The main prerequisite to observe capillary barriers in initially dry snow is a finer-over-coarser profile in layering. For this purpose, three combinations of grain size were considered here: (1) FC, i.e., fine-over-coarse snow; (2) FM, i.e., fine-over-medium snow; (3) MC, i.e., medium-over-coarse snow. We classify snow with 0.25 mm ≤ gS ≤ 0.5 mm as fine, snow with 1 mm ≤ gS ≤ 1.4 mm as medium, and snow with 2 mm ≤ gS ≤ 2.8 mm as coarse. Note that this nomenclature is convenient for the presented work, but it is not consistent with the international classification proposed by , which defines for instance medium snow grain size as 0.5 mm ≤ gS ≤ 1 mm.

Experimental evidence revealed that the area occupied by fingers in snow and the value of ψWE may be both functions of water input rate . In order for our conclusions to be more general, we carried out experiments with three different water inputs W: these are ∼ 10, 30 and 100 mm h-1. These water input rates are a compromise between the need for exploring the properties of capillary barriers over a broad range of W, expected melt rates in natural conditions , and operational constraints (specifically the expected duration of the tests). Because the saturated conductivity of snow is rather high compared with the chosen input rates, most existing instability criteria will predict unstable flow in these conditions. Accordingly, have already observed preferential infiltration in snow with average gS between 0.421 and 1.439 mm for different water input rates (21.7 mm h-1 ≤ W ≤ 205.5 mm h-1).

Nine samples were prepared in a cold room at -20 ∘C using refrozen melt forms (one sample for each of the three grain size combinations and three water input rates). Henceforth, numbers 1, 2, and 3 differentiate samples with the same grain size combination but subjected to different water input rate (10, 30, and 100 mm h-1, respectively). Fragmented snow particles were firstly partitioned in several classes of grain size. Afterwards, the three gS chosen were sieved a second time to prepare the samples. Snow was packed in a cylindrical container. The container was composed by a number of acrylic rings (height equal to 20 mm, diameter equal to 50 mm) that were previously taped on the external side. After sieving the lower layer, its dry density (ρD,L) was measured by gravimetry. The dry density of the upper layer (ρD,U) was measured by gravimetry at the end of sieving operations (by considering the difference between sample total weight and sample weight before sieving the upper layer). After preparation, each sample was moved to a second cold room at 0 ∘C, where it was stored for at least 12 h to reach initial conditions of dry snow at 0 ∘C.

We report in Table  the details of each experiment. Water input rates are reported both in mm h-1 and in g min-1 (samples diameter equal to 5 cm). The coefficients of variation of ρD,U and ρD,L read 0.06 and 0.03. We did not apply any tamping during sieving operations so we had no direct control on the values of ρD,U and ρD,L. Given the low variability of these two variables, we point out that this work investigates how capillary barrier effects and associated preferential flow vary with grain size only. Future investigations should focus on the generalization of this work to layers of different density. Some samples (namely, FC2, FM2, and MC2) are shorter than the others. However, the thickness of the upper layer is the same for all the samples. This is important as ponding occurs in the upper layer.

Before starting each experiment, we placed a thin cotton ring on the top of the sample to enable the point source of the tracer to spread over the surface of the upper layer. Then, each experiment was started by supplying dyed water into samples using a micro-tube pump. The dye used was blue ink, diluted by a factor of 10 in water at 0 ∘C. We monitored W during each experiment by automatically measuring the weight of the tracer reservoir (1 min resolution). Absolute relative differences between experimental (Table ) and reference (10, 30, and 100 mm h-1) values of W range between 6 and 19 % as it is difficult to apply a constant, low input rate.

When the tracer reached the base of each sample, tracer supply was stopped. The arrival time of the tracer at sample base (tt) was registered with a manual chronometer watch by visually inspecting samples during the experiments. Since samples had different heights, we define a specific travel time as τ=tt/h, with h equal to sample height. Note that τ is in min cm-1 as it is the reciprocal of velocity. After each experiment, pictures of the external sides of the sample were taken to estimate the approximate thickness of the upper layer marked by liquid water accumulation (p, in cm). Soon afterwards, we took pictures of the top section of each acrylic ring (by gradually removing them from the column, snow included). At the same time, the liquid water mass w, in grams, in each of the rings was measured using a portable calorimeter . These measurements were translated into profiles of volumetric liquid water content by converting w to θ. Fractions of wet areas over total area (f) were also estimated for each section by manually delimiting fingers in all the pictures taken and calculating their extension using the ImageJ software (http://imagej.nih.gov/ij/, v. 1.48; see ).

