The values of the snow and soil thermal conductivity,
Arctic permafrost contains large amounts of frozen organic matter (Hugelius
et al., 2014). Its thawing could lead to the microbial mineralisation of a
fraction of this carbon, resulting in the release of yet undetermined but
potentially very important amounts of greenhouse gases (CO
Obviously, soil thermal conductivity is also important in determining vertical heat fluxes. For a given soil composition, this variable depends mainly on temperature and content in liquid water and ice, so that its value may show considerable variations over time (Penner et al., 1975; Overduin et al., 2006). Modellers of the permafrost thermal regime have stressed that “monitoring of this parameter in the active layer all year-round would be useful if a more realistic numerical model is to be developed” (Buteau et al., 2004).
Location of the study site on Bylot Island in the Canadian Arctic archipelago and photograph of the monitoring station deployed. The polyethylene post with the three TP08 heated needle probes is in the foreground. The polyethylene post with the five thermistors is visible behind it. The radiometers, SR50 snow height gauge, cup anemometer, temperature and relative humidity gauge, and surface temperature sensors are visible on the tripod, from left to right. The CR1000 data loggers and batteries are in the metal box on the tripod. Batteries are recharged by solar panels and a wind mill in winter. Inset: detail of the lower two TP08 needles after their positions were lowered in July 2014.
Another important interest of studying snow physical properties such as thermal conductivity and the heat budget of the ground lies in the understanding of the conditions for subnivean life. For example, lemmings live under the snow most of the year at Bylot Island (Bilodeau et al., 2013), and the temperature at the base of the snowpack conditions their energy expenses to maintain body temperature. Furthermore, energy expense for subnivean travel in search of food depends on snow hardness and snow conditions have been invoked to help explain lemming population cycles (Kausrud et al., 2008; Bilodeau et al., 2013), even though no comprehensive snow studies have yet been performed to fully establish links between lemming populations and snow properties. Snow thermal conductivity has been shown to be well correlated with snow mechanical properties (Domine et al., 2011), so that monitoring snow thermal conductivity may help quantify the effort required by lemmings to access food and hence understand their population dynamics.
Given the importance of snow thermal conductivity to simulate the ground
thermal regime and to understand the conditions for subnivean life, we have
initiated continuous automatic measurements of the snow thermal conductivity
vertical profile at several Arctic sites using heated needle probes (NPs). The
first instrumented site was in low Arctic shrub tundra (Domine et al., 2015),
near Umiujaq on the eastern shore of Hudson Bay. We present here an additional
study with 2 years of snow thermal conductivity monitoring at Bylot Island,
a high Arctic herb tundra site (73
Stratigraphy and vertical profiles of snow physical properties near our study site on 12 May 2015. Visual grain sizes are indicated in the stratigraphy. Density data are for the middle of the 3 cm high sample. Snow type symbols are those of Fierz et al. (2009), except for the basal melt–freeze layer, which transformed into depth hoar to form an indurated layer, as detailed in the text.
