2.1 Study area

The Rikha Samba Glacier is a south-east-orientated, medium-sized glacier (5.5 km²) located in the Hidden Valley in western Nepal (Fig. 1a). The glacier was about 5.5 km long with an elevation ranging from 5420 to 6440 m a.s.l. in 2014. The glacier meltwater contributes to the Kali Gandaki Basin of the Ganges River. The Hidden Valley falls under a rain shadow climate, receiving the least precipitation in Nepal with an annual precipitation of 370 mm (Fujita and Nuimura, 2011). The annual mean temperature measured with an automatic weather station (AWS) in the vicinity of the glacier (Fig. 1a) was −5 ^∘C in 2014. The glacier was first visited in 1974 and it had been losing mass at a mean rate of −0.5 m w.e. yr⁻¹ between 1974 and 2010 (Fujita et al., 1997; Fujita and Nuimura, 2011).

Figure 1(a) Location and (b) map of Rikha Samba Glacier in the Hidden Valley catchment in Nepal. Radar tracks in 2015 (red dots), radar point measurements in 2010 (blue circles) and location of the thermistor chain (red diamond) are shown in (b). Background image of (a) is from Landsat 5 on 25 May 2010. Background image of (b) is from Google Earth (image © 2020 Maxar Technologies).

2.2 Ground-penetrating radar (GPR)

We used a Malå GeoScience ProEx impulse-type ground-penetrating radar (GPR) with a 30 MHz rough terrain antenna (RTA) with a frequency band from 15 to 45 MHz to measure the ice thickness and thermal regime of the Rikha Samba Glacier in 2015 (Fig. 1b). The RTA is an unshielded radar antenna in which the transmitter and receiver elements are mounted in a row in a flexible tube, making it suitable for rough terrain such as glacier surfaces.

DC offset was subtracted from the data, and a gain function as well as a bandpass Butterworth filter removing frequencies below 15 MHz and above 45 MHz were applied to the continuous radar profiles. The GPR reflectors from the assumed bedrock surface were manually picked from the data. Picked two-way travel times of the radar signal were converted to ice thickness using a wave velocity of 168 m µs⁻¹ (Robin, 1975). Strong scattering of the radar signal within the ice was interpreted as temperate, ice whereas ice without internal reflectors was classified as cold ice, and the interface between these zones was also manually picked in the data. In addition, we have used point data collected in 2010 with an impulse GPR transmitter (Ohio State University) with a set of half-wavelength 5 MHz dipole antennae (Fig. 1b). These data were used only for ice thickness. The time increment of 5 years between the GPR measurements was corrected by projecting the 2010 data to 2015. This was done by assuming that the glacier was thinning at the same rate between 2010 and 2015 as the long-term thinning rate for 100 m elevation bands obtained for a 12-year period between 1998 and 2010 (Fujita and Nuimura, 2011).

2.3 Glacier geometry and crevasse location

A digital elevation model (DEM) of 30 m resolution was generated for the Rikha Samba Glacier from Pleiades tri-stereo satellite imagery (0.5 m resolution) for 7 November 2014. Crevassed areas on the glacier were visually identified from the imagery and refined using Google Earth imagery (WorldView, 21 September 2012) in which the crevasses were more visible. The ice volume and bedrock topography were initially estimated by defining ice thicknesses as zero at the margins of the glacier and interpolating the ice thickness data measured with the GPR. For interpolation, we assumed a spherical semi-variogram (fitted to data with a sill of 3743 m² and a range of 784 m) and applied the kriging algorithm (Matheron, 1963). This method is widely used to interpolate ice thicknesses measured with a GPR to estimate the volumes of mountain glaciers (e.g. Fischer, 2009). Since the GPR data do not cover the entire glacier, they result in high uncertainties in the interpolated bedrock topography in some parts of the glacier. The initial bedrock topography is thus corrected using the ice flow model (Sect. 3.4.2).

2.4 Glacier mass balance, surface velocities and ice temperature

We constrained the surface mass balance model from stake network measurements in 2012 and 2013 (Gurung et al., 2016) and from the total volume change estimated by a geodetic survey (Fujita and Nuimura, 2011) over the period 1974–1994 (−0.57 m w.e. a⁻¹) and 1998–2010 (−0.48 m w.e. a⁻¹). Stakes displacement monitored during the 1998–1999 period shows horizontal surface velocities between 9 and 24 m a⁻¹ (Fujita et al., 2001). By comparing those velocities with the deformation velocities computed from shallow ice approximation and realistic flow rate factors, Fujita et al. (2001) concluded that observed velocities are greater than what could be explained by ice deformation alone, which thus implies basal sliding. The ice temperature at 10 m depth was measured with a thermistor chain manufactured by Stum Bohr AG Switzerland in the lower ablation area (5600 m a.s.l., Fig. 1b) for 2014–2015. The temperature was logged over a 1-year period in 6 h intervals. The air temperature and precipitation were observed with an AWS in the vicinity of the glacier at 5320 m a.s.l. (Fig. 1a).