We simulated the dynamics of each sample using the physically based 1-D multi-layer snow cover model SNOWPACK . These simulations aim at comparing observations of capillary barrier development with predictions by a physically based model, as previously done by, e.g., , , and mainly using field observations. The relatively high resolution of LWC measurements (2 cm) enables a rather detailed discussion of both the physical process and its simulation by a physically based model. This comparison will not include preferential flow patterns and arrival times, as the model does not include an explicit treatment of preferential flow regimes.

The model discretizes snow using a finite element grid. It simulates the evolution in time of a broad set of variables along a vertical profile of snow starting from external forcings. The original version of SNOWPACK considers a bucket-type approach to simulate water percolation in snow. Accordingly, water is retained at a given position in the profile until it exceeds a threshold (see ). After exceeding, excess water is transmitted downwards. have introduced in SNOWPACK a water transport model based on the model by and on an equilibrium approximation of water flow to tackle numerical instability (see for details). Recently, have also introduced a discretization of the Richards equation that significantly improves several aspects of liquid water content simulation in snow. We used the numerical scheme by in this paper (WE, SNOWPACK version 3.3, https://models.slf.ch/).

The initial spatial resolution of simulations was set to 2 cm. The time step was set to 1 min, but the numerical scheme by reduces this initial time step basing on an iteration rule (see for details). Snow initial conditions were chosen to replicate the grain type, size, and density (Table ), θ (initially dry), and temperature (0 ∘C) of the physical samples. Using the same sieves that we used here, obtained a median grain size (hereinafter, g¯S) for the class 0.25–0.5 mm and the class 1–1.4 mm equal to 0.406 and 1.463 mm, respectively. g¯S for medium snow is greater than the upper boundary of the sieve probably because snow grains used were not perfectly spherical. In the simulation, we therefore set g¯S = 0.406 mm for fine snow, g¯S = 1.463 mm for medium snow, and g¯S = 2.926 mm for coarse snow (by assuming this last value as 2 times the median medium grain size). Bond size was assumed equal to one-third of grain radius. Input data were chosen to replicate experimental conditions in the cold chamber, i.e., a constant precipitation flux (equal to the measured water input flux W, see Table ) and a fixed air temperature of 0∘C. The threshold temperature for classifying solid and liquid events was set to -0.01∘C, in order for W to be classified as liquid. Wind speed and solar radiation were set to 0, while incoming longwave radiation was calculated as σT4, where σ=5.67×10-8 W m-2 K-4 and T=273.15 K. Parametrizations for snow permeability, unsaturated hydraulic conductivity, water retention curve, residual water content, and averaging method for hydraulic conductivity at the interface were all kept at the default settings discussed for SNOWPACK in .

Particular attention is paid to the comparison between observed and simulated LWC peak over the interface between layers, as this is an important variable involved in capillary barrier formation and in wet snow avalanche forecasting . Another key feature of capillary barriers is the vertical profile in LWC . However, choosing a single snapshot of simulated LWC for the comparison with our observations is problematic, as the model is 1-D and, at this stage, does not include an explicit treatment of preferential flow patterns. These are expected to play a key role in water flow around capillary barriers as water concentrates in fingers that are characterized by a higher-than-average unsaturated hydraulic conductivity (due to a higher-than-average LWC). It follows that restricting this comparison to a single profile (i.e., only one time step) may be misleading as possible differences between observations and simulations might be due to a process that is currently not treated by the model. This may limit a comparison aiming at assessing the capability of a model to reproduce capillary barriers. Thus, we will compare observed profiles of LWC with model results at two different times. The first one (WE1) is the observed arrival time of water at sample base; the second one (WE2) is the simulated arrival time of water at sample base, which is chosen by identifying the instant when θ at sample base reaches ∼ 3 vol % in the simulation . Note that WE1 and WE2 for each sample were obtained from the same simulation.