Our study site is on Bylot Island, just north of Baffin Island in the
Canadian high Arctic. The actual site was at the bottom of Qalikturvik
valley, in an area of ice-wedge polygons (73
Three TP08 heated NPs from Hukseflux were positioned on a polyethylene post at heights above the ground of 7, 17 and 27 cm in July 2013. These heights were chosen somewhat arbitrarily before the snowpack structure could be observed. In July 2014 the TP08 needles were lowered to 2, 12 and 22 cm because we had realised during a field trip in May 2014 that the lowermost depth hoar layer could be thinner than 7 cm. On another nearby post, thermistors were placed at heights of 2, 7, 17, 27 and 37 cm. Heights intermediate between the TP08 heights allow the calculation of heat fluxes, using thermal conductivity and temperature values. Unfortunately, the cables of the thermistors at 7, 27 and 37 cm, although protected, were chewed by a fox in late summer 2014 and could only be replaced the following summer. In the ground, we also placed a TP08 NP at a depth of 10 cm with two thermistors at depths of 4.5 and 13.5 cm. In the immediate vicinity of the thermistors post, we placed 5TM sensors from Decagon to monitor ground temperature and volumetric liquid water content at depths of 2, 5, 10 and 15 cm. Water content sensors used the manufacturer's calibration for mineral soils and were not recalibrated, which may produce an error of up to 3 %. At that time, it was not possible to place deeper sensors because of the limited thaw depth. A few metres from these snow and ground instruments, we installed meteorological instruments to measure atmospheric variables including air temperature and relative humidity with a HC2S3 sensor from Rotronic and wind speed with a cup anemometer, both at 2.3 m height, and snow height with an SR50A acoustic snow height gauge (Campbell Scientific), an IR120 infrared surface temperature sensor and a CNR4 radiometer with a CNF4 heating/ventilating system from Kipp & Zonen which measured downwelling and upwelling short-wave and long-wave radiation. Heating and ventilating radiometers are mandatory to limit snow accumulation and the build-up of frost and freezing rain. Heating and ventilation were performed for 5 min every hour prior to measurements. This protocol of instrumentation and measurement is similar to that employed near Umiujaq (Domine et al., 2015), which was operated in the same manner with a CR1000 data logger (Campbell Scientific). Briefly, hourly measurements were recorded, except for the TP08 needles, whose operation is described in the next section.
The NP method has been used extensively for soils for a long time (Devries, 1952). Sturm and Johnson (1992) and Morin et al. (2010) discussed in detail the heated NP method in snow. The automatic operation of the TP08 needles in Arctic snow and the data analysis have been detailed in Domine et al. (2015) and only a brief summary will be given here.
For a measurement cycle, the 10 cm long needles were heated at constant
power (0.4 W m
In snow, NPs in fact measure an effective thermal conductivity,
The automatic routine developed by Domine et al. (2015) was used to obtain
Measurements of soil thermal conductivity
Measuring thermal conductivity in porous granular media such as snow has been
suspected of presenting biases and/or systematic errors (Calonne et al.,
2011; Riche and Schneebeli, 2013) and these have been discussed by Domine et
al. (2015). One concern is that the relevant metric for heat exchanges
between the ground and the atmosphere through the snow is the vertical heat
flux. Snow is anisotropic (Calonne et al., 2011; Riche and Schneebeli, 2013)
and NPs measure a mixture of vertical and horizontal heat fluxes. Anisotropy
depends on snow type, with, for example, depth hoar being more conductive in
the vertical direction and wind slabs in the horizontal direction. Moreover,
a systematic error caused by the granular nature of snow was invoked by Riche
and Schneebeli (2013). Domine et al. (2015) discussed the impact of these
processes on the accuracy of
For soils, the issues raised by Riche and Schneebeli (2013) are expected to
have no impact because soils are much denser with smaller grains and are
closer to a homogeneous medium at the scale of a NP. Anisotropy may, however,
be an issue in some soil types, but we have not developed this aspect here.
Based on the quality of our soil heating curves, we estimate that the error
on
Field measurements of snow
Soil grain size distribution is useful to understand its physical properties. Samples were taken from our instrumented site at 10 cm depth in July 2015. Particle size distribution was measured with a Horiba partica LA-950V2 laser scattering particle size analyser, which used a two-wavelength optical system, 405 and 650 nm. Five subsamples were analysed and averaged.
We used the Crocus model (Vionnet et al., 2012) coupled to the land surface model ISBA within the SURFEX interface (version 7.3) to simulate snow physical properties. We in fact used the simulations already described in Domine et al. (2016) for our Bylot Island site, but we analysed different output data. Very briefly, the model was forced with our meteorological data. When data were missing, ERA-Interim reanalysis data were used (Dee et al., 2011), corrected following the procedure of Vuichard and Papale (2015) to minimise the bias between measured and ERA data. Snow thermal conductivity is calculated from the equation of Yen (1981), based on a correlation between density and thermal conductivity.