2.5 Meteorological parameters

We used the ERA-Interim reanalysis data (Dee et al., 2011) at a daily timescale over the period 1980–2016 as input data for the mass balance model. We only used air temperature data and assumed a constant precipitation rate in time (no precipitation seasonality) to avoid complexity in the simulation. Temperature and precipitation were then distributed on the glacier according to altitudinal gradients to reproduce the observed mass balances. Bias in the ERA-Interim air temperature was corrected using a linear function determined from the comparison with the local AWS data (Fig. 1b) over the period 2011–2015.

3.1 Surface mass balance model

Mass balance is modelled using a degree-day method following Gilbert et al. (2016) in which we include the influence of the spatial variability of the shortwave radiation. The net annual surface mass balance is determined by

where B is the net annual surface mass balance (m w.e. a⁻¹), C is the annual snow accumulation, R is the rate of refreezing and M is the annual melting. Meltwater is computed from the sum of two components following Pellicciotti et al. (2005):

where m is the daily melt (m w.e. d⁻¹), T_a is the air temperature (K), T_th is a temperature threshold for melting (K), f_m is the melting factor (m w.e. K⁻¹ d⁻¹), S_pot is the potential solar radiation (W m⁻²) and f_rad is the radiative melting factor (m w.e. W⁻¹ m² d⁻¹). Following a similar approach as used in Gilbert et al. (2016), f_rad is computed from the radiative melting factors for snow and ice (fradsnow and fradice) and the ratio of annual accumulation to melt (r_s∕m) with

The annual ratio r_s∕m is computed assuming that

The annual snow accumulation (C) and the annual amount of melting (M) are computed with f_rad equal to fradsnow. Snow accumulation is calculated as a function of elevation (z, m a.s.l.) with

where P_ref is the annual precipitation rate (m w.e. a⁻¹) at the elevation zpref (m a.s.l.), A_P is the altitudinal precipitation factor (m⁻¹), z is the elevation (m a.s.l.) and T_snow is a temperature threshold that distinguishes between snow and rain (K). Assuming that refreezing in the previous year creates impermeable layers and thus occurs only to a depth equal to the annual accumulation rate, we write

where f_r is a refreezing factor. The determination of the parameter values is described in Sect. 4.2.

3.2 Thermo-mechanical model

The ice flow model solves the Stokes equations for incompressible flow adopting Glen's flow law for viscous isotropic ice (Cuffey and Paterson, 2010) and is coupled to an energy conservation equation using the enthalpy formulation (Aschwanden et al., 2012; Gilbert et al., 2014a)

where ρ is the firn or ice density (kg m⁻³), t is the time (s), H is the enthalpy (J kg⁻¹), v is the ice velocity vector, κ is the enthalpy diffusivity (kg m⁻¹ s⁻¹), Q_lat is the heating rate coming from meltwater refreezing (W m⁻³), and tr(σϵ˙) is the strain heating (W m⁻³) with σ and ϵ˙ respectively being the stress and strain rate tensors. A basal heat flux of 4.0×10-2 W m⁻² is assumed for the basal boundary condition. This value is not well-constrained, ranging from 2.0×10-2 W m⁻² (observed at Rongbuk Glacier in the Everest region; Zhang et al., 2013) to 8.0×10-2 W m⁻² as predicted by a large-scale model (Tao and Shen, 2008). The enthalpy is written in terms of ice temperature T_i (K) and water content ω with

where C_p is the heat capacity of ice (2.05×10³ J K⁻¹ kg⁻¹), T₀ is the reference temperature for enthalpy (set to 200 K), T_m is the melting point temperature (K), L is the latent heat of fusion (3.34×10⁵ J kg⁻¹) and p is the ice pressure (Pa). Changes in the glacier surface elevation are computed by solving a free surface equation (Gilbert et al., 2014a). We adopt a linear friction law as a basal boundary condition for the Stokes equation with

where τ_b is the basal shear stress (MPa), u_s is the sliding velocity (m a⁻¹) and β is the friction coefficient (MPa a m⁻¹).

