Manual measurement of snow water equivalent (SWE) is
still important today for several applications such as hydrological model
validation. This measurement can be performed with different types of snow
tube sampler or by a snow pit. Although these methods have been performed
for several decades, there is an apparent lack of information required to
have a consensus regarding the best reference for “true” SWE. We define
and estimate the uncertainty and measurement error of different methods of
snow pits and snow samplers used in a boreal biome. Analysis was based upon
measurements taken over five consecutive winters (2016–2020) from the same
flat and open area. This study compares two snow pit methods and three snow
samplers. In addition to including the Standard Federal sampler (SFS), this
study documents the first use of two new large diameter samplers, the
Hydro-Québec sampler (HQS) and Université Laval sampler (ULS). Large diameter samplers had the lowest uncertainty (2.6 % to 4.0 %). Snow pit methods had higher uncertainty due to
instruments (7.1 % to 11.4 %), close to that of the SFS (mean
The water equivalent of snow cover (SWE) is a key attribute in hydrological research and applications for watersheds that are supplied by snowmelt. SWE data are essential in different applications, such as forecasting spring freshets, estimating water supplies to hydroelectric dams, or calibrating hydrological models. Historically and still today, SWE data are acquired by manual measurements. Despite the installation of automatic SWE sensors in weather stations, manual SWE values are required to calibrate and evaluate the efficiency of these instruments. Whether SWE results are validated from a model or an automatic sensor, several manual methods and instruments can be used to obtain a reference value that is as close as possible to the “true” SWE value.
All manual SWE measurements are based upon the same principle, i.e.,
multiplying snow density by snow depth. Snow tube samplers or snow corers
are used widely to measure these two parameters (Goodison et
al., 1981). To our knowledge, the first documentation in English that
mentions the use of snow tubes dates from 1933 in the western United States
(Nevada), with the description of a 1
In the Province of Quebec (Canada), the Standard Federal sampler (SFS) is
most frequently used for obtaining SWE measurements over 100-plus
snow courses that are managed by the Ministère de l'Environnement et de la Lutte contre les changements climatiques (MELCC) (Ministère du Développement durable, de l'Environnement et des Parcs, 2008).
Initially designed for measurement in dense and deep snowpack
(Work et al., 1965), the SFS is widely used across North
America, given its many advantages, such as ease of transport and its use in
the field (Goodison et al., 1981). The SFS enables the
measurement of snow depth, mean snow density of the snow cover, and the SWE.
The two most common sources of uncertainty for this type of sampler are its
side slots and its cutter design (Dixon and Boon,
2012). Previous studies in western Canada showed that the SFS overestimates
SWE by 4.6 % to 10.5 % compared to SWE that is obtained by weighing all of
the snow within a large-sized test plot (
Among the elements that can explain differences in accuracy among different
samplers, the most common are the size of the sampler opening and its
ability to penetrate the snowpack without producing an ice plug
(Freeman, 1965). To obtain SWE measurements with greater accuracy,
snow samplers with a larger diameter have been designed. Larger samplers
will show better performance in snow covers in the presence of dense snow or
ice layers (Goodison, 1978; Dixon and Boon,
2012). Large diameter samplers, such as the ESC30 (i.d.
Another way of measuring SWE is by estimating it from measurements that are
acquired within a snow pit. The snow pit is generally considered to be a
good reference for the “true” SWE value (Sturm et
al., 2010). Snow pits, made using a variety of protocols and instruments,
has been frequently used as the SWE reference when evaluating the error of
snow samplers
(Sturm
et al., 2010; Dixon and Boon, 2012; López-Moreno et al., 2020) or
automatic SWE sensors (Choquette et al.,
2013; Kinar and Pomeroy, 2015b; Henkel et al., 2018; Mavrovic et al., 2020).
