ±26.2 0.01 Intercept only −53.9 ± 13.7 0.05
Topographic exposure only −287.3 ±26.3 0.01 Topographic exposure only −54.0 ± 13.7 0.05
Longitude only −290.4 ±26.5 0.01 Wind speed only −54.5 ± 13.6 0.05
Convection only −292.3 ±26.5 0.01 Longitude only −54.6 ± 13.7 0.05
Intercept only −293.3 ±26.5 0.01 Convection only −55.2 ± 13.6 0.05
Note. Models predict disturbance severity in beech-dominated mixed primary forests of Central and Eastern Europe.
Table 2
Model Comparison Organized by Expected Log Predictive Density Based on Leave-One-Out Cross-validataion

Journal of Geophysical Research: Atmospheres
severity in plots as both zero-inflated and beta parameter estimate credible intervals overlapped zero. High
values of topographic exposure increased the probability of disturbance. High values of time since the last
disturbance increased disturbance probability and severity. High values of the severity of the last distur-
bance and time period decreased the probability and severity of disturbance. At the stand level, only severity
of the last disturbance, time since last disturbance, and time period increased disturbance severity. The
directions of these relationships were the same as when analyzing at the plot level: high values of previous
disturbance severity, recent time periods, and longer times since last disturbance increased disturbance
severity observed.
3.3. Wind Moderators
When interactions between susceptibility variables and wind drivers were considered, we saw some gener-
alizable patterns. Regardless of the wind driver or the scale, intense wind conditions lost their impact over
time. This trend was much more pronounced for the “CAPE:time period” interaction than for the “wind
speed:time period” interaction. When the intense cyclone-induced wind was observed in plots or stands
lacking previous high severity disturbance, predicted disturbance severity was higher. Strong interactions
of wind drivers with time since the last disturbance was apparent at the plot level only. When increased
time since the last disturbance was paired with intense wind or CAPE conditions, disturbance severity in-
creased. When greater topographic exposure was paired with intense wind or CAPE conditions, the result
was greater severity or probability of disturbance at the plot level, but when plot level estimates of exposure
were averaged to the stand level, parameter estimate credible intervals began to overlap zero (FigureS3
andS4). Interestingly, the influence of high wind speeds produced from extratropical cyclones did not
taper with longitude as expected, but the influence of convective instability did increase with longitude as
expected.
When we plotted conditional effects of wind interactions with multiple forest susceptibility traits (e.g., time
since last disturbance, severity of last disturbance, and topographic exposure), it was apparent that winds
from large-scale cyclone-induced windstorms were more influential to disturbance severity in mixed beech
primary forests of Central and Eastern Europe than winds from small-scale convective instability, espe-
cially in susceptible plots and stands. Susceptible plots were defined as those with susceptibility variables
one standard deviation higher than the mean (e.g., plots with >77years since the last disturbance [66 for
stands], the severity of last disturbance removed>50% of the canopy area [26% for stands], and had an
exposure value of>−1.6° [−4.17° for stands]). At the plot and stand levels, the slope of the “wind speed-sus-
ceptible” predicted line was greater than that of the “convective instability-susceptible” line and higher
values of disturbance severity were reached (ca. 5% more severe at 20 severe days per year; see Figure4).
Wind from cyclone-induced windstorms was interacting with forest susceptibility at the plot level, and this
interaction was maintained even when aggregating to the stand level so that regardless of scale, higher cy-
clone-induced wind speeds caused higher severity disturbance in susceptible forests. However, when exam-
ining how convective instability interacts with forest susceptibility, more susceptible plots exhibited higher
severity disturbance when instability was high (Figure4a), but at the stand level, the impact of convective
storms were reduced (Figure4c).
4. Discussion
In this study, we showed that intense wind speed prevalence, driven by large-scale extratropical cy-
clones, was the main driver of windthrow disturbance especially in susceptible plots and stands of pri-
mary mixed forest in Central and Eastern Europe. The influence of smaller-scale convective instability
on disturbance was also supported by models. Based on these data, increases in intense wind speed
prevalence observed over the 20th century have not resulted in an increase in the scale of disturbance
events observed. Additionally, coincident reductions in the prevalence of intense convective storms,
have resulted in a net decline of disturbance severity. Because this study was conducted in fixed area
plots located only in primary forests, changes in forest area and management do not influence these
results.
PETTIT ET AL.
