The phenology of the subnivium

Abstract

The subnivium is a seasonal refuge that exists at the interface between the snowpack and the ground, and provides a haven for a diversity of species to survive extreme winter temperatures. Due to the fitness of many plants and animals being strongly influenced by winter conditions, much attention has been given to changes in the timing of snow cover extent and duration in seasonally snow-covered environments; however, these broad-scale characteristics do not capture the finer-scale dynamics of the subnivium. To study the factors associated with subnivium development, we quantified three critical phenophases of the subnivium: establishment, maintenance, and disintegration along a latitudinal and land cover gradient in the Great Lakes Region of North America. We hypothesized that subnivium phenophases would depend primarily on snow depth and air temperature, but that these would be mediated by latitude and land cover. We found that patterns in both establishment and disintegration were affected by latitude more than land cover, but that variability in the timing of early season snowfall events overrode the effects of both factors in subnivium establishment. In contrast, disintegration was predictably later in more northerly sites, regardless of interannual variation in weather patterns. We found that the subnivium was the result of a balance between ambient temperature, snow depth, and snow density, but that ambient temperatures constrained the system by contributing to the frequency of snowfall and inducing changes in snow depth and density. Areas in lake effect zones, characterized by high snow depths and persistent snow cover, may be the last refugia for subnivia-dependent species given the predicted shifting climate regimes of the 21st century.

Full text

Environ. Res. Lett. 13 (2018) 064037 https://doi.org/10.1088/1748-9326/aac670
LETTER
The phenology of the subnivium
Kimberly L Thompson1,3, Benjamin Zuckerberg1, Warren P Porter2and Jonathan N Pauli1
1Department of Forest and Wildlife Ecology, University of Wisconsin-Madison, 1630 Linden Drive, Madison, WI 53706, United States of
America
2Department of Integrative Biology, University of Wisconsin-Madison, 250 N. Mills Street, Madison, WI 53706, United States of America
3Author to whom any correspondence should be addressed.
OPEN ACCESS
RECEIVED
20 March 2018
REVISED
14 May 2018
ACCEPTED FOR PUBLICATION
21 May 2018
PUBLISHED
18 June 2018
Original content from
this work may be used
under the terms of t he
Creative Commons
Attribution 3. 0 licence.
Any further distribution
of this work must
maintain attrib ution to
the author(s) and the
title of the work, journal
citation and DOI.
E-mail: kthompson25@wisc.edu
Keywords: snow, climate change, winter climate, phenology, great lakes region
Supplementary material for this article is available online
Abstract
The subnivium is a seasonal refuge that exists at the interface between the snowpack and the ground,
and provides a haven for a diversity of species to survive extreme winter temperatures. Due to the
fitness of many plants and animals being strongly influenced by winter conditions, much attention
has been given to changes in the timing of snow cover extent and duration in seasonally snow-covered
environments; however, these broad-scale characteristics do not capture the finer-scale dynamics of
the subnivium. To study the factors associated with subnivium development, we quantified three
critical phenophases of the subnivium: establishment, maintenance, and disintegration along a
latitudinal and land cover gradient in the Great Lakes Region of North America. We hypothesized that
subnivium phenophases would depend primarily on snow depth and air temperature, but that these
would be mediated by latitude and land cover. We found that patterns in both establishment and
disintegration were affected by latitude more than land cover, but that variability in the timing of early
season snowfall events overrode the effects of both factors in subnivium establishment. In contrast,
disintegration was predictably later in more northerly sites, regardless of interannual variation in
weather patterns. We found that the subnivium was the result of a balance between ambient
temperature, snow depth, and snow density, but that ambient temperatures constrained the system by
contributing to the frequency of snowfall and inducing changes in snow depth and density. Areas in
lake effect zones, characterized by high snow depths and persistent snow cover, may be the last refugia
for subnivia-dependent species given the predicted shifting climate regimes of the 21st century.
Introduction
Snow covers roughly 24 million square kilometers of
the Northern Hemisphere annually, but its ephemeral
nature causes wide fluctuations in extent and per-
sistence (Lemke et al 2007). Snow cover extent has
declined in the Northern Hemisphere over the past 90
years, with the most significant reductions beginning
in the 1980s (Vaughn et al 2013). Since this time, snow
onset has advanced slightly, but the most significant
changes have been in snowmelt date, which has shifted
earlier by approximately two days per decade (Chen
et al 2016). These changes exhibit wide spatial het-
erogeneity, however, as the timing of snowmelt has
remained relatively stable in North America, but
advanced drastically in Eurasia (Peng et al 2013).
Although much attention has been given to these
broad-scale changes in snow cover (Chen et al 2016,
Wang et al 2017), they do not shed light on the condi-
tions of below-the-snow environments.
The subnivium is the seasonal refuge that exists at
the interface between the snowpack and the ground
(Pauli et al 2013), which provides a haven for a
diversity of plants (Bjork and Molau 2007), mammals
(Korslund and Steen 2006), amphibians (O’Connor
and Rittenhouse 2016), birds (Whitaker and Stauf-
fer 2003), and arthropods (Hagvar 2010) to survive
extreme winter temperatures. The formation of the
subnivium results from continual sublimation and
condensation occurring within a snowpack and the
upward migration of water vapor from areas of high
vapor density (closest to the ground) to low vapor den-
sity (near the snow surface) (Marchand 2013,Pinzer
and Schneebeli 2009). This movement of vapor reduces
© 2018 The Author(s). Published by IOP Publishing Ltd