One major advantage of making a snow pit is that it permits the observation
and measurement of the stratigraphy of the snowpack
(Kinar and Pomeroy, 2015a). A snow pit is simply an
opening that is manually excavated in the snow cover and is generally large
enough to enable a person to stand comfortably in it and make measurements
on the vertical face of the pit. Numerous snow density measurements are
taken with the depth of the snow pit using a sampler of specific volume,
which is known as a density cutter. These snow density measurements are then
used to estimate the snow cover SWE. Density cutters can assume different
shapes (cylinder, wedge, or box) and volumes (100 to 1000 cm
Snow pit SWE can be estimated according to two methods, which differ in their snow density sampling approach. The first method consists of considering stratification of the snow cover (Pomeroy and Gray, 1995; Sturm et al., 2010; Canadian Avalanche Association, 2016; Senese et al., 2018). Using the density and thickness measurements for each snow layer to estimate their respective SWE, snow pit SWE is calculated as the sum of all snow layer SWE. A second method of calculation disregards the stratification of the snow cover by taking density samples at regularly spaced intervals between the ground surface and the surface of the snowpack, i.e., continuous sampling strategy (Elder et al., 1991; Fassnacht et al., 2010; Dixon and Boon, 2012; Proksch et al., 2016; World Meteorological Organization, 2018). Thus, snow pit SWE is obtained as the product of snow depth and average snowpack density. Moreover, both snow pit methods require much more time, equipment, and expertise than snow tube sampling when it comes to estimating snow water equivalent (Pomeroy and Gray, 1995). A variation of this method measures all of the snow from the snow surface to the ground using a metal cylinder that is referred to as a glacier sampler. This method was used by the Western Snow Conference Metrication Committee as the SWE reference in many comparisons that evaluated different snow tube samplers (Farnes et al., 1983). Measurements are taken every 35–38 cm, where a metal plate is placed perpendicular to the snow pit to stop the glacier sampler between two snow samples. Regardless of the method that is used to calculate snow pit SWE from density cutters, they rely upon the sum of numerous measurements that are each prone to errors. Inevitably, estimates of snow cover SWE are dependent upon the sum of multiple uncertainties. To our knowledge, uncertainty of the snow pit SWE has never been estimated.
When considered accordingly, the manipulations that are necessary for the measurement of SWE from snow pits have many sources of uncertainty and inevitably generate errors. Therefore, this widely recognized method of reference for SWE measurement possibly could generate over- or underestimates of the “true” SWE. The objective of this study is to estimate the uncertainty and the measurement error of numerous snow pit and snow sampler methods used in a boreal biome to identify which would represent the most appropriate method of reference for the “true” SWE. We posit that once all errors are considered, large size snow tube samplers would yield consistent measurements and results that are closest to the “true” SWE. We further expect snow pits to exhibit great variability and, therefore, not to be representative of the most appropriate reference for “true” SWE. Although the concepts of uncertainty and measurement error seem basic, it is possible to find different interpretations in the literature (JCGM, 2008). This confusion leads to difficulty in understanding and comparing different SWE measurement methods. In order to avoid a misinterpretation of the results presented in this study, the calculated statistical values will be supported by definitions and equations from the literature.
Forêt Montmorency (FM) covers 397 km
For five winters (January 2016 to May 2020), manual SWE measurements of the snow cover were performed on a weekly or bi-weekly basis. SWE data were taken from January (winters of 2016, 2017, and 2018) or from November (winters of 2019 and 2020) until the snowpack melted completely, which generally occurred in mid-May. During each field visit, three SWE measurements were made with each snow tube sampler, i.e., the Standard Federal sampler (SFS), the Hydro-Québec sampler (HQS), and the Université Laval sampler (ULS) (Fig. 1). SFS and ULS were used throughout the study, while the HQS was added to the measurement campaign from winter 2018 onward. All snow measurements were collected exclusively by meticulously trained scientific observers, of which the lead author and a technician were respectively responsible for 22 % and 66 % of the field visits.
Snow samplers that were used in this study. From left to right:
The SFS was designed in 1935 by the U.S. Soil Conservation Service
(Work et al., 1965). This snow sampler consists of
0.76 m long sections of aluminum tubing (i.d.