10.1029/2020JD033929
11 of 17

Journal of Geophysical Research: Atmospheres
4.1. Intense Convective Instability
The highest-ranked models for plots and stands both included evidence of convective storms which indi-
cates that, in mixed forests of Central and Eastern Europe, thunderstorms with microbursts and tornadoes
are significantly impacting forest dynamics, a fact that has largely been ignored in the European windthrow
literature (Antonescu etal.,2017; Gardiner etal.,2010; Schelhaas etal.,2003; Usbeck etal.,2010). The influ-
ence of small-scale convective storms on forest disturbance has only been recorded in mixed forests of the
Dinaric Mountains of Southeastern Europe (Nagel etal.,2017), mixed forests of the Romanian Carpathian
Mountains (Furtuna etal.,2018), and in the deciduous forests of North America (Canham et al.,2001;
Peterson & Pickett,1991). Interestingly, these studies as well as the current study all focus on mixed forests
dominated by a deciduous species which likely reduces the susceptibility of forests to windthrow in winter
when trees lack leaves. Thus, we would not expect to see the same relative influence of convective storms
in coniferous forests lacking deciduous trees. Additionally, the lack of convective forest disturbance studies
is probably due to the smaller and more heterogeneous average footprint of convective-induced windthrow
events, usually much less than 1km2 (Canham etal.,2001; Nagel etal.,2017). Thus, because the average
area covered by stands analyzed here was 1km2, the influence of convective winds was not as influential
at the stand scale. Disturbance from small-scale convective storms may have only affected one plot or a
few trees and could be averaged out when calculating the disturbance severity at the stand scale. However,
small-scale disturbance dynamics such as those created by small-scale convective storms cannot be ignored
as they create forests of high structural complexity by opening up gaps in which recruitment can diversify
the age and size profile of the forest (Franklin etal.,2002; Lorimer & Halpin,2014; Meigs etal.,2017; Tepley
etal.,2013). These localized events increase forest horizontal and vertical heterogeneity by creating gaps for
seedling/sapling recruitment.
PETTIT ET AL.
10.1029/2020JD033929
12 of 17
Figure 4. Predicted disturbance values per number of intense wind speed (panel a and c) and CAPE days (panels b
and d) from the global zero-inflated beta models fit to plot (panels a and c) and stand (panels b and d) data from beech
dominated mixed primary forests of Central and Eastern Europe. Susceptible plot and stand lines represent those
that have a topographic exposure and time since last disturbance values one sd above the mean and severity of last
disturbance one sd below the respective mean observed for those variables. CAPE: convective available potential energy.

Journal of Geophysical Research: Atmospheres
In addition to the smaller size of the intense wind footprint of convective storms, dynamics specific to con-
vective instability also played a part in reducing the predicted impact of convective instability in forests. The
CAPE values from 20th century reanalysis data used as our measure of convective instability are modeled
on a 1°×1° grid and represent conditions favorable for producing uplift in the atmosphere but should not
be interpreted as implying that a disturbance-inducing storm will form or that a storm will cover the entire
grid cell. Thus, high values of CAPE are not always associated with forest disturbance and only increase the
probability of exposure to high winds. We accounted for this in two ways: (1) We used zero-inflated models
in order to account for the probability of observing no disturbance despite having elevated CAPE and (2) we
used the number of extreme events per year as an independent variable which has been shown to influence
windthrow (Leckebusch etal.,2008; Usbeck etal.,2010).
Convective storms had a greater influence with increasing longitude, potentially indicating the influence
of a more continental climate. Also, there was a strong and consistent interaction between longitude and
intense CAPE observed in the plot level global model (FigureS4). Thus, the forests of Romania are more
vulnerable to convective storms, probably because convective instability is more likely to be intense in many
areas of Romania compared to Slovakia (Taszarek etal., 2019). This pattern of more intense conditions
with continentality has been noted in previous research on storm prevalence across Europe (Antonescu
etal.,2017; Brooks etal.,2003) and is probably playing a role in driving this pattern. However, this is one of
the first studies to show this pattern with empirical forest disturbance data.
4.2. Intense Wind Speed
Intense wind speeds were a primary driver of disturbance severity regardless of the scale measured. This
pattern was true across the longitudinal cline of this study making this one of a limited number of studies at-
tributing large-scale cyclone-induced windstorms to forest disturbance in Romania (Gardiner etal.,2010).
The interactions between the prevalence of intense cyclone-induced wind speeds and susceptibility varia-
bles were also stronger and more consistent than those of the smaller scale intense winds created by small-
scale convective instability, likely reflecting the larger footprint of cyclones (Brázdil etal.,2018; Gardiner
etal.,2010; Leckebusch etal.,2008; Leckebusch & Ulbrich,2004; Usbeck etal.,2010). It is even likely that
this pattern would have been observed at the landscape scale based on the average size of large-scale cy-
clone-induced windstorms (Brázdil etal.,2004) and the strength of these patterns in these data.
The fact that cyclone-induced windstorms are driving disturbance dynamics at stand scales has implications
for stand structure. Large severe disturbances can reduce local structural complexity (Janda etal.,2017;
Meigs etal.,2017), however, even the most severe stand level disturbance observed here would only be
classified as low or possibly moderate severity in the global/European context (see predicted disturbance
severities around 0.15 proportion canopy area removed in Figure4). So, even the stand scale disturbances
observed here are still increasing structural complexity of stands through gap creation and patch dynamics,
only the gap sizes are likely a bit larger than those produced through convective instability.