Environ. Res. Lett. 13 (2018) 064037
the size of the ice crystals in the bottommost snow
layer, creating a network of loosely-connected crystals
whose low density traps heat released from the soil
(Marchand 2013). When snow depths are sufficiently
high, the low thermal conductivity of the snowpack
insulates the subnivium, creating a warmer and more
stable microclimate compared to external air tempera-
tures (Pruitt 2005).
While climate limits the distribution of the sub-
nivium at regional scales, other factors such as
vegetation, microtopography, and wind can influence
the formation of the subnivium at local scales (Heffer-
nan et al 2014, Petty et al 2015). Snow accumulation
and density, both of which directly control the for-
mation and persistence of the subnivium, are affected
by ambient temperature, wind, snowfall, and radiation
fluxes (Harestad and Bunnell 1981). Deep, low density
snow is most effective in maintaining the thermal st abil-
ity of the subnivium (Ge and Gong 2010). With warmer
ambient temperatures, ablation increases, reducing
depth and increasing snow density throughout the
entire snowpack (Zhang et al 2016). Under colder
conditions, the temperature gradient between the bot-
tom layer of snow and the air temperature increases,
which increases snow density at the surface of the
snowpack (Kim and Jamieson 2014, Pfeffer and Mru-
gala 2002). Despite increased density at the surface,
air temperatures at or below 0 ◦C prevent the occur-
rence of surface snow melt and support the retention
of snow depth (Marchand 2013). Wind influences snow
depth and density by reducing snow accumulation
through erosion and increased sublimation, but can
also increase accumulation and compaction through
redistribution (e.g. snowdrifts) (Gascoin et al 2013,
Pomeroy and Goodison 1997). Finally, the energy bal-
ance between net long-wave and incoming short-wave
radiation influences snow characteristics as absorption
of radiation raises snow temperatures and increases
melt rates (Essery et al 2008,Varholaet al 2010).
Many local-scale factors influencing snow cover
characteristics are modulated by land cover. Canopy
cover intercepts up to 40%–60% of snowfall (Hedstrom
and Pomeroy 1998, Pomeroy et al 1998). Intercepted
snowfall is exposed to wind, unprotected from solar
radiation, and subsequently vulnerable to sublima-
tion back to the atmosphere (Martin et al 2013). The
effects of forest cover on accumulation are particularly
important in coniferous forests, where snow can be
retained in the canopy for days to months (Pomeroy
et al 2002, Pomeroy and Schmidt 1993). Although
interception may reduce snow depth, forest cover also
promotes snow retention once a snowpack has been
developed by providing a buffer against wind and
blocking the snow from incoming shortwave radia-
tion (Essery et al 2008, Lundquist et al 2013,Varhola
et al 2010). Since the amount of longwave radiation
emitted by forests generally does not exceed incoming
shortwave radiation (Essery et al 2008), the thermal
protections provided by trees can promote longer-
lasting snow conditions.
We investigated the phenology of the subnivium
across three different cover types (coniferous forests,
deciduous forests, and open sites) and three latitudi-
nal ranges in the North American Great Lakes Region.
While previous work on the subnivium has focused
on the role of threshold depth (Pruitt 2005) and den-
sity (Marchand 2013), the life cycle of this ephemeral
habitat has not been fully explored. Our goal was to
quantify, for the first time, three critical phenophases of
the subnivium: establishment, maintenance, and disin-
tegration, and to evaluate how latitude and land cover
mediate these phenophases. We predicted that sub-
nivium conditions would depend primarily on snow
depth and air temperature, with delayed establishment
in coniferous sites due to increased canopy intercep-
tion of snowfall, and in more southerly latitudes due to
less snowfall. Once established, however, we expected
that the duration of the subnivium would be longest
in northerly, forested sites due to later snowmelt dates.
Conversely, the duration of the subnivium would be
reduced in southerly open sites characterized by shal-
lower snow depths and greater exposure to shortwave
radiation. Similarly, we hypothesized that the thermal
stability of the subnivium would be lowest in southerly
open areas, while northerly areas that have less ambi-
ent temperature variation during the winter season
would have the most stable subnivia. Understanding
the phenology of this seasonal refuge—the timing of
its establishment, maintenance, and disintegration—is
important for characterizing the current conditions to
which species are exposed, and for predicting future
exposure.
Methods
Study area
The climate of the Great Lakes Region is typified by
cold winters and warm-to-hot summers (Peel et al
2007), but varies according to latitude and proxim-
ity to the Great Lakes (Andresen et al 2014). The
greatest range of temperatures occurs from Decem-
ber to February and the coldest overall temperatures
occur in northern interior areas, away from the Great
Lakes (Andresen et al 2014). Areas in the lake effect
zone (i.e. downwind of the lakes) are typically wet-
ter with more moderate climates and large amounts
of snowfall (Burnett et al 2003, Changnon and Jones
1972,ScottandHuff1996).Tocapturetherangeof
variation in subnivium phenology across the northern
Great Lakes Region, we selected three latit udinal ranges
(lower latitude: 42◦–44◦, middle latitude: 44◦–46◦,
and upper latitude: 46◦+), and within each examined
three cover types (deciduous, coniferous, and open)
for a total of nine sites (figure 1, table S1 available at
stacks.iop.org/ERL/13/064037/mmedia).
2

Environ. Res. Lett. 13 (2018) 064037
Figure 1. Study sites and land cover in the Upper Midwest Great Lakes study area.
Data collection
During the winters of 2015–2016 and 2016–2017, we
deployed three temperature strings at each site, with
each string containing four individual temperature
probes for a total of 12 probes per site. Each probe
was separated by 0.3 meters, and we positioned strings
with roughly 5 meters spacing. We staked the strings
to ensure that the probes were flush with the ground.
Probes recorded ground temperatures at five-minute
intervals from November through April. As a result
of the microtopography at several sites, 25 individual
probes in winter 2015/16 and four probes in win-
ter 2016/17 became encased in ice or submerged in
standing water during all or part of the winter season
and provided faulty temperature data; data from these
probes were excluded from analyses.
We established three weather stations at each site
to which we affixed a heated rain gauge for measuring
the amount of liquid precipitation (Davis Instruments
Corp Rain Collector and Rain Collector Heater),
an anemometer for measuring wind speeds (Davis
Instruments Corp Anemometer, mounted at mean
height of 1.8 meters), and a temperature probe housed
in a radiation shield (Davis Instruments Corp External
Temperature Sensor, mounted at mean height of
1.6 meters) for measuring ambient temperature.
Each rain gauge possessed a wetness sensor (Davis
Instruments Corp Leaf Wetness Sensor) to identify
the beginning and end of individual precipitation
events with greater precision. Similar to the subnivium
temperature probes, all weather station instruments
recorded data from November through April, but
at one-minute intervals. Between winter 2015/16
and 2016/17, we also installed a snow depth sensor
(HRXL-Max Sonar WRS Series Ultrasonic Snow
Depth Sensor, typical accuracy of 1%) at each site
to obtain finer scale snow depth data, which was
measured at one-minute intervals from November
through April in winter 2016/17. For winter 2016/17,
we used our field-measured snow depth values; in
winter 2015/16, we used the midpoint of the daily snow
depth range from the National Operational Hydro-
logic Remote Sensing Center’s (NOHRSC) Interactive
Snow Information map (www.nohrsc.noaa.gov/
interactive/html/map.html). NOHRSC uses a
physically-based snow model to produce spatially
explicit estimates of daily snow depth, which are
reported as ranges of accumulation (e.g. trace to
5.1 cm, 5.1–10 cm, 10–20 cm, 20–30 cm, etc.). We
confirmed that depths obtained from NOHRSC were
3