The Hydro-Québec sampler (HQS) was designed by the provincial electrical utility to provide their employees with an alternative SWE measurement method,
particularly when snow conditions caused SFS measurements to be unreliable,
such as the presence of melt–freeze crust or ice layers that would clog the SFS.
The HQS is constructed of a 1.45 m long aluminum tube (i.d.
The Forest Hydrology Laboratory (Université Laval) constructed a large diameter snow sampler, hereafter referred to as the Université Laval sampler (ULS). This sampler is
fabricated from a 1.52 m long polyvinyl chloride (PVC) tube (i.d.
For a SWE measurement that is made with a snow tube sampler to be considered
valid, it is generally recognized that the ratio between snow core length
and snow depth must be
A snow pit was dug at each field visit to provide a SWE value and a
description of snowpack stratification similar to those documented by Fierz et al. (2009). Density measurements were made with a wedge
density cutter with a volume of 250 cm
For the average density method (method 2), the principle is based upon
calculating a mean snow cover density from multiple samples that were
collected at regular depth intervals in the snow pit, i.e., continuous
sampling strategy (Farnes et al., 1983;
Elder et al., 1991; World Meteorological Organization, 2018). Such regular depth interval sampling is not
exempt from errors since ice layers are often neglected given the
difficulty of them being sampled without bias
(Proksch et al., 2016). Some layers of
diverging density and thickness also can be under- or over-represented.
Despite a strictly continuous sampling strategy not being used in this study,
method 2 was used to calculate SWE from the average density of multiple
density samples that were described earlier and from total snow depth that
was measured in the snow pit (
In order for the statistical results to be correctly interpreted, it is important to clearly define what this study means by uncertainty and measurement error. Produced by the Joint Committee for Guides in Metrology (JCGM), the guide “Evaluation of measurement data – Guide to the expression of uncertainty in measurement” was used for this study in order to have an adequate definition of the statistical concepts discussed (JCGM, 2008). Uncertainty of a measurement method represents the dispersion of values that are attributable to the measuring instrument (JCGM, 2008). To characterize the uncertainty of each snow sampler, the coefficient of variation (CV) was calculated as the ratio between the standard deviation and the mean. Because the three snow samplers allow the measurement of snow depth, snow density, and SWE, a CV was calculated for these three variables. Since repeated measurements that were taken with each snow sampler were in close proximity to one another on open and flat terrain, sources of uncertainty should be due to the instrument and not to random effects that are induced by spatial and temporal variation. This remains a purely theoretical assumption in which measurement conditions are set in place to minimize random effects, yet the impact of these effects on the uncertainty cannot be precisely measured. The spatial variability in the snow properties was calculated by the CV of the average snow depth and snow density for each measurement day. The CV calculation was made for each snow sampler to best estimate the spatial variability in the snow conditions on the NEIGE site, therefore without the variability between the different SWE estimation methods used. In order to analyze if there is a temporal influence on the uncertainty of the SWE measurement, the CV was calculated depending on whether the measurements are taken during periods of snow accumulation or snowmelt.
For the snow pit, the uncertainty of snow layer density has been evaluated
from the CV of repeated density cutter measurements. However, this
uncertainty estimate did not match the overall uncertainty of the SWE
measurement that was derived from the snow pit. First, since only one snow
pit was excavated at each field visit, it technically forbids the
calculation of a CV of the SWE. Second, the CV was not necessarily the most
appropriate metric to facilitate robust comparisons of the uncertainty of
methods that were based upon single integrative measurements (snow samplers)
to methods that were based upon cumulation or averaging of multiple
separated measurements (snow pit). Therefore, uncertainty was evaluated from
the estimates of the precision of each instrument that was used to calculate
SWE of each snow sampler and snow pit method. According to the statistical
principles of propagation of uncertainties, the different uncertainties must
be considered in either a relative or an absolute manner, according to the
formula that is used (Lindberg, 2000). For each equation that
was used for the SWE calculation, different measuring instruments with their
precision are used to estimate different variables. According to these
statistical principles, it was possible to calculate a theoretical
uncertainty that was attributable to instruments for the SWE obtained
according to the propagation of these uncertainties. Thus, the method-related
uncertainty estimate would only be influenced by the instruments that are
being used. The uncertainty due to instruments, therefore, would be excluded
from random effects such as errors that were associated with the
manipulation of instruments by different operators. For snow samplers,
uncertainty was associated with the precision of the handheld suspended
scale that was used to measure snow core mass. The snow core mass (
These two approaches to evaluate the uncertainties of a measurement provide additional information that cannot be directly compared with one another. When it is calculated with the CV, the uncertainty value takes into account both random and systematic effects (JCGM, 2008). It then considers that for a measurement under the same conditions, there is an uncertainty associated with the instrument used, the observer, and the spatial and temporal variation, among others. In contrast, the uncertainty due to instruments ignores random errors, and, therefore, it concerns only a portion of the uncertainty. It is only associated with the precision of the instruments that are used. Conceptually, this uncertainty due to instruments would always be the same under different conditions, while the uncertainty that is calculated with the CV may vary according to random effects.