4.3. Trends Over Time
The link between both convective storms and cyclone-induced windstorms and forest disturbance sever-
ity has weakened over the 20th century. This trend is more understandable for intense CAPE, which has
decreased in prevalence over the 20th century and will likely continue to do so (Figure1). Reductions in
CAPE values are caused by reductions in relative humidity with warming and the subsequent changes in
air parcel buoyancy based on adiabatic lapse rates (Riemann-Campe etal.,2009). As temperatures continue
to increase, atmospheric humidity values will increase but relative humidity values will continue to decline
(Hartmann etal.,2013). Rising atmospheric humidity will cause intense wind speed prevalence to rise,
though some uncertainty in the extent of this pattern exists due to changes in cyclone storm path changes
(Sepp etal.,2005). Despite the observed increases in intense wind speed prevalence over the 20th century,
we observed a reduction in the influence of wind on disturbance severity (i.e. the same prevalence of intense
wind causes lower severity disturbance today than in the past). This result is in direct contrast to hypoth-
eses made in previous studies of windthrow disturbance in European forests that, based on data from sal-
vage logging and reported damage, surmise increasing prevalence of intense winds may be responsible for
PETTIT ET AL.
10.1029/2020JD033929
13 of 17

Journal of Geophysical Research: Atmospheres
disproportionate increases in catastrophic windthrow damage (Gardiner etal.,2010; Schelhaas etal.,2003;
Usbeck etal.,2010). However, these studies rightly list the caveat that windthrow and salvage logging re-
ports are spatially and temporally inconsistent and reports may be disproportionately more common from
intensively managed stands and in recent periods, reducing the applicability of their results for unmanaged
primary forest ecosystems (Everham & Brokaw,1996). These studies, however, are able to incorporate data
from recent periods which our study was not able to do, due to the ca. 30years time delay required when
using dendrochronological methods to reconstruct disturbance history. Thus, we could not incorporate data
from recent severe windstorms like Lothar, Martin, and Kyrill, which would have almost certainly inflated
the impact of recent wind speed increases on forest disturbance. All time periods analyzed here maintain
a sufficient presence of trees with the potential to record disturbance (FigureS5). Based on the number of
trees with the ability to record disturbance, we see that the recent portion of the 20th century that we ana-
lyzed was well represented and if the disturbance was present, it could have been recorded by trees.
Other hypotheses given for the elevated levels of recently reported damage in Western Europe windthrow
studies are an increase in forest cover across Europe (Gardiner etal.,2010) and increases in the age of the
average European forest (Schelhaas etal., 2003). Increases in forest area likely led to increases in forest
disturbance, however by using data from a plot census in this study, the area represented by disturbance re-
cording trees has not drastically changed. Thus, our data do not directly test this hypothesis but do exclude
forest cover increases as a potential bias. As for the increasing age of European forests due to the length-
ening of rotation periods, this may make even-aged managed forests more vulnerable to wind disturbance.
However, primary forest plots are usually older on average and their lack of stand-leveling disturbance
events implies that older trees remain present within forests. Despite this presence of old trees, which sug-
gests primary forests should be more vulnerable, we observed a decrease in disturbance severity. Thus,
based on these data, it may be more likely that disproportionate increases in wind disturbance observed in
Western Europe may be due both to forest area increases and the prevalence of maturing, even-aged mon-
ocultures in commercial forests there.
4.4. Structural Complexity
Previous studies have hypothesized the potential for stands high in structural complexity to have increased
resistance to windthrow (Everham & Brokaw,1996; Gardiner etal.,2010; Mitchell,2013). Stands in this
study show increasing resistance to windthrow and likely have high structural complexity induced by
changes in the mixed-severity disturbance regime (Figures 3 b and 3c). Across the time period analyzed, dis-
turbance severity decreased and the time since disturbance increased on average. Both of these variable tra-
jectories can be interpreted as increases in forest structural complexity based on previous research in spruce
forests (Janda etal.,2017; Meigs etal.,2017). Also, as the forests in this study are aging and increasingly
exhibit old-growth structure traits (e.g. increased canopy height roughness), their resistance to wind distur-
bance may be increasing as well (Mitchell,1995). However, the link between structure and disturbance was
not directly measured here and should be interpreted with caution. Previous studies that have attempted
to account for stand structure as a susceptibility variable have not shown a clear and obvious influence
of structure on windthrow susceptibility (Barry Gardiner etal., 2005; Mitchell,2013). Regardless, the fact
that disturbance probability is lower recently even when the intense wind is controlled for (Figure3e), is
evidence that something about these forests has changed and increased resistance. Thus, future research on
the influence of structural complexity on windthrow vulnerability is warranted.