Environ. Res. Lett. 13 (2018) 064037
Figure 2. Phenophases of th e subnivium applied to a temperature prob e at one of the forested st udy sites during the winter of 2016/17.
Subnivium establishment occurs on the day when daily temperature stabilizes (i.e. the slope of the line segment thereafter ≈0). The
subnivium is then maintained for the duration of time starting from the day of the first occurrence of a zero slope to the day of the last
occurrence of a zero slope. The subnivium disintegration date is the day following the last occurrence of a stable slope.
appropriate to use for winter 2015/16 by investigating
the correlations between our field-measured depth
data in winter 2016/17 and the midpoints of the daily
snow depth range modeled by NOHRSC for the same
time period (table S2). From these data we determined
the date of the first snow event that resulted in at least
10 cm of snow accumulation for each site in each
season, as well as a snow-off date. We obtained daily
snow density values for each site for both winters from
NOHRSC data to have an estimate that covered the
entire depth of the existing snowpack.
Establishment, maintenance, and disintegration of
the subnivium
We aggregated all ground temperature probe data to
daily mean ground temperatures for each probe and
fit piecewise linear models to the ground tempera-
tures of each probe (Zeileis et al 2003) to determine
the number of breakpoints for the entire winter sea-
son (Ryan and Porth 2007,Zeileiset al 2003,Zeileis
et al 2002). Breakpoints represent the day where the
slope of the modeled temperature changed signifi cantly
(Ryan and Porth 2007) and indicate shifts in the tem-
perature profile of the subnivium (Zeileis et al 2002).
For any number of breakpoints p,therearep+1seg-
ments in which the regression coefficient 𝛽is constant;
therefore, the segmented model takes the form
𝑦𝑖𝑘 =𝛽𝑗𝑘𝑥𝑖+𝑢𝑗𝑘(𝑖=𝑡initial ,…,𝑡
final,
𝑗=1,…,𝑝+1,𝑘=1,2,…,12)
where iis the day, jis the segment index, and kis
the probe identifier (Zeileis et al 2002). Thus, yik is
the subnivium temperature at day ifor probe k,x𝑖is
the date, 𝛽jk is the regression coefficient and ujk is the
intercept for the segment.
We trialed multiple piecewise regression models for
each probe by varying the number of breakpoints and
selected the model which minimized the total residual
sum of squares for each segment (∑𝑝+1
𝑗=1(𝑦ik −𝑦ik ) 2)
and had the lowest Bayesian Information Criterion
(BIC) value (Zeileis et al 2003). We used BIC because
it is appropriate for selection when describing all
possible models (i.e. breakpoints) (Zeileis et al 2003)
and the Akaike information criterion has been shown
to overestimate the number of breakpoints (Bai and
Perron 2003).
Once the number of breakpoints was identified,
we obtained the slopes for each segment and used
these to define three phenophases of the subnivium:
establishment, maintenance, and disintegration. We
considered the subnivium established on the day when
daily temperature stabilized (i.e. the slope demon-
strated stability for at least three days: 0.15 ≤slope
≤0.15), given that the first snowfall with accumu-
lation of at least 10 cm had occurred on or prior
to that day (figure 2). Although Pruitt (2005)iden-
tified 20 cm as the minimum depth required for
subnivium establishment, we found evidence for sta-
bilized temperatures at depths less than this amount
when analyzing ground temperatures at hourly time
scales. Therefore, we used a threshold depth of 10 cm
to include these shorter-term stabilizations and their
potential effects on subnivium establishment. Addi-
tionally, piecewise regression requires a minimum of
three observations, or in this case days, to fit an appro-
priate line segment.
4