As mentioned by JCGM (2008), accuracy and measurement error are different
concepts sometimes confused or misinterpreted. Accuracy represents the
ability of an instrument to estimate a value that is as close as possible to
the “true” value (JCGM, 2008). By this definition,
accuracy is only a qualitative concept of a measurement method in which the
measurement error is considered low. What has been calculated in this study
is the error of measurement corresponding by definition to the difference
between the measured values and a reference value (JCGM,
2008). In theory, the “true” value is obtained during a perfect
measurement. Since it is not possible to know the “true” SWE value, a
method was chosen, according to our best information, to be considered as
the reference. To estimate the measurement error of each method, the largest
sampler, the ULS, was considered as the reference for SWE measurement.
Although bulky and cumbersome, it demonstrated the greatest reliability to
execute a constant and robust sampling of the snow cover in the field. Mean
bias error (MBE) was calculated for each method to represent the measurement
error compared to the reference. The MBE (%) has been calculated using
the following equation:
To compare the different methods further, the sampled volume for each
measurement was calculated. For snow samplers, the sampled volume was
calculated as the product of the inner area of the snow sampler multiplied
by snow depth. For the snow pit, the number of density samples that were
taken was counted. The cumulative volume of the three measurements that were
performed per sampled snow layer was considered. Since the density cutter
that was used had a fixed volume of 250 cm
During five winters of field campaigns at the NEIGE site, snow measurements
were taken in a wide variability of snow depth, snow density, and SWE (Fig. 2). The mean snow depth, snow density, and SWE of all measurements were
respectively 99 cm
Distribution of
In order to represent the NEIGE site spatial variability, a coefficient of variation of the measurement of snow depth and snow density was calculated for each field visit for the three snow samplers used. The average CV for the snow depth is 2.8 %, 2.3 %, and 2.0 % for SFS, HQS, and ULS respectively. For the snow density, the average CV is 5.2 %, 3.8 %, and 4.0 % for SFS, HQS, and ULS respectively. On the study site, there were from 0 to 10 ice layers per snowpack, averaging 5 ice layers. Their mean cumulative thickness was 15.4 cm.
From 2016 to 2020, a total of 91 snow pits were excavated, whereby 398 snow
layers were sampled, for a total of 1194 snow density measurements. Snow
density ranged between 0.090 and 0.590 g cm
Average coefficient of variation (CV) of three replicate measurements of snow density that were measured with a wedge density cutter for different snow density classes.
Measurements in low-density layers
Over the five winters of the study, 606 snow cores were collected for SWE
measurements with three snow samplers (Table 2). SWE values that were
measured with the three samplers ranged from 50 to 481 mm (mean
SWE measurements for snow samplers and snow pits.
Coefficient of variation of SWE that was measured with three snow samplers according to the period of snow accumulation or melting: the Standard Federal sampler (SFS), the Hydro-Québec sampler (HQS), and the Université Laval sampler (ULS). Boxes represent 25th and 75th percentiles. Bars inform of the lowest and highest quartiles, excluding the outliers represented by the dots. The middle line in each box shows the median of the data.