4.5. Susceptibility to Windthrow
Here we controlled for five variables that we hypothesized moderated wind-induced disturbance in beech
dominated mixed forests of Central and Eastern Europe, all influenced disturbance in the hypothesized
manner, but patterns were scale specific or only interacted with one of the two wind drivers. Longitude
only interacted with small-scale convective instability at smaller, plot scales. The fact that longitude did
not show strong interactions with the larger-scale cyclone-induced windstorms indicates that intense
winds were ubiquitous across the Carpathian Mountains and the influence of intense wind speeds was
consistent across the longitudinal gradient. Despite evidence that extratropical cyclones are not likely to
reach deep into continental Europe due to the East-Central European High (Di Rita etal.,2018) and the
PETTIT ET AL.
10.1029/2020JD033929
14 of 17

Journal of Geophysical Research: Atmospheres
poleward movement trend of extratropical cyclones making landfall in Europe (Leckebusch etal.,2008;
Sepp etal.,2005), the absolute area of elevated wind speeds that cyclones produce covers the full gradient of
longitudes studied here (14°–25°). Topographic exposure was consistent as a moderator of wind disturbance
regardless of the wind driver but only at the plot scale. This result has been observed before and, thus, many
windthrow susceptibility models include distance-limited topographic exposure as a predictor (Quine &
White,1998; Ruel etal.,2002; Schmidt etal.,2010). The reduced interaction between winds and topograph-
ic exposure at the stand scale was expected because stand measures of topographic exposure represented
an average of plot level exposure, which reduced differences between stand measurements. Beyond topo-
graphic exposure, topographic roughness has been shown to moderate the scale of disturbances, which may
be an additional reason that we did not observe severe disturbances at the larger stand scale as our stands
are mostly in or near mountainous areas (Senf & Seidl,2018). Previous disturbance also influenced future
disturbance severity as expected, with shorter time interval since the last disturbance and higher severity of
the last disturbance associated with higher severity forest leveling, i.e., when more trees are present, more
forest can be disturbed (Janda etal.,2017; Meigs etal.,2017).
There are variables that we do not include in this analysis that are known moderators of windthrow distur-
bance. These include soil depth, soil moisture, tree height, and stand density (Canham etal.,2001; Gardiner
etal.,2008; Nicoll etal.,2008; Usbeck etal.,2010), yet reliably reconstructing these variables over the anal-
ysis period is a large feat outside of the scope of this analysis. Because our main goals were to determine
if large-scale cyclone-induced storms were the only source of wind disturbance in Central and Eastern
European forests and to describe changes in the influence of wind drivers over the 20th century, maximiz-
ing model fit was not essential to achieve these goals. Inclusion of other drivers and moderators of forest
dynamics would almost certainly increase the fit of models, and the low estimated R2 values observed for
plot scale models are evidence of this, but finding strong and significant trends of cyclones, convection, and
wind moderators across the 20th century had never been accomplished prior to this study. Thus, this study
provides essential information based on readily available forest positions and past disturbance data that can
be used to predict the risk of future windthrow disturbance.
Though large-scale cyclone-induced windstorms are driving small- and large-scale disturbances in primary
forests, the influence of convective storms cannot be ignored. The fact that both intense windstorms and
intense convective instability are less influential in recent years may be evidence that primary forests are
less susceptible to windthrow due to changes in the disturbance regime. Forest susceptibility variables used
here can be extrapolated to many forests. Forest exposure, time since the last disturbance, and severity of
the last disturbance can be incorporated into censuses to map and monitor forest vulnerability in response
to predicted increases in intense cyclone activity.
Data Availability Statement
Datasets for this research are available in these in-text data citation references (Pettit,2020), with CC 4.0
license https://doi.org/10.6084/m9.figshare.12983003.v1.
References
Altman, J., Ukhvatkina, O. N., Omelko, A. M., Macek, M., Plener, T., Pejcha, V., etal. (2018). Poleward migration of the destructive effects
of tropical cyclones during the 20th century. Proceedings of the National Academy of Sciences 115, 11543–11548. https://doi.org/10.1073/
pnas.1808979115
Antonescu, B., Schultz, D. M., Holzer, A., & Groenemeijer, P. (2017). Tornadoes in Europe: An underestimated threat. Bulletin of the Amer-
ican Meteorological Society, 98(4), 713–728. https://doi.org/10.1175/bams-d-16-0171.1
Anyomi, K. A., Mitchell, S. J., & Ruel, J.-C. (2016). Windthrow modeling in old-growth and multi-layered boreal forests. Ecological Model-
ling, 327, 105–114. https://doi.org/10.1016/j.ecolmodel.2016.02.003
Bengtsson, J., Nilsson, S. G., Franc, A., & Menozzi, P. (2000). Biodiversity, disturbances, ecosystem function and management of European
forests. Forest Ecology and Management, 132(1), 39–50. https://doi.org/10.1016/s0378-1127(00)00378-9
Bett, P. E., Thornton, H. E., & Clark, R. T. (2017). Using the twentieth century reanalysis to assess climate variability for the European wind
industry. Theoretical and Applied Climatology, 127(1–2), 61–80. https://doi.org/10.1007/s00704-015-1591-y
Brázdil, R., Blennow, K., Dobrovolný, P., Štekl, J., Kotyza, O., Valášek, H., & Jež, J. (2004). History of weather and climate in the Czech lands
VI: Strong winds.