On the one hand, WE1 is advantageous as supplied mass in experiments and in simulations matches because both profiles refer to the same time step. On the other hand, flow in the lower layer will be at the beginning spatially variable, strongly accelerated, and highly fingered , which are all features that are not explicitly included in the 1-D model and that might hamper the application of the Richards equation. Thus, when considering WE1, we will focus on the profile over the interface, where the peak of LWC develops. Conversely, WE2 enables a comparison of a full profile of LWC, but the simulated mass of liquid water will be greater than observed due to the possible mismatch between observed and expected arrival time of water in simulations . This is particularly evident in the upper layer of FC and FM samples, as water speed in fine snow during matrix flow is slow. Thus, when considering WE2, we will focus on the profile in the lower layer. Because liquid water flow in MC samples turned out to be highly fingered (see next section), observations in these samples are compared only with WE2 (full profile), which is probably less affected by effects due to preferential flow, including the time arrival of water at a certain point of the profile. Note that both model version and the evaluation methodology are different from the preliminary results reported in the discussion paper .

Figure  reports the horizontal sections of all the samples (2 cm vertical resolution) at the end of the experiments (i.e., when dyed water arrived at sample base). Dyed water is visible as blue stains. Generally, the darker the color is, the greater local LWC is . We report in Fig.  three examples of samples at the end of the experiment. These are FC2 (as an example of FC tests), FM2 (as an example of FM experiments), and MC1 (as an example of MC experiments).

Table  reports observations in terms of thickness of the upper layer marked by liquid water accumulation (p), arrival time of water at the base of each sample (tt), and specific travel time of water in snow (τ). In Figs.  and , profiles of wet snow fractions f and LWC are given. In Fig. , each point represents bulk LWC in the underlying 2 cm. As an example, any value reported at a depth equal to 8 cm is bulk LWC between 8 and 10 cm. This represents the LWC measured immediately over the interface between layers.

Figure  compares observed and SNOWPACK-based profiles of volumetric LWC for each sample. Each point represents bulk LWC in the underlying 2 cm. As described in Sect. , two simulated profiles are reported for FC and FM samples (WE1 and WE2), whereas only WE2 is reported for MC samples. Dashed lines indicate the parts of the simulated profiles that are not considered in the discussion (see again Sect. ).

Figure  confirms that liquid water movement through a finer-over-coarser snow texture is subjected to ponding and horizontal diversion of water when the wetting front comes to the textural interface. In FC and FM samples, horizontal spreading of water at the interface introduces a clear textural transition in wetness between finer and coarser layers (see Fig. ). In four out of six samples of these two classes, a homogeneously blue area is observed even at a depth equal to 8 cm (i.e., 2 cm above the interface). MC samples show a more variable behavior. Indeed, almost no water spreading was observed for MC1 (see also Fig. ), whereas marked horizontal redistribution of water is visible in MC2 and MC3. MC samples show a smaller p than FC and FM samples.

The difference between FC–FM and MC samples in terms of ponding behavior may be explained considering the retention properties of snow with different grain size. ψWE for medium and coarse snow can be estimated from grain size using the relation reported in and : ψWE∼0.025 m in coarse snow and ∼0.04 m in medium snow. In contrast, ψ in fine snow for 5 and 10 % LWC is ∼ 0.22 and 0.21 m, respectively, and ψ in medium snow for 5 and 10 % LWC is ∼ 0.09 and 0.08 m, respectively. This implies that for typical low saturation values in snow, the difference between the suction pressure in the finer snow and the water entry pressure of the coarser snow is larger for fine snow. Furthermore, unsaturated conductivity of coarser snow is likely to quickly decrease with increasing ψ, as already observed in soils . It follows that, at the same W, the greater the mismatch between unsaturated properties of finer and coarser snow is, the larger the mass of water accumulated and horizontally diverted until the underlying layer is able to convey the supplied flux. These approximate values of suction in snow were estimated using the WRC parametrization in snow by , assuming a residual LWC equal to 2.4 vol % and considering 5 and 10 vol % as reference values for relatively low saturation. Note that the WRC by refers to a drying process and this may cause some additional uncertainty when estimating ψ for a wetting process, due to hysteresis. Indeed, ψ for a primary wetting process is expected to be smaller than ψ for a primary drying process. Available data of hysteresis in snow are, however, very preliminary . A specific discussion about the role of hysteresis for interpreting these results is given in Sect. .