When the snow samplers were analyzed individually, only the ULS showed a
significant difference between the two period with a lower CV during the
accumulation period (
By calculating the uncertainty due to instruments using the principles of propagation of uncertainties, it was possible to estimate the uncertainty of the snow pit, in addition to the snow samplers (Table 3). For snow samplers, the absolute value in millimeters is a constant value associated with the instruments and not dependent upon the snow depth or SWE. Therefore, a measurement that is made with the SFS is always associated with an uncertainty of 20 mm of SWE regardless of the estimated SWE value. For the snow pit methods, the snow density is not estimated with a single measurement for the total snow cover but by several measurements in each snow layer. Therefore, the absolute uncertainty in millimeters of SWE would be dependent upon the number of snow layers. For example, uncertainty (in mm) that was associated with a SWE measurement of a snow pit would be 2-fold greater if there were 10 snow layers rather than 5 snow layers. There is no direct relationship between uncertainty and snow depth, except that there will be generally more snow layers in a deeper snow cover.
Uncertainties due to instruments of the snow pit and snow samplers.
Compared to the relative uncertainty due to instruments (%), the SFS displayed the greatest value for snow samplers, i.e., more than twice the HQS uncertainty, and almost 4 times the ULS uncertainty. To provide some perspective, the minimum uncertainties of SFS and snow pit methods 1-a and 1-b were higher than the average uncertainty of the HQS and ULS. For the snow pit methods 1-a and 1-b, the results are shown in the same column in Table 3 because the two calculation methods used the same measurements, i.e., density and thickness of each measurable snow layer. With respect to the two snow pit calculation methods, snow pit method 2 had an average uncertainty due to instruments that was lower than snow pit methods 1-a and 1-b but also higher than those of the snow samplers.
Measurement error was calculated using the ULS as a reference. The
mean bias error (MBE) results suggest that all SWE measurement methods
overestimate SWE compared to the ULS (Fig. 4). A mean bias error value close
to zero means a small error, corresponding to a very close agreement with the
ULS. A high positive value means the method overestimates SWE, while a
negative value means that SWE is underestimated. The snow sampler method
with the lowest error is the HQS (1.60 %). SFS has a significantly greater
MBE than the HQS (
Mean bias error of the different SWE measurement methods relative to the ULS, which was used as the SWE reference. The value above each boxplot is the average MBE for each method. Boxes represent 25th and 75th percentiles. Bars inform of the lowest and highest quartiles, excluding the outliers represented by the dots. The middle line in each box shows the median of the data.
Depending upon the sampling method that is used to estimate SWE, the sampled
snow volume that is required to obtain the former varies greatly from one
method to the next (Fig. 5). For the three snow samplers that were used in
this study, the sampled volume exhibits a strong linear relationship with
the increase in snow depth (
Volume of snow that was sampled using each SWE estimation method in relation to snow depth.
The volume of snow that was sampled for snow pits depends not only upon snow
depth but also upon the number of distinct snow layers. Since all layers
that were thinner than 5 cm were not sampled, while all layers that were
thicker than 5 cm were sampled three times, regardless of their thickness,
the volumes that were sampled originated from 1 layer (750 cm
Using data that were taken at the same location for five consecutive
winters, this study allows a unique comparison among different methods of
SWE measurement. With about 88 % of the measurements being made by two
observers, the bias that is induced by differences between observers is low.