Brázdil, R., Stucki, P., Szabó, P., Řezníčková, L., Dolák, L., Dobrovolný, P., etal. (2018). Windstorms and forest disturbances in the Czech
Lands: 1801-2015. Agricultural and Forest Meteorology, 250–251, 47–63. https://doi.org/10.1016/j.agrformet.2017.11.036
PETTIT ET AL.
10.1029/2020JD033929
15 of 17
Acknowledgments
We would like to thank the Maramureş
Mountains Nature Park and Semen-
ic-Caraș Gorge National Park for their
assistance with site access and surveys.

Journal of Geophysical Research: Atmospheres
Brooks, H. E., Lee, J. W., & Craven, J. P. (2003). The spatial distribution of severe thunderstorm and tornado environments from global
reanalysis data. Atmospheric Research, 67–68(68), 73–94. https://doi.org/10.1016/s0169-8095(03)00045-0
Burrascano, S., Keeton, W. S., Sabatini, F. M., & Blasi, C. (2013). Commonality and variability in the structural attributes of moist temperate
old-growth forests: A global review. Forest Ecology and Management, 291, 458–479. https://doi.org/10.1016/j.foreco.2012.11.020
Čada, V., Trotsiuk, V., Janda, P., Mikoláš, M., Bače, R., Nagel, T. A., etal. (2020). Quantifying natural disturbances using a large-scale den-
drochronological reconstruction to guide forest management Ecological Applications.
Canham, C. D., Papaik, M. J., & Latty, E. F. (2001). Interspecif ic variation in susceptibility to windthrow as a function of tree size and storm
severity for northern temperate tree species. Canadian Journal of Forestry and Research, 31(1), 1–10. https://doi.org/10.1139/x00-124
Carey, E. V., Sala, A., Keane, R., & Callaway, R. M. (2001). Are old forests underestimated as global carbon sinks? Global Change Biology,
7(4), 339–344. https://doi.org/10.1046/j.1365-2486.2001.00418.x
Compo, G. P. J. S., & Whitaker, P. D. (2006). Feasibility of a 100-year reanalysis using only surface pressure data. Bulletin of American Me-
teorological Society, 87(2), 175–190. https://doi.org/10.1175/bams-87-2-175
Compo, G. P., Whitaker, J. S., Sardeshmukh, P. D., Matsui, N., Allan, R. J., Yin, X., etal. (2011). The Twentieth Century Reanalysis Project.
Quarterly Journal of the Royal Meteorological Society, 137(654), 1–28. https://doi.org/10.1002/qj.776
Cook, E. (Ed.). (2019). Agriculture, forestry and fishery statistics, 2019 edition. Publications Office of the European Union.
Del Genio, A. D., Yao, M.-S., & Jonas, J. (2007). Will moist convection be stronger in a warmer climate?: Convection strength in a warmer
climate. Geophysical Research Letters, 34(16). https://doi.org/10.1029/2007gl030525
Di Rita, F., Fletcher, W. J., Aranbarri, J., Margaritelli, G., Lirer, F., & Magri, D. (2018). Holocene forest dynamics in central and west-
ern Mediterranean: Periodicity, spatio-temporal patterns and climate influence. Scientific Reports, 8(1). https://doi.org/10.1038/
s41598-018-27056-2
Donat, M. G., Renggli, D., Wild, S., Alexander, L. V., Leckebusch, G. C., & Ulbrich, U. (2011). Reanalysis suggests long-term upward trends
in European storminess since 1871. Geophysical Research Letters, 38(14). https://doi.org/10.1029/2011gl047995
Everham, E. M., & Brokaw, N. V. L. (1996). Forest damage and recovery from catastrophic wind. The Botanical Review, 62(2), 113–185.