Environ. Res. Lett. 13 (2018) 064037
We defined subnivium maintenance as the dura-
tion of time from the day of the first occurrence of
a stable slope to the day of the last occurrence of a
stable slope (figure 2). Since stable temperatures can
occur when snow is not present, we only considered
slope segments that fell between the day of the first
major snow event and the snow-off date to ensure
that we were capturing subnivium conditions. To
assess the stability of the subnivium, we calculated
the range between maximum and minimum ground
temperatures during the maintenance period. Finally,
we defined subnivium disintegration as the day that
directly followed the last stable slope segment (fig-
ure 2). We then created pseudo-Julian dates with
November 1st corresponding to 1, and represented each
phenophase as a continuous date variable.
The responses of establishment and disintegration
date, maintenance period, and subnivium temperature
range, were analyzed using generalized linear mixed
effects models (GLMM) with Gaussian distributions
and temperature string as a random effect (Bolker
et al 2009). All models were fit using the lme4 pack-
age in R (Bates et al 2015,RCoreTeam2015).
For maintenance period and subnivium temperature
range, we ran separate mixed models for each year of
data collection to account for interannual variability.
Predictors of subnivium phenology
Predictors for all responses (establishment and dis-
integration date, maintenance period, and subnivium
temperature range) included latitude, land cover, and
year. Covariates of minimum and maximum snow
depth (cm), minimum and maximum air tempera-
ture (◦C), mean snow density (g cm−3), and mean
wind speed (m s−1) were also included for mainte-
nance periods and subnivium temperature range. Prior
to inclusion in the models, we standardized all con-
tinuous predictors and checked for multicollinearity.
We sequentially removed covariates from inclusion
based on their variance inflation factors (VIF) until
all covariates had VIFs less than 4 (Zuur et al 2010).
For each response, we developed a set of candidate
models that included additive combinations of the
remaining explanatory variables. We included inter-
actions between continuous covariates that we had
hypothesized to be relevant, but did not test all combi-
nations to avoid overparameterization. For this reason,
and because we were missing one latitude and land
cover combination in Year 1 (upper latitude, decid-
uous), we did not include an interaction of latitude
and land cover in models for maintenance period or
subnivium temperature range. We assessed all mod-
els with the Akaike information criteria adjusted for
small sample size (AICc). We considered models that
were within 2 ΔAICcof the top model as competitive;
however, models that contained uninformative param-
eters were excluded from the model selection (Arnold
2010). We averaged predictions for responses that had
more than one competitive model based on AICc
weights (w𝑖), but avoided model averaging parameter
estimates given recent concerns (Cade 2015).
Results
Timing of subnivium establishment and disintegra-
tion
Subnivia established at 79 out of 83 temperature probes
in winter 2015/16 (Year 1) and 91 out of 104 probes
in winter 2016/17 (Year 2) in all cover types and lati-
tudes. To investigate the effects of latitude, cover type,
and interannual variability on subnivium establish-
ment and disintegration dates (represented as days
after November 1st), we tested a set of 13 mod-
els (tables S3 and S4). For sites and probes where
subnivia developed, the best model for establishment
date included an interaction between latitude and
winter season (w𝑖= 0.61) (table 1,figure3(a)). Estab-
lishment date did not vary between latitudes during
Year 2; however, during Year 1 subnivia established
approximately two weeks earlier at the northerly sites
(Year 1 Lower: 70.37 ±2.50, Middle: 56.88 ±2.17,
Upper: 46.53 ±2.48; Year 2 Lower: 43.53 ±2.27, Mid-
dle: 47.64 ±2.10, Upper: 47.32 ±2.24). Interannual
variability produced differences in the timing of estab-
lishment in lower and middle latitudes, but the early
establishment of subnivia in upper latitudes was con-
sistent across years (figure 3(a)).
Although we predicted that snow interception
from conifers would lead to later subnivium estab-
lishment, we found no differences in establishment
dates at conifer sites across latitudes (Conifer Lower:
55.93 ±3.44, Conifer Middle: 53.73 ±2.74, Conifer
Upper: 46.06 ±3.03, figure 3(b)). Similarly, com-
parisons of establishment in open and deciduous
sites across latitudes showed no differences (Open
Lower: 45.38 ±3.57, Middle: 52.75 ±2.62, Upper:
45.48 ±2.68; Deciduous Lower: 61.5 ±2.62, Middle:
49.45 ±2.88, Upper: 50.17 ±3.71, figure 3(b)). In gen-
eral, the establishment of the subnivium was similar
between cover types of similar latitudes, except for
lower latitude deciduous and open sites. In our most
southerly sites, deciduous areas had later establishment
dates than open sites, yet the establishment dates for
coniferous sites overlapped with both deciduous and
open areas (Open: 45.38 ±3.57, Conifer: 55.93 ±3.44,
Deciduous: 61.50 ±2.62, figure 3(b)).
The best model for subnivium disintegration date
included an interaction between latitude and season
and between latitude and cover type (w𝑖= 0.99) (table
1,figures3(c)and(d)). Disintegration date did not
exhibit as much interannual variation as establish-
ment date and was later in upper latitudes (Year 1
Lower: 106.36 ±2.53, Middle: 124.94 ±2.06, Upper:
144.75 ±2.71; Year 2 Lower: 79.17 ±2.20, Middle:
115.64 ±1.98, Upper: 123.76 ±2.18, figure 3(c)). Dis-
integration dates in middle latitudes were consistent
across years and cover types, while in lower latitudes
5

Environ. Res. Lett. 13 (2018) 064037
Table 1. Model selection results for each subnivium response variable measured during the winters of 2015/2016 (Year 1) and 2016/17 (Year 2) in the Upper Midwest. Best-supported model and models within two ΔAICcare shown.
Also provided are the number of model parameters (k) and AICcweights (wi), which indicate the most likely model. Cov=Land cover type, Lat=Latitudinal Range, Tair ,min=Minimum Air Temperature, Tair,max=Maximum Air
Temperature, Densitymean =Mean Snow Density, Depthmin =Minimum Snow Depth, Depthmax=Maximum Snow Depth, and Windmean =Mean Wind Speed. For marginal and conditional R2values see table S7.
Year Response Model k AICcΔAICcwiLog-likelihood
Pooled Establishment Date Year∗Lat 8 1304.12 0.00 0.61 −643.61
Year∗Lat +Year∗Cov 12 1305.09 0.96 0.37 −639.55
Pooled Disintegration Date Year∗Lat +Cov∗Lat 14 1333.54 0.00 0.99 −651.41
1 Maintenance Period Lat +Tair,min ∗Depthmax +Depthmax∗Densitymean +Depthmin +Windmean 11 536.82 0.00 0.40 −255.44
Cov +Tair,min ∗Depthmax +Depthmax∗Densitym ean +Depthmin +Windmean 11 537.19 0.37 0.33 −255.62
Cov +Lat +Tair,min ∗Depthmax +Depthmax∗Densitymean +Windmean 13 537.73 0.92 0.25 −253.07
2 Maintenance Period Cov +Tair,min ∗Depthmax +Depthmax ∗Densitymean +Windmean 12 668.51 0.00 0.85 −320.25
1 Temperature Range Cov +Lat +Tair,min +Depthmin 9 327.22 0.00 0.48 −153.31
Lat +Tair,min ∗Depthmin +Windmean +Densitymean 10 329.02 1.79 0.20 −152.89
2 Temperature Range Lat +Depthmin ∗Densitymean +Tair ,min +Windmean 10 364.39 0.00 0.64 −170.82
Lat +Tair,min ∗Depthmin +Windmean +Densitymean 10 366.21 1.82 0.26 −171.73
6