The winter 2013–2014 was a low snow year so that two out of three NPs were not covered. Data are more complete for the following year and we therefore start with data from the 2014–2015 season.
Describing the structure and physical properties of the snowpack at Bylot
Island helps to understand the monitoring data. Observations made on
12 May 2015 close to our monitoring site are shown in Fig. 2. Vegetation was
observed to be mostly flattened by snow, with some sedge or graminoids stems
still upright, but they did not seem to have impacted snow structure. We
observed a basal depth hoar layer 8 to 10 cm thick, overlaid by a wind slab
11 to 12 cm thick, with in-between a thin layer of faceting crystals. Above
the wind slab were thin layers of small rounded grains, decomposing
precipitation crystals and a thin wind crust. The depth hoar layer was
divided into two sublayers. The lower sublayer was slightly indurated and
harder, although hardness and other properties appeared spatially very
variable. Indurated depth hoar is a snow type seldom mentioned, as it does
not form in alpine or temperate snow, and it is not described in the
international snow classification (Fierz et al., 2009) despite its
widespread presence in the Arctic, where it has been observed without being
named for decades. Hall et al. (1991) described “solid-type depth hoar”,
presumably indurated depth hoar. Sturm et al. (2002) mentioned “wind slab to
depth hoar” layers, presumably also indurated depth hoar. Its most detailed
description is probably that of Sturm et al. (2008): This type of layer is generally not seen outside of the Arctic. It arises in
the Arctic because the temperature gradients are so strong that even dense,
fine-grained layers of wind slab eventually metamorphose into large, faceted,
and striated depth hoar grains. A key characteristic of these slab-to-hoar
layers is that they are tough, not fragile like most depth hoar. If comprised
of mainly depth hoar crystals, yet still cohesive, we called the layers
“indurated”. If a significant number of small wind grains remained, we
called them “slab-like”. Indurated depth hoar …forms in dense wind slabs under very high
temperature gradients not encountered in alpine snow. Its density can exceed
400 kg m
The indurated depth hoar here showed signs of early season melting in the
form of rounded grains that were partially to almost totally transformed into
depth hoar, but bonds were stronger than for regular depth hoar. There are
therefore two main precursors to indurated depth hoar: wind slabs as detailed
by Sturm et al. (2008) and refrozen layers as observed here and to the best
of our knowledge only previously described in Domine et al. (2016). The
indurated depth hoar that forms in refrozen snow is slightly different from
that observed in wind slabs, in that no small grains are present and it does
not have a milky aspect. In line with Domine et al. (2016), we represent this
type of depth hoar with a symbol that does not exist in the classification of
Fierz et al. (2009), as that classification is ideal for alpine snow but is
not detailed enough to represent many Arctic snow types. The depth hoar upper
sublayer was very soft, appeared more homogeneous and showed no signs of
melting. Vertical profiles of density, SSA and thermal conductivity are also
shown in Fig. 2. Several density measurements were made in the lower depth
hoar layer, yielding values between 172 and 260 kg m
Figure 3 shows time series for
Variations in values of thermal conductivity,
The lowest NP, at 2 cm, was covered by the first snowfall on
12 September 2014. That initial snow partially melted and the first data
point that was almost certainly obtained in snow was on 24 September. After
initial values around 0.04 W m
Snow temperature, air temperature, wind speed, snow height and snow thermal conductivity at three heights during the 2014–2015 winter season at Bylot Island. The levels of the three thermal conductivity needle probes (NPs) are indicated in the snow height panel. Note that the snow height gauge and the NPs were about 5 m away, so that snow height between both spots may have been different. Snow temperatures were measured with the NPs every other day at 05:00.
The NP at 12 cm was probably in fresh snow on 1 November. On 10 to
11 November, the strongest wind storm of the winter, reaching a speed of 12.9
m s
The NP at 22 cm was definitely covered by snow on 21 January 2015.