https://doi.org/10.1007/bf02857920
Firm, D., Nagel, T. A., & Diaci, J. (2009). Disturbance history and dynamics of an old-growth mixed species mountain forest in the Slove-
nian Alps. Forest Ecology and Management, 257(9), 1893–1901. https://doi.org/10.1016/j.foreco.2008.09.034
Franklin, J. F., Spies, T. A., Pelt, R. V., Carey, A. B., Thornburgh, D. A., Berg, D. R., etal. (2002). Disturbances and structural development
of natural forest ecosystems with silvicultural implications, using Douglas-fir forests as an example. Forest Ecology and Management,
155(1–3), 399–423. https://doi.org/10.1016/s0378-1127(01)00575-8
Fraver, S., & White, A. S. (2005a). Disturbance dynamics of old-growth Picea rubens forests of northern Maine. Journal of Vegetation Sci-
ence, 16(5), 597. https://doi.org/10.1111/j.1654-1103.2005.tb02401.x
Fraver, S., & White, A. S. (2005b). Identifying growth releases in dendrochronological studies of forest disturbance. Canadian Journal of
Forestry and Research, 35(7), 1648–1656. https://doi.org/10.1139/x05-092
Furtuna, P., Haidu, I., & Maier, N. (2018). Synoptic processes generating windthrow. A case study for Apuseni mountains (Romania).
Geographia Technica, 13(2), 52–61. https://doi.org/10.21163/gt_2018.132.04
Gardiner, B, Blennow, K., Carnus, J.-M., Fleischer, P., Ingemarson, F., Landmann, G., etal. (2010). Destructive storms in European forests:
Past and Forthcoming impacts, 138.
Gardiner, B., Byrne, K., Hale, S., Kamimura, K., Mitchell, S. J., Peltola, H., & Ruel, J.-C. (2008). A review of mechanistic modeling of wind
damage risk to forests. Forestry, 81(3), 447–463. https://doi.org/10.1093/forestry/cpn022
Gardiner, B., Marshall, B., Achim, A., Belcher, R., & Wood, C. (2005). The stability of different silvicultural systems: a wind-tunnel investi-
gation. Forestry: An International Journal of Forest Research, 78(5), 471–484. https://doi.org/10.1093/forestry/cpi053
Harmon, M. E., Ferrell, W. K., & Franklin, J. F. (1990). Effects on carbon storage of conversion of old-growth forests to young forests. Sci-
ence, 247(4943), 699–702. https://doi.org/10.1126/science.247.4943.699
Hartmann, D. L., Tank, A. M. K., Rusticucci, M., Alexander, L. V., Brönnimann, S., Charabi, Y. A. R., etal. (2013). In Climate change 2013
the physical science basis: Working group I contribution to the fifth assessment report of the intergovernmental panel on climate change (pp.
159–254). Cambridge University Press. Observations: Atmosphere and surface.
Holmes, R. L. (1983). Computer-assisted quality control in tree-ring dating and measurement. Tree-Ring Bulletin, 43, 69–78.
Janda, P., Trotsiuk, V., Mikoláš, M., Bače, R., Nagel, T. A., Seidl, R., etal. (2017). The historical disturbance regime of mountain Norway
spruce forests in the Western Carpathians and its influence on current forest structure and composition. Forest Ecology and Manage-
ment, 388, 67–78. https://doi.org/10.1016/j.foreco.2016.08.014
Kozák, D., Mikoláš, M., Svitok, M., Bače, R., Paillet, Y., Larrieu, L., etal. (2018). Profile of tree-related microhabitats in European primary
beech-dominated forests. Forest Ecology and Management, 429, 363–374. https://doi.org/10.1016/j.foreco.2018.07.021
Larsson, L. (2003). CDendro: Cybis Dendro dating program. öbaden,Cybis Elektronik & Data AB.
Leckebusch, G. C., & Ulbrich, U. (2004). On the relationship between cyclones and extreme windstorm events over Europe under climate
change. Global and Planetary Change, 44(1–4), 181–193. https://doi.org/10.1016/j.gloplacha.2004.06.011
Leckebusch, G. C., Weimer, A., Pinto, J. G., Reyers, M., & Speth, P. (2008). Extreme wind storms over Europe in present and future climate:
A cluster analysis approach. Metz, 17(1), 67–82. https://doi.org/10.1127/0941-2948/2008/0266
Leverkus, A. B., Puerta-Piñero, C., Guzmán-Álvarez, J. R., Navarro, J., & Castro, J. (2012). Post-fire salvage logging increases restoration
costs in a Mediterranean mountain ecosystem. New Forests, 43(5–6), 601–613. https://doi.org/10.1007/s11056-012-9327-7
Lorimer, C. G., & Frelich, L. E. (1989). A methodology for estimating canopy disturbance frequency and intensity in dense temperate
forests. Canadian Journal of Forest and Research, 20(5), 615615. https://doi.org/10.1139/x90-082
Lorimer, C. G., & Halpin, C. R. (2014). Classification and dynamics of developmental stages in late-successional temperate forests. Forest
Ecology and Management, 334, 344–357. https://doi.org/10.1016/j.foreco.2014.09.003
Luyssaert, S., Schulze, E.-D., Börner, A., Knohl, A., Hessenmöller, D., Law, B. E., etal. (2008). Old-growth forests as global carbon sinks.