Environ. Res. Lett. 13 (2018) 064037
Figure 3. Subnivium establishment and disintegration dates and 95% confidence intervals based on data collected in the Upper
Midwest during winter 2015/16 (Year 1) and winter 2016/17 (Year 2). Pseudo-julian date is shown on the left y-axis (November 1st =
1) with corresponding dates on the right y-axis. (a) Relative to the lower latitudes (42◦–44◦), establishment occurred approximately
one month earlier in upper latitudes (46◦+) and 2 weeks earlier in middle latitudes (44◦–46◦), during Year 1. Establishment dates for
Year 2 were more similar across latitudes, demonstrating the importance of interannual variation in weather patterns for subnivium
establishment. (b) Establishment date did not vary by cover type between latitudinal ranges. (c) Disintegration became later as both
latitude increased (lower latitude: 42◦–44◦, middle latitude: 44◦–46◦, and upper latitude: 46◦+), and (d)thisresponsewassimilar
across land cover types (open, deciduous, and conifer), with the exception of upper latitude conifer areas. In these areas subnivia
disintegrated significantly earlier than in upper latitude deciduous and open areas.
disintegration varied by about a month between
years. The disintegration of subnivia in the northerly
sites displayed variability in both years and cover
types, with a three-week difference in timing between
years and a two-month difference between cover
types. Upper-latitude deciduous sites had the latest
disintegration dates (157.25 ±3.72), while subnivia dis-
integrated two weeks earlier in upper-latitude open
sites (140.52 ±2.51) and two months earlier in upper-
latitude conifer sites (101.10 ±2.87, figure 3(d)).
Duration of subnivium maintenance
Out of 22 candidate models (table S5), the model that
best explained maintenance period in Year 1 included
latitude, an interaction between minimum air tem-
perature and maximum snow depth, an interaction
between maximum snow depth and mean snow den-
sity, minimum snow depth, and mean wind speed
(w𝑖= 0.40) (table 1, figure S1). The top model for Year
2 included cover type, an interaction between min-
imum air temperature and maximum snow depth,
an interaction between maximum snow depth and
mean snow density, and mean wind speed (w𝑖= 0.85)
(table 1,figureS1).
Maintenance periods were similar throughout
deciduous and coniferous sites in Years 1 and 2
(Year 1 Deciduous: 53.19 ±2.72, Conifer: 62.61 ±2.43;
Year 2 Deciduous: 78.51 ±4.73, Conifer: 69.67 ±4.39),
whereas open sites differed between years (Year 1:
69.62 ±3.37, Year 2: 57.89 ±8.07, figure S3). Lati-
tude was only present in the top model for Year
1, but as expected due to higher levels of snowfall,
middle and upper latitudes had longer maintenance
periods than lower latitudes (Lower: 52.10 ±2.56,
Middle: 69.57 ±3.43, Upper: 78.80 ±5.40). Similarly,
higher snow depths (cm) supported longer mainte-
nance periods (Year 1: 6.49 ±2.99, Year 2: 47.24 ±3.96,
figure 4(a)). Higher minimum temperatures (◦C) pro-
duced shorter maintenance periods in both years (Year
1: −2.77 ±0.82, Year 2: −16.04 ±3.12, figure 4(b)).
Higher snow density (g cm−3) also resulted in shorter
maintenance periods, although the magnitude of this
7

Environ. Res. Lett. 13 (2018) 064037
Figure 4. Predictions of subnivium maintenance period for a range of (a) maximum snow depths and (b) minimum air t emperatures.
As maximum snow depth increases, the maintenance period also increases. As minimum air temperature increases, maintenance
period decreases. The magnitude of this effect is much stronger in year 2.
effect was much stronger in the second year (Year 1:
−2.53 ±2.69, Year 2: −18.86 ±3.51).
The importance of minimum air temperature,
maximum snow depth, and mean snow densit y became
clearer when investigating their interactions. Warmer
air temperatures produced shorter maintenance peri-
ods, but this effect was lessened by higher snow depths,
and was most extreme at colder air temperatures
(figure 5(a)). At minimum air temperatures of −30 ◦C,
a wide range of snow depths were able to maintain
subnivia, but the length of this maintenance period
fluctuated by depth from around 75 days at low depths
of 5–15 cm to around 150 days at high depths of
80–90 cm. The differences in maintenance periods
produced by depth were most extreme at low air tem-
peratures, but as temperature increased the variation
between maintenance periods produced by differing
depths decreased.
Depth and temperature alone did not sufficiently
describe the conditions necessary for subnivium main-
tenance because their effect was mediated by snow
density. Although the interaction of temperature and
depth showed that longer maintenance periods were
possible with higher snow depths and colder air tem-
peratures, high snow densities counteracted this effect
(figure 5(b)). At low snow depths, all density val-
ues had similar predicted maintenance periods of
around 30 days, but as snow depth increased, main-
tenance periods rapidly increased for snowpacks with
low densities (0.10–0.20 g cm−3 )(figure5(b)). At
intermediate densities (0.22–0.26 g cm−3), the increase
in maintenance periods under increasing snow depths
reduced in magnitude, and was almost flat at a density
of 0.26 g cm−3.Finally,highsnowdensities(0.28–
0.30 g cm−3) overrode the positive effect of snow depth
on subnivium duration (figure 5(b)).
Subnivium temperature range
Out of 15 possible models (table S6), two candidate
models of subnivium temperature range emerged for
each year. The top model for Year 1 included cover
type, latitude, minimum air temperature and mini-
mum snow depth (w𝑖= 0.48), while the second-best
model included latitude, an interaction between mini-
mum air temperature and minimum depth, mean wind
speed, and mean density (w𝑖= 0.20) (table 1,figure
S2). The top model for Year 2 included latitude, an
interaction between minimum snow depth and mean
snow density, minimum air temperature, and mean
wind speed (w𝑖= 0.64), while the second-best model
included latitude, an interaction between minimum air
temperature and minimum snow depth, mean wind
speed, and mean density (w𝑖= 0.26) (table 1,figure
S2).
Subnivium temperature range represented a mea-
sure of stability, with smaller ranges corresponding to
more stable subnivia. Consequently, since subnivia in
middle latitudes had the smallest temperature ranges
in both years, they could be considered the most stable
(Year 1: 3.96 ±0.78, Year 2: 2.98 ±0.82). Temperatures
in lower latitude subnivia were the least stable (Year
1: 7.01 ±0.64, Year 2: 5.74 ±0.56). In Year 1, upper
latitudes displayed ranges similar to lower latitudes
(6.24 ±0.83), while in Year 2 upper latitude ranges
were similar to middle latitudes (3.35 ±0.64). Thus,
while subnivia in upper latitudes were stable and com-
parable to those in middle latitudes in Year 2, overall
bothupperandlowerlatitudesweremoresusceptible
to interannual variability in factors such as snow depth,
snow density, and air temperature.
The predictors of subnivium temperature stabil-
ity were similar between years. Increases in minimum
air temperature, mean wind speed, and minimum
8