All layering types are characterized by similar LWC profiles with the only difference in absolute values for LWC (Fig. ). LWC increases with depth in the upper layer, it presents a marked peak at the textural boundary, and it decreases again below the capillary barrier. Peaks in LWC at the interface may be associated with capillary barriers as water ponds until ψ reaches ψWE and the underlying layer becomes conductive : similar examples are reported or discussed in, e.g., , , , , and . All FC–FM samples yield a similar LWC in the upper layer at the interface: ∼ 33 vol % in FC samples and ∼ 34–36 vol % in FM samples. Again, this may be explained by considering that the peak of LWC at the boundary is ruled by the retention properties of snow with different grain size and by their contrast (see for a similar discussion in soils). Note that such values of LWC are much greater than those usually reported in field profiles . LWC drives, among others, snow settling and wet snow metamorphism . Both processes experience a dramatic acceleration with increasing LWC. This supports the idea that capillary barriers may play an essential role for snow stability in wet conditions .

In MC samples, a smaller, but distinct, peak in LWC was measured at the interface: in MC1, LWC over the boundary is ∼ 4.5 vol %, whereas LWC values immediately above and below are 2.7 and 1.7 vol %, respectively. In MC2 and MC3, the peak in LWC over the interface is ∼9 vol %. A smaller peak of LWC in MC layering is attributed to the small difference between unsaturated hydraulic properties in medium and coarse snow (for example, a smaller difference between ψ in medium snow and ψWE in coarse snow). Note again that this difference could be even smaller than expected if hysteresis were explicitly taken into account . The observed peaks in MC samples are generally greater than the peak value observed by during snowmelt infiltration through a 1.5 over 2.5 mm transition in an artificially sieved snowpack. This difference may be due to (microstructural) heterogeneity in snow, infiltration rate, experiment durations, and the larger measurement area of the TDR system used by compared to a calorimeter (see Sects.  and ).

The occurrence of capillary barrier causes horizontal redistribution of water. Thus, spatial homogenization of liquid water patterns at the interface is promoted. Indeed, f increases with depth over the boundary (where f=1 for all FC and FM samples; see Fig. ). However, sections in Fig.  reveal a remarkable spatial variability of this process at centimeter scale. For example, some pockets of dry snow persist at depths equal to 8 cm in samples FC2 and FC3 (i.e., 2 cm above the interface). Other indications are the observed spatial variability of p in each sample (Table ) and the differences of coloring in some sections at depths equal to 8 or 10 cm (e.g., FC2, FC3, and FM3), which can be linked to differences in LWC . Isolated clusters of liquid water surrounded by dry snow are also visible in MC1 and MC2. All these observations show that the distribution of liquid water above a capillary barrier has a marked 2-D (or even 3-D) structure at local scale, probably due to heterogeneity in snow microstructure. The difference in wet areas in MC samples for different water input rates might be on the contrary an effect of water input rate on ψWE, as observed in snow by .

Observed profiles of f suggest that water movement in samples was marked by high spatial variability and that this variability is lower in fine snow layers than in medium or coarse snow. Overall, preferential flow turns out as the predominant pattern of water infiltration in snow . We observed that new fingers created during the percolation, and that sometimes fingers stopped their vertical percolation at some locations but continued to develop at others. The movement of fingers in the lower layer was very rapid and represented a small fraction of the total duration of each experiment (typically in the range of minutes).

report that the total area of preferential flow (hence f) in snow samples made by vertically homogeneous snow decreases with increasing grain size but increases with increasing input flux. We also observed a decrease in f with increasing gS, whereas a clear increase of f with increasing W was detected only for MC samples. On the one hand, the expected dependency of f on sublayer unsaturated conductivity and the possible relation between ψWE and velocity support the existence of a relation between f and W (albeit both effects have been mainly observed only in soils). On the other hand, these experiments included a capillary barrier, contrary to experiments by , and this represents a major driver of liquid water content patterns at a more local scale (see the previous Section). Accordingly, water speed in the upper layer was locally affected by ponding, whereas inflow rate in the sublayer was driven by breakthrough of water when reaching ψWE. Both processes limit the impact of external water input rate on f, at least until steady conditions are reached. In MC samples, the difference in retention properties between layers is lower; thus, the effect of a capillary barrier is spatially very localized. This may explain why observations in MC samples agree with previous observations in homogeneous snow.

Considering outcomes of different experiments in sand, reports that finger width might increase with both very high (i.e., close to saturated conductivity) and very low supplied fluxes, while finger width keeps constant for a broad range of supplied flux in between. Water input rates in our experiments span 11 and 113 mm h-1; these values are very small when compared with expected values of saturated conductivity in snow (∼ 104–106 mm h-1; see ). We therefore suggest that future developments of this work should investigate the relation between flux and f extensively, i.e., enlarging the range of W considered during the experiments and/or reaching steady conditions. Furthermore, additional experiments should be carried out using containers of different size in order to assess whether the experimental geometry used may induce possible boundary effects (see also on this).