Three soil pits were dug in the summers 2013 to 2015 down to the thaw front
in the polygon where our instruments are located to measure soil physical
properties and another two pits were dug just for observations. An organic
litter layer 3.5 to 6 cm thick was observed. Lower down was a layer of
organic-rich silt-looking material. Figure 4 shows vertical profile of soil
temperature, thermal conductivity,
Vertical profiles of soil physical variables in pits dug in the
polygon where our instrument station is located, during summers 2013 to 2015.
Seasonal evolution of the thermal conductivity, temperature and volume water content of the soil at 10 cm depth for the 2014–2015 season. The 5TM probe which measures both temperature and water content hourly is about 2 m from the TP08 NP, which measures thermal conductivity and temperature every 2 days.
Figure 5 shows the evolution of the
Figure 6 shows the value of
Snow temperature, air temperature, wind speed, snow height and snow thermal conductivity at 7 cm height during the 2013–2014 winter season at Bylot Island. The levels of the three thermal conductivity needle probes (NPs) are indicated in the snow height panel, showing that only the lowermost NP was covered.
Stratigraphy and vertical profiles of snow physical properties near our study site on 14 May 2014. Fresh snow is just a 2 mm thick sprinkling, also visible in the SSA profile. Density data are for the middle of the 3 cm high sample. Snow type symbols are those of Fierz et al. (2009), except for the lower wind slab, which transformed into depth hoar to form an indurated layer, as detailed in the text. Red-filled black squares in the stratigraphy indicate where thermal conductivity measurements were made.
Gaps in the snow in the basal depth hoar layer.
The snow stratigraphy was observed on 14 May 2014 about 50 cm from the NP
post and vertical profiles of density, specific surface area and thermal
conductivity were measured and are shown in Fig. 7. The stratigraphy was
spatially extremely variable and complex with frequent alternation of hard
and soft layers. The basal layer of columnar depth hoar was very soft and
collapsed at the slightest contact so that we were not able to measure its
density. By comparison with other observations, it was definitely
< 200 kg m
The intermediate layer of faceted crystals around 10 cm indicates an extended period of low wind weather, as visible in Fig. 6 between 21 January and 5 March, during which temperature gradient metamorphism could proceed without perturbation by any wind compaction episode. Few precipitation events took place that winter, as indicated by the small amount of snow observed in May 2014. The snow gauge (Fig. 6) also indicates little precipitation, although many wind-erosion episodes at our gauge spot limit our ability to evaluate precipitation in 2013–2014.
Snow cover in 2013 started late, on 12 October. Our NP at 7 cm recorded a
first significant
Photograph of the snow stratigraphy taken on 14 May 2014. The NPs are 7 and 17 cm above the ground. The various depth hoar and indurated depth hoar layers between 0 and about 11 cm are clearly visible, as well as the wind slab between 11 and 16 cm.
On 14 May 2014, we excavated the snow around the NPs, essentially ending our
time series. A photograph of the snow profile is shown in Fig. 9. The NP at
7 cm was in an indurated depth hoar layer, but very close to the border with
a thin depth hoar layer. Above that was a layer of faceted crystals/depth
hoar. Given the stratigraphy, the 7 cm NP had been completely buried for
months and changes in
Figure 10 shows soil data for the 2013–2014 season. Before the initiation of
freezing, the soil volume water content at 10 cm depth was
56.6
Seasonal evolution of the thermal conductivity, temperature and volume water content of the soil at 10 cm depth for the 2013–2014 season. The 5TM probe which measures both temperature and water content hourly is about 2 m from the TP08 NP, which measures thermal conductivity and temperature every 2 days.
Time series of the temperature gradient in the snow. Values were obtained from the heated needle probes at 2, 12 and 22 cm, with a data point every 2 days. Thermistors at 2 and 17 cm also measured temperature every hour, and the values are shown with an hourly resolution. The different start dates of each curve are determined by the date where the snow height reached the relevant level.