Environ. Res. Lett. 13 (2018) 064037
Figure 5. Predictions of subnivium maintenance period highlighting the interaction between (a) minimum air temperature and
maximum snow depth and (b) maximum snow depth and mean density. As minimum temperature increases, maintenance periods
decrease; however the rate of this decline varies according to depth. As maximum snow depth increases maintenance periods increase
drastically at low densities (0.10–0.22 g cm−3), increase slightly at intermediate densities (0.22–0.26 g cm−3), and decrease at high
densities (>0.26 g cm−3).
snow depth all resulted in smaller temperature vari-
ability (Temperature: Year 1: −1.04 ±0.31, Year 2:
−0.23 ±0.25; Wind: Year 1: −1.92 ±0.56, Year 2:
−1.18 ±0.25; Depth: Year 1: −0.63 ±0.24, Year 2:
−0.27 ±0.22). Colder air temperatures increased tem-
perature variability in the subnivium, but since snow
depth insulated the subnivium it minimized the effect
of ambient temperature (figure 6(a)). Density fur-
ther mediated the magnitude of this effect; higher
densities (0.26–0.30 g cm−3) produced increases in
temperature variability even with increasing snow
depth (figure 6(b)).
Discussion
In describing three important phenophases of
the subnivium—establishment, maintenance, and
disintegration—we found that while latitude, land
cover, and interannual variability influenced sub-
nivium phenology, optimal subnivium conditions
depended on the relationship between air tempera-
ture, snow depth, and snow density. Although the
relative impact of these three factors varied across
the different subnivium phenophases, air tempera-
ture was a universal constraint through its effect on
snowpack depth and density, and varied according to
phenophase. Above-freezing temperatures promoted
rapid subnivium establishment, while below-freezing
temperatures reduced snowmelt and supported longer
subnivium maintenance periods. Concurrently, cold
air temperatures increased temperature variability in
the subnivium during the maintenance phase. Thus,
the creation of conditions for a stable and longer
subnivium period required that air temperatures fell
within a certain range.
Establishment and disintegration date
In subnivium establishment, interannual variability
was most important in lower and middle latitudes;
whereas, in upper latitudes establishment date was
similar between years. While the timing of establish-
ment progressed gradually in winter 2015/16 (Year 1),
a region-wide snow storm occurred in winter 2016/17
(Year 2), causing spatial synchronization of establish-
ment dates acrossall latitudes. Additionally, contrary to
our expectations, there was no difference in establish-
ment between cover types within or between latitudes.
This suggests that latitude, as a proxy for the amount
of snowfall an area typically receives, plays a larger role
than cover type in subnivium establishment, but that
variability in early season snow storms can override the
onset of subnivium conditions.
Disintegration was not as prone to variability in
weather patterns and was more synchronized between
years and cover types within the same latitudinal range.
As predicted, subnivium disintegration exhibited a
gradual pattern of later disintegration at more nort herly
sites. Middle latitudes exhibited the highest consistency
in disintegration dates, which unlike lower and upper
latitudes, fell within a similar range when comparing
across years and cover types. This suggests that con-
ditions in middle latitudes during the maintenance
period leading up to disintegration were such that
neither interannual variability nor differences among
cover types were sufficient to override the subnivium
conditions created by the balance of air temperature,
snow depth, and snow density.
Duration of subnivium maintenance
We predicted that subnivium conditions would
depend primarily on snow depth and air tempera-
ture. While this was partially supported, we found
9

Environ. Res. Lett. 13 (2018) 064037
Figure6. Predictions of subnivium temperature rangehighlighting the interaction between(a) minimum air temperature and minimum
snow depth and (b) minimum snow depth and mean density. As minimum air temperature increases, subnivium temperature range
decreases, which means that the subnivium becomes more stable. The most stable subnivia (i.e. temperature range = 0 ◦C) occur
when the minimum air temperature is at or greater than −20 ◦C. As minimum snow depth increases, the variability in subnivium
temperatures decreases at most densities. Higher densities promote an increase in temperature variability at higher snow depths;
however, this increase is relatively small in magnitude.
that air temperature provided the overarching con-
straint on subnivium existence, and each subnivium
phenophase resulted from a balance between air tem-
perature, snow depth, and snow density. Consequently,
describing the subnivium in terms of a threshold of
one of these factors alone is insufficient due to their
interdependency.
Colder air temperatures were required to promote
subnivium maintenance. Warmer air temperatures
increase ablation, which reduces depth, increases
density, and reduces the overall insulative capacity of
snow cover (Ge and Gong 2010). Warmer air tempera-
tures led to shorter subnivium maintenance periods,
but this effect was reduced at higher snow depths
(figure 5(a)). Since higher snow depths mitigated
the effect of warmer air temperatures, they promoted
longer maintenance periods, but this outcome was off-
set by high snow density (figure 5(b)). The potential
for snow density to counteract the effect of snow depth
became stronger in the later part of the winter season
since snow density increases over time due to changes
in the snowpack induced by air temperature, like melt-
ing and refreezing (Judson and Doesken 2000). Our
results show that at high densities (e.g. >0.25 g cm−3)
maintenance periods decreased even with increasing
snow depth, reducing subnivium maintenance from
approximately 30 days at shallow depths (5–25 cm) to
10–15 days at midrange depths (25–45 cm, figure 5(b)).
In fact, the range of snow densities at our surveyed
sites (0.10–0.30 g cm−3) represents a low to moder-
ate range compared to those found in other studies
at similar latitudes (Dawson et al 2017, Derksen et al
2014).
Subnivium stability
We measured thermal stability during the maintenance
period by quantifying overall temperature variabil-
ity. We found that warmer air temperatures reduced
subnivium temperature variability by reducing the gra-
dient between the bottom layer of snow and the air;
however, without sufficient snow depth, the subnivium
lacked insulation and was vulnerable to swings in air
temperature outside of the optimal range for minimiz-
ing thermal variability (figure 6(a)).
By exploring differences in subnivium tempera-
ture ranges between latitudes, we found that middle
latitude sites (44◦–46◦) had optimal ranges of air tem-
perature, snow depth, and density relative to upper
and lower latitudes. Subnivia in these sites had the
narrowest temperature range and this result was con-
sistent in both years, demonstrating that conditions in
these areas buffered the subnivia from the effects of
interannual variability and differences in land cover.
It is, therefore, possible that sites with more sta-
ble subnivium temperatures represented the trailing
range boundary of optimal conditions for subnivium
maintenance; however, it is likely this range bound-
ary will shift northward given the projected northward
shift from snow- to rain-dominated regions (Ning
and Bradley 2015).
Caveats
Although our study represents a novel exploration of
the subnivium and its phenology, we recognize sev-
eral potential limitations. First, given the number of
parameters in our models of maintenance period and
subnivium stability, we were unable to model explicit
10