Nature, 455(7210), 213–215. https://doi.org/10.1038/nature07276
Meigs, G. W., Morrissey, R. C., Bače, R., Chaskovskyy, O., Čada, V., Després, T., etal. (2017). More ways than one: Mixed-severity dis-
turbance regimes foster structural complexity via multiple developmental pathways. Forest Ecology and Management, 406, 410–426.
https://doi.org/10.1016/j.foreco.2017.07.051
Mikoláš, M., Ujházy, K., Jasík, M., Wiezik, M., Gallay, I., Polák, P., et al. (2019). Primary forest distribution and representation
in a Central European landscape: Results of a large-scale field-based census. Forest Ecology and Management, 449, 117466.
https://doi.org/10.1016/j.foreco.2019.117466
PETTIT ET AL.
10.1029/2020JD033929
16 of 17

Journal of Geophysical Research: Atmospheres
Mitchell, S. J. (1995). The windthrow triangle: A relative windthrow hazard assessment procedure for forest managers. The Forestry Chron-
icle, 71(4), 446–450. https://doi.org/10.5558/tfc71446-4
Mitchell, S. J. (2013). Wind as a natural disturbance agent in forests: A synthesis. Forestry, 86(2), 147–157. https://doi.org/10.1093/forestry/
cps058
Müller, J., Noss, R. F., Thorn, S., Bässler, C., Leverkus, A. B., & Lindenmayer, D. (2019). Increasing disturbance demands new policies to
conserve intact forest. Conservation Letters, 12(1), e12449. https://doi.org/10.1111/conl.12449
Nagel, T. A., Mikac, S., Dolinar, M., Klopcic, M., Keren, S., Svoboda, M., etal. (2017). The natural disturbance regime in forests of the
Dinaric Mountains: A synthesis of evidence. Forest Ecology and Management, 388, 29–42. https://doi.org/10.1016/j.foreco.2016.07.047
Nagel, T. A., Svoboda, M., & Diaci, J. (2006). Regeneration patterns after intermediate wind disturbance in an old-growth Fagus-Abies
forest in southeastern Slovenia. Forest Ecology and Management, 226(1–3), 268–278. https://doi.org/10.1016/j.foreco.2006.01.039
Nicoll, B. C., Gardiner, B. A., & Peace, A. J. (2008). Improvements in anchorage provided by the acclimation of forest trees to wind stress.
Forestry, 81(3), 389–398. https://doi.org/10.1093/forestry/cpn021
Peterson, C. J., & Pickett, S. T. A. (1991). Treefall and resprouting following catastrophic windthrow in an old-growth hemlock-hardwoods
forest. Forest Ecology and Management, 42(3–4), 205–217. https://doi.org/10.1016/0378-1127(91)90025-q
Pettit, J. (2020). climdf.csv. https://doi.org/10.6084/m9.figshare.12983003.v1
Quine, C., & White, I. M. S. (1998). The potential of distance-limited topex in the prediction of site windiness. Forestry, 71(4), 325–332.
https://doi.org/10.1093/forestry/71.4.325
Quine, & Gardiner, B. A. (2007). Understanding how the interaction of wind and trees results in windthrow, stem breakage, and canopy
gap formation. In Plant Disturbance Ecology (pp. 103–155). Elsevier. https://doi.org/10.1016/B978-012088778-1/50006-6
Riemann-Campe, K., Fraedrich, K., & Lunkeit, F. (2009). Global climatology of convective available potential energy (CAPE) and con-
vective inhibition (CIN) in ERA-40 reanalysis. Atmospheric Research, 93(1–3), 534–545. https://doi.org/10.1016/j.atmosres.2008.09.037
Ruel, J.-C., Mitchell, S. J., & Dornier, M. (2002). A GIS based approach to map wind exposure for windthrow hazard rating. Northern Jour-
nal of Applied Forestry, 19(4), 183–187. https://doi.org/10.1093/njaf/19.4.183
Sabatini, F. M., Burrascano, S., Keeton, W. S., Levers, C., Lindner, M., Pötzschner, F., etal. (2018). Where are Europe's last primary forests?