τ increases with decreasing W, as clearly expected (Table ). In the case of FM2 (fine-over-medium snow, W=27.7 mm h-1), we can compare the τ measured during the experiment (2.2 min cm-1) with the τ observed during the experiment by , since this is the only gS-W combination that these two works share. The τ measured by for fine snow and W=22.3 mm h-1 is equal to 1.7 min cm-1, while the τ for medium snow and W=21.7 mm h-1 is equal to 0.7 min cm-1. These results suggest that τ in a FM sample is higher than the τ observed in a homogeneous sample composed by medium snow. This is clearly expected since permeability in medium snow is higher than in fine snow. However, this τ is even higher than the specific travel time observed in a homogeneous sample made by fine snow, which is marked by a low saturated conductivity. This comparison helps to quantify the relevance of capillary effects in ruling water speed in snow and the arrival time of meltwater at snow base.

According to Fig. , SNOWPACK clearly reproduces an increasing LWC with depth in the upper layer and a peak of LWC at the interface at WE1. Furthermore, LWC profiles below the barrier are generally in good agreement, once water has reached the base in the simulation (WE2). Point differences between observed and simulated LWC at the interface at WE1 read ∼ 2–5 vol % in FC1-FC2, ∼ 3–8 vol % in FM1-FM2, and ∼ 0.1–5 vol % in MC samples. A larger difference (∼ 9–13 vol %) is found for higher input rates. However, note that FC3 and FM3 were subjected to an extremely high water input rate compared with natural conditions.

Previous evaluations of SNOWPACK already show that the inclusion of the Darcy–Buckingham equation in snow enables a correct prediction of the onset of capillary barriers at textural discontinuities . Here, we enlarged previous findings by considering a broad set of snow textures and input rates and a relatively high resolution of measurements. Note that the version of SNOWPACK used does not implement a parametrization of water entry suction , but the model is anyway able to provide a sufficiently good performance in reproducing the profile around a capillary barrier. Results by , , , and show that ψ and θ at an infiltrating front (hence, ψWE) may follow a wetting WRC. Both variables are also strictly coupled with unsaturated conductivity and the impedance mismatch given by the low unsaturated conductivity of the coarser layer compared to the applied flux plays a key role in delaying water on the barrier. SNOWPACK currently includes a parametrization of both a WRC and unsaturated conductivity and solves the Richards equation. It may be that implementing the Richards equation in 1-D is sufficient to mimic some essential features of capillary barriers in snow (e.g., ponding) even without an explicit calculation of ψWE. Note that an increase in LWC with depth as well as abrupt transitions in LWC at the interface may occur even in equilibrium, i.e., when suction increases with height. This is due to the different retention properties of fine, medium, and coarse snow. Furthermore, suction profiles over the barrier may depend on applied flux too . Thus, a more detailed analysis of capillary barrier dynamics in snow necessarily needs observations of suction profiles during infiltration; this represents an important step of future research.

Another important limitation for this discussion may be the present lack of an exhaustive investigation of WRC hysteresis in snow . As already noted, the absence of a proper parametrization of hysteresis may hamper the estimation of expected LWC at ψWE during ponding (i.e., wetting), if different WRCs for a wetting and a drying process are not known. Also, the hysteretic behavior of the WRC is considered an important factor in driving preferential flow in general, since for example it promotes the persistence of fingers in soils . The magnitude of hysteresis is also related with the magnitude of capillary and saturation overshoot , although it is not the prime cause of instability . In this context, note that, according to , hysteresis plays a less important role than the difference in unsaturated hydraulic properties between the finer and the coarser layer when studying the general properties of capillary barriers in soils and how they depend on layer parameters; this may again support the idea that the existing implementation of unsaturated flow in a complex 1-D model may be sufficient to mimic LWC distribution around a capillary barrier. A possible improvement may be represented by the set of parametric models proposed by for porous media, which includes hysteresis.