Environ. Res. Lett. 13 (2018) 064037
interactions between latitude and land cover due to
issues of model overparameterization. For example,
collapsing all sites of a particular cover type throughout
our entire latitudinal gradient (42◦−46◦+) into a sin-
gle unit avoided problems with model convergence.
Second,wewereunabletotakedirectmeasuresof
solar radiation and were not able to explicitly account
for this variable, however, many of the abiotic con-
ditions we captured (e.g. ambient temperature, snow
density) change in tandem with solar radiation and
likely account for these effects.
Biological implications
Understanding the phenology of this critical, yet
understudied, seasonal refuge is crucial for predict-
ing how global change will affect the subnivium
and overwintering species that depend on the sub-
nivium. Although overwintering species have different
physiological tolerances and coping strategies, longer
maintenance periods with consistent and predictable
temperature stability will provide thermal protec-
tion and fitness benefits to a wide diversity of
subnivium-dependent species, while also sustaining
other ecological processes. For example, the subnivium
ecosystem influences patterns and frequencies of soil
freeze-thaw cycles (Groffman et al 2001), which in
turn regulate soil biogeochemistry (Kreyling et al
2012). Disturbances to the subnivium in either extent,
duration, or thermal stability can therefore disrupt
these prevailing regimes and result in phenological
mismatches (Wheeler et al 2015), leading to root
damage in plants (Sanders-DeMott et al 2017)and
microbial mortality (Schimel et al 2004). Subnivium
disturbances also produce phenological mismatches
in wildlife, with decreased survival rates in freeze-
tolerant wood frogs (O’Connor and Rittenhouse
2016), arthropod communities (Bokhorst et al 2012),
and small rodents (Korslund and Steen 2006).
Future climate change scenarios predict warmer
winter temperatures, which are often accompanied by
an increase in precipitation falling as rain rather than
snow and an overall increase in air temperature vari-
ability (e.g. cold air outbreaks) (Vaughn et al 2013).
ThismeansthatmostoftheGreatLakesRegionwill
experience decreased snow depth, increased snow den-
sity, and temperature variability, which will reduce
the extent, duration, and thermal stability of the sub-
nivium. In this future scenario, areas that receive high
amounts of lake-effect snow may be important refu-
gia for subnivium-dependent species. Future research
needs to quantify how climate change will affect sub-
nivium phenology to understand the impacts of these
changes on the myriad of species that depend on this
seasonal habitat.
Acknowledgments
We thank M Moore for sharing his engineering
expertise and M Fitzpatrick, S Petty, A Mosloff, and
L Werner for their help in data collection, as well as B
Ace, J Clare, J Grauer, M Garces-Restrepo, C Latimer,
C Lane, and A Shipley. We thank Dane County Parks,
University of Wisconsin-Madison Arboretum, Univer-
sity of Wisconsin-Milwaukee Field Station, US Forest
Service, University of Wisconsin-Steven’s Point, Hunt
Hill Audubon Sanctuary, University of Minnesota
Cloquet Forestry Center, and Michigan Technological
University Ford Center for access to their land. Fund-
ing for the project was provided by National Science
Foundation’s Macrosystems Biology program, grant
EF-1340632.
ORCID iDs
Kimberly L Thompson https://orcid.org/0000-0002-
9644-3855
References
Andresen J A et al 2014 Historical climate and climate trends in the
midwestern United States Climate Change in the Midwest: A
Synthesis Report for the National Climate Assessment ed J A
Winkler et al (Washington, DC: Islan d Press)
Arnold T W 2010 Uninformative parameters and model selection
using akaike’s information criterion J. Wildl. Manage. 74
1175–8
Bai J and Perron P 2003 Computation and analysis of multiple
structural change models J. Appl. Econ. 18 1–22
Bates D et al 2015 Fitting Linear Mixed-Effects Models Using lme4
J. Stat. Softw. 67 1–48
Bjork R G and Molau U 2007 Ecology of alpine snowbeds and the
impact of global change Arct. Antarct. Alpine Res. 39 34–43
Bokhorst S et al 2012 Extreme winter warming events more
negatively impact small rather than large soil fauna: shift in
community composition explained by traits not taxa Glob.
Change Biol. 18 1152–62
Bolker B M et al 2009 Generalized linear mixed models: a practical
guide for ecology and evolution Trends Ecol. Evol. 24 127–35
Burnett A W et al 2003 Increasing great lake-effect snowfall during
the twentieth century: a regional response to global warming?
J. Clim. 16 3535–42
Cade B S 2015 Model averaging and muddled multimodel
inferences Ecology 96 2370–82
Changnon S A and Jones D M A 1972 Review of infl uences of Great
Lakes on weather Water Res. Res. 8360
Chen X N et al 2016 Satellite observed changes in the northern
Hemisphere snow cover phenology and the associated
radiative forcing and feedback between 1982 and 2013
Environ. Res. Lett. 11 084002
Dawson N et al 2017 A new snow density parameterization for land
data initialization J. Hydrometeorol. 18 197–207
Derksen C et al 2014 Physical properties of Arctic versus subarctic
snow: Implications for high latitude passive microwave snow
water equivalent retrievals J. Geophys. Res. Atmos. 119
7254–70
11