Although this result is not presented, the comparison of the results of
uncertainty and measurement error between the two main observers did not
show any significant differences (
For the 91 d of snow sampling, it is interesting to note that there is a
difference in the number of samples between the SFS and the ULS, whereas the
two samplers should have taken the same number of snow cores. This
difference of 70 additional measurements for the ULS is explained by the
difference in diameter between the two samplers. When ice layers are found
in the snow cover, a small diameter sampler like the SFS had a greater
chance of forming a plug blocking the opening than would larger diameter
samplers. The average ratio that was measured between snow core length and
snow cover thickness supports this hypothesis. It is further strengthened by
the fact that the sampling method, which was applied meticulously, rejected
all samples showing possible snow loss during core extraction. The lower
average ratio that was obtained for the SFS (77.9 %) is related to the
highest number of days that the SFS was rejected, i.e., 19 d of
measurements. For 19 field trips, it was not possible to obtain a minimum of
three snow cores with a ratio between snow core length and the snow cover
thickness
This study made it possible to compare the uncertainty of different methods of estimating SWE. The uncertainty of the studied methods is quantified from two measures of dispersion, i.e., the coefficient of variation (CV) and the uncertainty due to instruments, which leads to different interpretations. For snow samplers, for which several measurements were made every field trip, a calculation of the standard deviation and the coefficient of variation could be done to represent the uncertainty. While only one snow pit was produced each field trip, it was not possible to calculate a coefficient of variation. To allow all methods to be compared with one another, the uncertainty due to instruments was calculated for each method according to the principles of propagation of uncertainties. The CV is evaluated from the repeatability of the measurements that are carried out and therefore is included without distinguishing many random and systematic effects that are associated with uncertainty, such as the effects of snow and weather conditions when data are collected or the bias that is related to the observer who is performing the measurement. We can therefore assume that the CV overestimates the absolute uncertainty of a measurement method. Yet, the uncertainty due to instruments can underestimate the uncertainty of a measurement method (Lindberg, 2000). It only considers the precision of the instruments that are used to calculate the SWE. This corresponds to a theoretical uncertainty that can be calculated without being based upon field measurements but only on the methodology being employed (i.e., instruments and equations being used). This difference may explain why snow samplers, for which the uncertainty was estimated using both methods, had an absolute uncertainty due to instruments lower than the CV. Since the uncertainty due to instruments for snow samplers always has the same absolute value regardless of the snow depth, the SWE needs to exceed 336, 212, and 131 mm for SFS, HQS, and ULS respectively for the uncertainty estimated by the CV to be higher than the uncertainty due to instruments.
From the estimated coefficients of variation, it was possible to compare the
uncertainty of each snow sampler that was used. Similar to López-Moreno
et al. (2020), our results showed that larger diameter samplers, i.e., HQS and
ULS, allowed SWE measurement with lower uncertainty than a small diameter
sampler, i.e., SFS. Although the two large diameter samplers are made of
different materials, their CV values do not show any significant difference. This
result suggests that the choice of aluminum or PVC in the construction of a
snow sampler does not affect the variability in repeated measurements for
large snow samplers. Further, this suggests that the higher CV that was
obtained for SFS is not attributable to the materials that were used in its
design. Two reasons may explain the higher CV for the SFS, namely the low
precision of the spring balance that was used and the smaller diameter of the
sampler. For each measurement that is made with the SFS, there is an
uncertainty of
A portion of the uncertainty of the snow pit can be explained by the
uncertainty of density measurements that are made using the density cutter.
The density cutter that we used (i.e., wedged, volume
For each method of SWE estimation that was used in this study, uncertainty
due to instruments has been calculated. The method with the greatest
uncertainty is the snow pit based upon cumulative layers (methods 1-a and
1-b), with a value of 11.40 %. Although this value is high, recall that
measurements taken to estimate the density of each snow layer are precise.
For a single snow layer, the average uncertainty of the density that is
attributable to the digital scale being used is 3.52 %. This uncertainty
is similar to the HQS and ULS, which have the lowest uncertainties. The high
uncertainty for snow pit methods 1-a and 1-b can be explained by two
factors. First, unlike snow samplers with which the SWE estimate is made from a
single measurement, the snow pit requires numerous measurements in each snow
layer to obtain a SWE value. Although each measurement that is taken
individually has a low uncertainty, the snow pit SWE calculation requires
the sum of SWE for each snow layer, as well as their uncertainties, according
to the statistical principles of propagation of uncertainties. Second, the
absolute uncertainty for the measurement of snow depth is always
For snow pit method 2, average density method, the results have an
uncertainty lower than snow pit methods 1-a and 1-b but higher than those
of the three snow samplers. This method resulted in a lower uncertainty due
to instruments compared to the other snow pit methods for two reasons.