The snowpack structure observed at Bylot Island, especially in 2015, is
frequently encountered on Arctic tundra (Benson and Sturm, 1993; Domine et al.,
2002; Sturm and Benson, 2004; Sturm et al., 2008), especially in areas of
moderate wind, and mostly consists of a lower depth hoar layer and an upper
wind slab. The depth hoar layer forms because of the elevated temperature
gradient at the beginning of the season. Figure 11 shows the temperature
gradients in the 2–12 and 12–22 cm snow height ranges, as obtained from
the NPs temperature measurements every other day at 05:00. Higher time
resolution measurements would have been desirable, but most of the
thermistors that logged temperature every hour were damaged by a fox.
Figure 11 nevertheless shows that NP data are similar to the gradient derived
from thermistor data at 2 and 17 cm, so that reasoning on NP data is still
adequate. Values barely reach 100 K m
With regards to metamorphism, the actual variable of interest is the water
vapour flux rather than just the temperature gradient. This flux is the
product of the diffusion coefficient of water vapour in snow,
Time series of the water vapour flux at two levels in the snowpack. Positive fluxes are upward.
It is interesting to evaluate whether calculated fluxes can explain the mass
loss leading to snow collapse. Assuming all the flux comes from a 5 cm thick
depth hoar layer of density 250 kg m
Figure 12 shows that in the upper region, the water vapour flux was much lower
and apparently insufficient to allow depth hoar formation. For most of the
snow season, there was a continuous upward water vapour flux, leading to
overall water vapour loss to the atmosphere. The late season reversal of the
flux direction in early May lasted only about a month and was insufficient to
reverse the overall loss trend. This reversal coincides with the change in
the trend of evolution of
There is also a temperature gradient, and hence a water vapour flux, between
the soil and the snow, as already mentioned by Sturm and Benson (1997) from
observations in interior Alaska. The resulting soil water vapour loss is
detectable in Fig. 10. In late summer 2013, the liquid water content in the
soil at 10 cm depth was about 58 %. After thawing in early July 2014,
the water content only rises back to 31 %, meaning that almost half of the
water present in the soil the previous summer has been lost by sublimation
during the snow season. If this value applies to the top 10 cm of the soil,
then it lost 17 kg m
The presence of a soft depth hoar layer clearly facilitates subnivean travel and food search. The softer the layer is, the easier the travel and presumably the better the feeding and reproductive success of subnivean species. Factors that adversely affect the softness of this layer include wind packing and melt–freeze events. The strength of the temperature gradient may allow the transformation of wind slabs and ice crusts into indurated depth hoar, but such depth hoar is much harder than that formed in softer snow (Domine et al., 2009, 2012). This is probably what happened in autumn 2014, as signs of early season melt–freeze cycling were still observable in May 2015 (Fig. 3). Based on these observations, it appears that feeding conditions for subnivean species may have been slightly better in 2013–2014 than in 2014–2015.
The recent work of Fauteux et al. (2016) is consistent with our observations.
These authors measured lemming abundance very close to our snow study site
right after snow melt in a large 9 ha exclosure to minimise predator impact
on populations. The exclosure data are therefore expected to be more likely to
be affected mostly by just snow conditions. Their data show counts of
6 lemmings ha
Our time series of
It is clear that simulations and measurements yield very different results.
Measurements show low values for the lower depth hoar layers and high values
for the upper wind slabs. On the contrary, simulations show a value around
0.26 W m
Vertical profiles of
Our proposed interpretation of these differences, aided by a detailed
analysis of Crocus output data, is as follows. Crocus simulated a melting
episode in late September, giving the basal layer a high thermal
conductivity. This is consistent with our observations of a melt–freeze relic
in the basal depth hoar layer. However, Crocus cannot predict the
transformation of a melt–freeze layer into depth hoar because it does not
simulate the required vapour fluxes. These fluxes lead to mass loss in the
lower layers and mass gain in the upper ones. This is an important process
that contributes to the observed inverted density profiles (Fig. 2) (Sturm
and Benson, 1997) and hence the inverted thermal conductivity profiles
because thermal conductivity is calculated from density only. The differences
highlighted in Fig. 13 are not due to specificities of Crocus. Simulations
with SNOWPACK version 3.30 driven by North American Regional Reanalysis
(NARR) data (
A detailed evaluation of the ability of Crocus to reproduce the ground thermal regime is in order. However, ground temperature also depends on soil properties so that coupling to a land surface scheme is required for full testing. Crocus is currently coupled to the land surface scheme ISBA through the SURFEX interface. Improved snow and soil schemes for ISBA are being tested (Decharme et al., 2016) and the evaluation of these new schemes will be the subject of future work.