Environ. Res. Lett. 13 (2018) 064037
Essery R et al 2008 Radiative transfer modeling of a coniferous
canopy characterized by airborne remote sensing
J. Hydrometeorol. 9228–41
Gascoin S et al 2013 Wind effects on snow cover in Pascua-Lama,
Dry Andes of Chile Adv. Water Resour. 55 25–39
Ge Y and Gong G 2010 Land surface insulation response to snow
depth variability J. Geophys. Res. Atmos. 115 D08107
Groffman P M et al 2001 Colder soils in a warmer world: a snow
manipulation study in a northern hardwood forest ecosystem
Biogeochemistry 56 135–50
Hagvar S 2010 A review of Fennoscandian arthropods living on and
in snow Eur. J. Entomol. 107 281–98
Harestad A S and Bunnell F L 1981 Prediction of snow-water
equivalents in coniferous forests Can. J. Forest Res. 11 854–7
Hedstrom N R and Pomeroy J W 1998 Measurements and
modelling of snow interception in the boreal forest Hydrol.
Process. 12 1611–25
Heffernan J B et al 2014 Macrosystems ecology: understanding
ecological patterns and processes at continental scales Front.
Ecol. Environ. 12 5–14
Judson A and Doesken N 2000 Density of freshly fallen snow in the
Central Rocky Mountains Bull. Am. Meteorol. Soc. 81 1577–87
Kim D and Jamieson B 2014 Estimating the density of dry snow
layers from hardness and hardness from density Int. Snow Sci.
Workshop Banff 499–506
Korslund L and Steen H 2006 Small rodent winter survival: snow
conditions limit access to food resources J. Anim. Ecol. 75
156–66
Kreyling J et al 2012 Absence of snow cover reduces understory
plant cover and alters plant community composition in boreal
forests Oecologia 168 577–87
Lemke P et al 2007 Observations: changes in snow, ice and frozen
ground Climate Change 2007: the Physical Science Basis
Contribution of Working Group I to the Fourth Assessment
Report of the Intergovernmental Panel on Climate Change
ed S Solomon et al (Cambridge: Cambridge University Press)
pp 337–83
Lundquist J D et al 2013 Lower forest density enhances snow
retention in regions with warmer winters: a global framework
developed from plot-scale observations and modeling Water
Resour. Res. 49 6356–70
Marchand P J 2013 Life in the Cold: An Introduction to Winter
Ecology (Lebanon, NH: University Press of New England)
Martin K A et al 2013 Development and testing of a snow
interceptometer to quantify canopy water storage and
interception processes in the rain/snow transition zone of the
North Cascades, Washington, USA Water Resour. Res. 49
3243–56
Ning L and Bradley R S 2015 Snow occurrence changes over the
central and eastern United States under future warming
scenarios Sci. Rep. 517073
O’Connor J H and Rittenhouse T A G 2016 Snow cover and late fall
movement influence wood frog survival during an unusually
cold winter Oecologia 181 635–44
Pauli J N et al 2013 The subnivium: a deteriorating seasonal
refugium Front. Ecol. Environ. 11 260–7
Peel M C et al 2007 Updated world map of the Koppen-Geiger
climate classification Hydrol.EarthSyst.Sci.11 1633–44
Peng S S et al 2013 Change in snow phenology and its potential
feedback to temperature in the northern Hemisphere over the
last three decades Environ. Res. Lett. 8014008
Petty S K et al 2015 Winter conditions and land cover structure the
subnivium, a seasonal refuge beneath the snow PLoS ONE 10
e0127613
Pfeffer W T and Mrugala R 2002 Temperature gradient and initial
snow density as controlling factors in the formation and
structure of hard depth hoar J. Glaciol. 48 485–94
Pinzer B R and Schneebeli M 2009 Snow metamorphism under
alternating temperature gradients: morphology and
recrystallization in surface snow Geophys. Res. Lett. 36
L23503
Pomeroy J W and Goodison B E et al 1997 Winter and snow The
Surface Climates of Canada ed W G Bailey (Montreal:
McGill-Queen’s University Press)
Pomeroy J W et al 2002 Prediction of seasonal snow accumulation
in cold climate forests Hydrol. Process. 16 3543–58
Pomeroy J W et al 1998 Coupled modelling of forest snow
interception and sublimation Hydrol. Process. 12 2317–37
Pomeroy J W and Schmidt R A 1993 The use of fractal geometry in
modeling intercepted snow accumulation and sublimation
Proceedings of the Sixty-First Annual Western Snow
Conference ed M Ferrick and T Pangburn (Quebec: 61st
Western Snow Conference) pp 1–11
Pruitt W O 2005 Why and how to study a snowcover Can. Field
Nat. 119 118–28
R Core Team 2015 R: A Language and Environment for Statistical
Computing (Vienna: R Foundation for Statistical Computing)
Ryan S E and Porth L S 2007 A Tutorial on Piecewise Regression
Approach Applied to Bedload Transport Data ed U S D O
Agriculture (Fort Collins, CO: Forest Service, Rocky
Mountain Research Station)
Sanders-DeMott R et al 2017 Snow depth, soil temperature and
plant-herbivore interactions mediate plant response to climate
change J. Ecol. 1–12
Schimel J P et al 2004 Increased snow depth affects microbial
activity and nitrogen mineralization in two Arctic tundra
communities Soil Biol. Biochem. 36 217–27
Scott R W and Huff F A 1996 Impacts of the Great Lakes on
regional climate conditions J. Great Lakes Res. 22 845–63
Varhola A et al 2010 Forest canopy effects on snow accumulation
and ablation: An integrative review of empirical results J.
Hydrol. 392 219–33
Vaughn D G et al 2013 Observtions: cryosphere Climate Change
2013: The Physical Science Basis. Contribution of Working
Group I to the Fifth Assessment Report of the
Intergovernmental Panel on Climate Change ed T F Stocker
et al (Cambridge: Cambridge University Press) pp 317–82
Wang S Y et al 2017 Complex responses of spring alpine vegetation
phenology to snow cover dynamics over the Tibetan Plateau,
China Sci. Total Environ. 593 449–61
Wheeler H C et al 2015 Phenological mismatch with abiotic
conditions-implications for flowering in Arctic plants Ecology
96 775–87
Whitaker D M and Stauffer D F 2003 Night roost selection during
winter by ruffed grouse in the central appalachians
Southeastern Nat. 2377–92
Zeileis A et al 2003 Testing and dating of structural changes in
practice Comput. Stat. Data Anal. 44 109–23
Zeileis A et al 2002 Strucchange: an R package for testing for
structural change in linear regression models J. Stat. Softw. 7
1–38
Zhang W et al 2016 Investigations on physical properties and
ablation processes of snow cover during the spring snowmelt
period in the headwater region of the Irtysh River, Chinese
Altai Mountains Environ. Earth Sci. 75 199
Zuur A F et al 2010 A protocol for data exploration to avoid
common statistical problems Methods Ecol. Evol. 1
3–14
12