Diversity and Distributions, 24(10), 1426–1439. https://doi.org/10.1111/ddi.12778
Schelhaas, M.-J. (2008). Impacts of natural disturbances on the development of European forest resources: Application of model approach-
es from tree and stand levels to large-scale scenarios. Dissertationes Forestales, 2008(56). https://doi.org/10.14214/df.56
Schelhaas, M.-J., Nabuurs, G.-J., & Schuck, A. (2003). Natural disturbances in the European forests in the 19th and 20th centuries. Global
Change Biology, 9(11), 1620–1633. https://doi.org/10.1046/j.1365-2486.2003.00684.x
Schemm, S., Sprenger, M., Martius, O., Wernli, H., & Zimmer, M. (2017). Increase in the number of extremely strong fronts over Europe? A
study based on ERA-Interim reanalysis (1979-2014). Geophysical Research Letters, 44(1), 553–561. https://doi.org/10.1002/2016gl071451
Schmidt, M., Hanewinkel, M., Kändler, G., Kublin, E., & Kohnle, U. (2010). An inventory-based approach for modeling single-tree storm
damage - experiences with the winter storm of 1999 in southwestern Germany. Canadian Journal of Forestry and Research, 40(8),
1636–1652. https://doi.org/10.1139/x10-099
Schurman, J. S., Trotsiuk, V., Bače, R., Čada, V., Fraver, S., Janda, P., etal. (2018). Large-scale disturbance legacies and the climate sensitiv-
ity of primary Picea abies forests. Global Change Biology, 24(5), 2169–2181. https://doi.org/10.1111/gcb.14041
Seedre, M., Janda, P., Trotsiuk, V., Hedwall, P.-O., Morrissey, R. C., Mikoláš, M., etal. (2020). Biomass carbon accumulation patterns
throughout stand development in primary uneven-aged forest driven by mixed-severity natural disturbances. Forest Ecology and Man-
agement, 455, 117676. https://doi.org/10.1016/j.foreco.2019.117676
Seidl, R., Thom, D., Kautz, M., Martin-Benito, D., Peltoniemi, M., Vacchiano, G., etal. (2017). Forest disturbances under climate change.
Nature Climate Change, 7(6), 395–402. https://doi.org/10.1038/nclimate3303
Senf, C., & Seidl, R. (2018). Natural disturbances are spatially diverse but temporally synchronized across temperate forest landscapes in
Europe. Global Change Biology, 24(3), 1201–1211. https://doi.org/10.1111/gcb.13897
Sepp, M., Post, P., & Jaagus, J. (2005). Long-term changes in the frequency of cyclones and their trajectories in Central and Northern Eu-
rope. Hydrology Research, 36(4–5), 297–309. https://doi.org/10.2166/nh.2005.0023
Siedlecki, M. (2009). Selected instability indices in Europe. Theoretical and Applied Climatology, 96(1–2), 85–94. https://doi.org/10.1007/
s00704-008-0034-4
Slivinski, L. C., Compo, G. P., Whitaker, J. S., Sardeshmukh, P. D., Giese, B. S., McColl, C., etal. (2019). Toward a more reliable historical
reanalysis: Improvements for version 3 of the twentieth century reanalysis system. Quarterly Journal of the Royal Meteorological Society,
145(724), 2876–2908. https://doi.org/10.1002/qj.3598
Sommerfeld, A., Senf, C., Buma, B., D'Amato, A. W., Després, T., Díaz-Hormazábal, I., etal. (2018). Patterns and drivers of recent distur-
bances across the temperate forest biome. Nature Communications, 9(1). https://doi.org/10.1038/s41467-018-06788-9
Synek, M., Janda, P., Mikoláš, M., Nagel, T. A., Schurman, J. S., Pettit, J. L., etal. (2020). Contrasting patterns of natural mortality in primary
Picea forests of the Carpathian mountains. Forest Ecology and Management, 457, 117734. https://doi.org/10.1016/j.foreco.2019.117734
Taszarek, M., Allen, J., Púčik, T., Groenemeijer, P., Czernecki, B., Kolendowicz, L., etal. (2019). A climatology of thunderstorms across
Europe from a synthesis of multiple data sources. Journal of Climate, 32(6), 1813–1837. https://doi.org/10.1175/jcli-d-18-0372.1
Tepley, A. J., Swanson, F. J., & Spies, T. A. (2013). Fire-mediated pathways of stand development in Douglas-fir/western hemlock forests of
the Pacific Northwest, USA. Ecology, 94(8), 1729–1743. https://doi.org/10.1890/12-1506.1
Trotsiuk, V., Svoboda, M., Janda, P., Mikolas, M., Bace, R., Rejzek, J., etal. (2014). A mixed severity disturbance regime in the primary
Picea abies (L.) Karst. forests of the Ukrainian Carpathians. Forest Ecology and Management, 334, 144–153. https://doi.org/10.1016/j.
foreco.2014.09.005
Usbeck, T., Wohlgemuth, T., Pfister, C., Volz, R., Beniston, M., & Dobbertin, M. (2010). Wind speed measurements and forest damage in
Canton Zurich (Central Europe) from 1891 to winter 2007. International Journal of Climatology. https://doi.org/10.1002/joc.1895
Zielonka, T., Holeksa, J., Fleischer, P., & Kapusta, P. (2010). A tree-ring reconstruction of wind disturbances in a forest of the Slovakian
Tatra Mountains, Western Carpathians. Journal of Vegetation Science, 21(1), 31–42. https://doi.org/10.1111/j.1654-1103.2009.01121.x
PETTIT ET AL.
10.1029/2020JD033929
17 of 17