Lastly, the parameterization of Yen (1981) to calculate thermal conductivity
from density may not be suitable for Arctic snow. For a density
representative of the depth hoar studied, 200 kg m
Understanding and predicting the seasonal variations of
A tempting approximation would be to use a bimodal distribution of thermal
conductivity values, as we find transition periods of less than 10 days. The
transition from thawed to frozen
We feel that the following points are important conclusions of this study:
Vertical water vapour fluxes induced by the temperature gradients in the
soil and the snowpack strongly determine snow conditions, soil dehydration
and the water budget of the surface. Water vapour fluxes also determine the snow thermal conductivity
profile and the ground thermal regime. The comparison of observed vs.
simulated thermal conductivity profile demonstrates that omitting these
fluxes leads to a radically different snow thermal conductivity profile. Major snow models (Crocus, SNOWPACK) do not describe water vapour
fluxes. The consequences on the water budget, on the ground thermal regime,
on the energy budget of the surface and possibly on climate may be quite
significant. For both years studied, a layer of soft depth hoar was present at
the base of the snowpack, which seems to be favourable conditions for
subnivean life. In the second year, a melt–freeze layer at the very base of
the snow pack may have rendered conditions somewhat less favourable for a few
weeks, but Soil thermal conductivity showed transition periods of just a few
days between the thawed and the frozen values. Modelling soil thermal
conductivity with a step function may therefore be tested. A hysteresis
process was observed, with the change from thawed to frozen value taking
place at low ice content and the change from frozen to thawed values at high
ice content. Further work is needed to determine whether these processes need
to be taken into account for adequate simulation of the permafrost thermal
regime at our study site.
Florent Domine designed research. Denis Sarrazin built and deployed the instruments with assistance from Florent Domine. Mathieu Barrere and Florent Domine performed the field measurements. Florent Domine and Mathieu Barrere analysed the field data. Mathieu Barrere performed the model simulations. Florent Domine prepared the manuscript with comments from Mathieu Barrere and Denis Sarrazin.
This work was supported by the French Polar Institute (IPEV) through grant 1042 to FD and by NSERC through the discovery grant program. We thank Laurent Arnaud for advice in writing the program to run the needle probes. The Polar Continental Shelf Program (PCSP) efficiently provided logistical support for the research at Bylot Island. We are grateful to Gilles Gauthier and Marie-Christine Cadieux for their decades-long efforts to build and maintain the research base of the Centre d'Etudes Nordiques at Bylot Island. Winter field trips were shared with the group of Dominique Berteaux, who helped make this research much more efficient and fun. Bylot Island is located within Sirmilik National Park, and we thank Parks Canada and the Pond Inlet community (Mittimatalik) for permission to work there. The assistance of Matthieu Lafaysse, Samuel Morin and Vincent Vionnet at Centre d'Etudes de la Neige (Météo France-CNRS) for the use of Crocus is gratefully acknowledged. We thank J.-B. Madore and A. Langlois (University of Sherbrooke) for sharing with us their results on SNOWPACK simulations. Constructive comments by Martin Schneebeli, Matthew Sturm, an anonymous reviewer, and the editor Ross Brown are gratefully acknowledged.Edited by: R. Brown Reviewed by: M. Sturm, M. Schneebeli and one anonymous referee