First, method 2 does not use the measurement of the thickness of each snow
layer; rather, it uses total snow depth. Although both methods used the same
tape measure with the same precision (
All results and observations that were made during this study demonstrated
that the ULS was the method that best represented the SWE reference
in the snow conditions of the boreal biome. The results of uncertainties due
to instruments or according to the coefficient of variation show that the
ULS has the lowest uncertainty. The large diameter of its opening makes it
less sensitive to ice layers when taking measurements than when using the
SFS. In addition, the ULS allows a larger snow volume to be collected for
estimating SWE, which suggests that its estimate is more accurate. The
results of the ratio between snow core length and snow depth of 94.9 %
suggest that the entire snow cover was collected in a SWE measurement.
These observations agree with the study by Farnes et al. (1983),
who reported that SWE measurements made with a PVC tube (i.d.
As shown in Fig. 4, mean bias error estimates reveal that almost all other
methods overestimate SWE when compared with the ULS. Sharing characteristics
with ULS, the HQS has the smallest error, with MBE values that are close to
zero. Although the MBE of the SFS suggests that SWE is overestimated, the
box-and-whisker plot in Fig. 4 illustrates high variability, where 23 % of
the results are underestimated compared to ULS. This result is related to
the high uncertainty of the Standard Federal sampler. Although it is
possible under certain conditions to obtain accurate results, its design
generates greater variability in the presence of ice layers, for example,
which makes it less accurate overall than a sampler with a larger diameter.
For the MBE results that were obtained from snow pits, it is possible to
explain the overestimation of these methods by the calculation of the ice
layers. Due to the low precision of the tape measure that was used (
When comparing the SFS to SWE that was obtained by either a volumetric pit or with the Glacier sampler, Farnes et al. (1980, 1983) showed that SFS was overestimated by 10 %, similar to the results that were obtained in this study. Yet, these results are different from those that were obtained by Dixon and Boon (2012), in which the SFS exhibits values that are similar to the SWE reference. In this study, it is the samplers with larger diameter, i.e., the Meteorological Service of Canada sampler (MSC) and SnowHydro, which underestimate the SWE by 6 % to 12 %. We assume that this difference is due to the choice of the SWE reference. In Dixon and Boon (2012), the selected SWE reference was the snow pit, which they consider to be the most accurate method. If we had also chosen the snow pit as a SWE reference, we would have obtained results following the same trend as Dixon and Boon (2012), in which large samplers underestimate SWE. This difference highlights the importance of the choice of the reference for evaluating the error of SWE measurement methods.
While this study takes a different look at the snow pit method, we believe more studies are needed. It would be interesting to compare the different methods employing snow pits from data where several snow pits would be excavated during the same field trip, whereas only one was done in this study. It would be possible to better compare the difference between the cumulative layer method and the average density method, i.e., snow pit methods 1 and 2 respectively. Our results made it possible to answer our second objective, which is to identify the method representing the most appropriate reference of the “true” SWE. Based upon our uncertainty and measurement error results, we believe that large snow samplers are better methods for estimating the “true” SWE than the snow pit in a boreal biome, especially when they are conducted according to the usual methods that we have studied. Our results for the SFS suggest that different jurisdictions using it should consider replacing it. Although large snow samplers require more effort and time in the field to take measurements, they have the considerable advantage of providing a much more consistent and more accurate estimate of SWE than does SFS.
In the context of the boreal biome, which is different from an arctic or alpine environment, the “true” SWE of the snowpack is frequently determined in a snow pit with a non-continuous sampling strategy using a small-sized density cutter. The objective of the study was to compare this method with snow samplers already used in the field (SFS and HQS) and a larger sampler developed for research purposes (ULS). The novelty of the study originates in analyzing the snow pit data at the same level as the samplers instead of considering it as a reference. The snow-pit-based method has been used to measure the SWE of the snowpack, but in the literature, there is no evaluation of its uncertainty and its measurement error at this level. This study made it possible to compare different snow samplers with one another in terms of uncertainty and measurement error. With analyzes that were based on data taken in five consecutive winters, and always in the same specific location and under varying snow conditions, it was possible to quantitatively describe the performance of different SWE estimation methods by reducing environmental and temporal effects as much as possible.