Observations show that both breakthrough of liquid water below a capillary barrier and wet conditions in the upper layer or in fingers may present a high spatial variability at centimeter scale. This is because natural snow is spatially heterogeneous and this may affect 3-D patterns of capillary barriers (e.g., see the already discussed pockets of dry snow in FM3). Alternation of dry and wet snow can sensibly decrease the bulk LWC in a ring, although local LWC can still be very high. This may partially explain some differences between observations and 1-D simulations. For example, predicted peak LWC in FC3 and FM3 is ∼ 43–46 vol %, which is close to saturated conditions (the porosity of fine snow in both samples is ∼ 0.5) but greater than observations. An approximate estimation of LWC at ψWE in fine snow (obtained assuming continuity of suction at the interface) reads 50 vol %, which is closer to SNOWPACK simulations than data. Thus, saturated conditions might be reached at a very local scale, while bulk LWC in each ring can be lower due to heterogeneity in wetness at a larger scale. Another example is water flow below the interface, which showed a high degree of spatial variability. The good agreement between the model and the data (at WE2) might suggest that at this measurement resolution differences in LWC between a highly channeled flow and a matrix-only simulation balance; that is, fingers are usually highly saturated but occupy only a small fraction of total volume. Thus, the average LWC at ring scale is much lower than saturation and close to matrix conditions.

This result suggests that an exhaustive process understanding of the physics of capillary barriers in snow and water flow instability may need that a proper measurement and/or modeling scale are established to clearly separate model–data significant discrepancies and effects due to the sampling strategy (see for a definition). Importantly, the spatial resolution needed to capture 3-D patterns of capillary barriers might be smaller than that usually used to sample LWC in the field (see again Fig. ). Increasing the spatial resolution of LWC measurements is challenging as measuring LWC in snow is still marked by high uncertainties . It is only recently that undisturbed, non-destructive, and repetitive measurements of LWC have been obtained . A promising alternative might be given by pore-scale measurements of liquid water flow .

This discussion also reveals the role played by heterogeneity in introducing possible differences in LWC between 3-D (bulk) and 1-D conditions. Additional uncertainty in this comparison may be caused by instrumental precision (see next Section), ambiguity in the identification of the correct snapshot of LWC for this comparison, possible air trapped in voids at saturation , and possible boundary effects due to the experimental geometry. To summarize, we suggest that additional investigations should be carried out to establish proper frameworks for the high-resolution comparison of complex models and laboratory (or field) observations.

A mass balance between supplied and measured liquid water mass reveals that the measured mass ranges between 93 and 176 % of supplied mass in eight out of nine samples, while in MC1 measured mass is 434 % of supplied mass. Note that in this last sample the total mass supplied is nonetheless very small due to the short duration of this experiment (∼ 2.88 g).

This discrepancy can be explained by instrumental noise. Melting calorimetry has been widely used to measure LWC for decades , but points out that this method may be inaccurate as it implies the calculation of a difference between large numbers . According to , absolute errors in measuring LWC using calorimetry span 1 and 5 %. The instrument we used here (the so-called Endo-type snow-water content meter) was proposed by . They note that measured LWC span ±2 % of known LWC (by weight) in 87 % of the cases. By comparing measurements by the Endo-type calorimeter with those by a dielectric device in snow pits (see for details), they note that this device returns alternatively higher or lower LWC if compared with high and low readings by the dielectric device.

We estimated an absolute error for these experiments by comparing the height-integrated LWC measured using calorimetry within each sample with the ratio between supplied liquid water volume and total volume of samples. The absolute difference spans 0.8 and 2.97 vol %, thus it is consistent with the literature . Measured and simulated LWC by SNOWPACK are also in fair agreement (see previous section) which underlines the above mentioned range of absolute error, since SNOWPACK bases an energy conservation on mass.

Capillary barriers and associated preferential flow represent a large challenge for LWC measurements. On the one hand, peaks in LWC at the interface are rather high and this is a problem for those instruments that may lose accuracy for high LWC, such as a snow fork . On the other hand, bulk LWC in fingered snow may be very low, as water accelerates and occupies a small fraction of total volume. This means that such experiments need an instrument that guarantees a comparable performance for both high and low LWC. This may represent a benefit of the Endo calorimeter (see Fig. 4 in ), which seems also appropriate given the small dimension of each ring. Furthermore, report that the absolute error in measuring water content using dielectric methods spans 0.2 and 0.9 vol %, while note that the expected difference between measurements taken using a Denoth meter and a snow fork is ∼ 1 vol %. Thus, measuring low LWC is generally very challenging for several existing techniques. Finally, a highly fingered flow may be missed and/or disturbed by using larger instruments . For future experiments, we will also consider alternative portable techniques, such as a dilution method .