Contrary to literature reports, snow pits are using sampling methods that generate a high SWE measurement uncertainty, based upon small snow samples taken with density cutters, thereby resulting in overestimated and inaccurate SWE values. Although snow density measurements that are taken individually in each snow layer of the snow pit have relatively low uncertainty, the weaknesses of these methods arise from applying the principles of propagation of errors; the summation of numerous measurements that are performed for the many layers constituting the complete snow cover cumulate these individual uncertainties. Although the snow pits measured in this study were based on regional protocols, the conclusions obtained remain relevant and can be applied also to other snow pit protocols. The application of the methodology proposed by this study for the analysis of uncertainty and measurement error could be extended to other methods and areas, and it will help to address the lack of certainty in the literature on what is the most appropriate method of reference for the “true” SWE of the snow cover. While it was not realized in this study, it would be beneficial in future studies to document the measurement error of snow pits made with continuous sampling strategies when used to estimate the SWE of entire snowpack snow cover. Despite its higher uncertainty and measurement error in estimating SWE compared to large-sized samplers, snow pits remain a highly pertinent method for a better understanding of snowpack stratification.
The results that were produced by this study made it possible to reassess the uncertainty and measurement error of SWE measurements that were obtained using the Standard Federal sampler (SFS), in addition to documenting for the first time the performance of the Hydro-Québec sampler (HQS) and Université Laval sampler (ULS), which both use a plate at the ground surface to prevent snow loss. For organizations wishing to evaluate the performance of hydrological models or automatic SWE sensors, the results that have been produced by this study bring a better understanding to the methods that are already in place or which they plan to use. Uncertainty and measurement error results demonstrate that large diameter samplers, such as HQS and ULS, are the best methods for estimating “true” SWE in a boreal biome. Due to the sampling of the large volume of snow at each measurement, an uncertainty less than 5 %, and the ability to take reliable measurements under different snow conditions, large diameter samplers can be used with confidence in obtaining a reference SWE value. Given the great variability in snow conditions present in the cryosphere, it must be considered that using large diameter samplers is environment related. Large diameter samplers will not be as well suited for all environments, like in deep snow conditions, because they are designed for shallower snow cover.
Data are available from the authors upon request.
SJ conceptualized and initiated the study. MBG and SJ carried out field work. MBG performed the data processing and analysis, prepared the figures and tables, and wrote the manuscript. SJ contributed to review and improve the manuscript.
The contact author has declared that neither of the authors has any competing interests.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
We would like to thank all partners supporting the projects that were conducted at the NEIGE site, namely the Ministère du Développement Durable et de la Lutte contre les Changements Climatiques (MDDELCC), Environment and Climate Change Canada (ECCC), Global Water Futures (GWF), and Campbell Scientific. We thank the Forêt Montmorency team for their help with the work on the NEIGE site. We also thank Hydro-Québec, especially Alexandre Vidal and Christian Bouchard, for their participation and the donation of the Hydro-Québec snow sampler (“Super Carottier”), which could be added to the protocol. We thank William F. J. Parsons who did a linguistic revision of this article. We also thank the reviewers and the scientific community for their constructive comments during the peer review process. Finally, special thanks are due to all those individuals who contributed to the collection of numerous high-quality manual snow data over five winters: Cédric Gilbert, Amandine Pierre, Guillaume Arbor, Frédéric Poirier, Charles Villeneuve, Benjamin Bouchard, Kathy Pouliot, and Olivier Ferland.
This research has been supported by the Global Water Futures and Environment and Climate Change Canada (grant no. GCXE21M014).
This paper was edited by Guillaume Chambon and reviewed by Leena Leppänen and one anonymous referee.