The model is solved using the finite-element software Elmer/Ice (Gagliardini et al., 2013) on a 3D mesh with a 50 m horizontal resolution and 15 vertical layers, giving about 10 m of vertical resolution. The mesh is constructed from a 2D triangular linear element extruded in the vertical direction. It gives a mesh made of a triangular prism unstructured in the horizontal direction and structured in the vertical direction.

3.3 Modelling crevasse influence via water percolation

In order to determine the areas where crevasses are likely to form on the glacier, we compute the maximal principal Cauchy stress σ_I (MPa) at the glacier surface from the stress tensor. We compare it with a threshold value σ_th (MPa) to identify where damage production occurs (Pralong and Funk, 2005; Krug et al., 2014) and define the crevassed areas where σ_I>σ_th. Crevasses are commonly considered to be the place where surface meltwater can enter the body of the glacier (Fountain and Walder, 1998). The development of a hydrologically driven crack network initiated by water-filled crevasses seems to be able to transport liquid water down to the bedrock even through cold ice (Boon and Sharp, 2003; Van Der Veen, 2007). In our model, we therefore make an assumption of free vertical percolation of the meltwater down to the bedrock in which local surface meltwater in the crevassed area is the only source of liquid water percolating into the crevasses. This means that we leave out any water coming from the surface runoff outside of the crevassed area and draining into the crevassed area. Assuming that water refreezes in the first cold layer, we compute a latent heat heating rate Q_lat (see Eq. 7) from the available annual meltwater and the ice temperature of the current iteration. At each vertical layer i of the model, Q_lat is computed from the amount of refreezing water r_i each year (kg m⁻²) with

where dz_i is the thickness of the layer i (m) and dt is the time step (s). The amount of refreezing water r_i is distributed from top to bottom of the glacier with the condition that the enthalpy for a certain layer has first reached the enthalpy of fusion H_f before the water can access the next layer downwards. Starting from the surface melt r₁=m, the amount of liquid water available for refreezing in the next layer r_i+1 is computed following

In adopting this approach, we assume that all the energy used to melt ice at the surface in a crevassed area is released into the deeper ice body. It can be seen as an energy conservation approach rather than the modelling of water routing through crevasses.

3.4 Strategy for steady-state glacier

3.4.1 Enthalpy surface boundary condition including firn and snow influence

For this study, we develop a new method in order to determine surface boundary conditions for enthalpy. We use the 1D semi-parameterised approach developed in Gilbert et al. (2014b) and distribute it over the entire glacier. The method takes into account water percolation and refreezing in both firn and seasonal snow to determine the adequate surface boundary condition for the 3D model (Gilbert et al., 2012). The 1D model is solved on a one-dimensional 10 m depth vertical profile at each surface node of the 3D model independently of all other surface nodes. It allows a higher temporal and vertical spatial resolution in order to explicitly solve percolation and refreezing processes in the snow and firn. Parameters of the model are shown in Table 1.

Starting from an initial uniform temperature profile, firn and ice temperature is solved at daily time steps along the vertical profile with a 0.06 m resolution. The vertical grid is updated as the surface elevation evolves and all the variables are interpolated from the old grid to the new one at each time step. The 1D model is forced by air temperature and by the surface mass balance model that provides snow accumulation and surface melting for the corresponding surface node. To compute the steady-state condition, the model is driven by a mean annual cycle of air temperature which is determined at the daily timescale from the meteorological data. A Gaussian random noise is added to the computed mean annual temperature cycle to plausibly represent the daily temperature variability. The standard deviation of the Gaussian function is adjusted to match the number of positive degree days in our mean annual cycle to the mean number in the data in order to conserve the amount of modelled melting. It provides a strong constraint on this parameter, making our results independent of this choice. The 1D model runs for several years with the same cycle until the 10 m depth temperature (approximate limit of the thermally active layer) reaches a mean annual equilibrium value that will be used as the surface boundary condition of the thermo-mechanical model.

At each surface node, the density profile is calculated from the firn thickness F (m w.e.) which is computed as a function of time as follows:

(12) F t + d t = max F t + c t - d f F Δ t ; 0 ,

where c(t) is the daily net surface accumulation (m w.e. yr⁻¹), d_f is a firn densification rate parameter (yr⁻¹) and Δt the time step (yr). The initial value of F is set to F_ref, which is computed from the steady-state mass balance assuming

(13) F ref = B d f if B ≥ 0 0 if B < 0 .

The density is then calculated assuming a linear evolution of density with depth between the surface density ρ₀ (kg m⁻³) and the ice density ρ_ice (kg m⁻³) at the firn–ice interface. It gives

(14) ρ z = ρ 0 + ρ ice - ρ 0 ( z s - z ice ) z s - z ,

where z is the vertical coordinate (m a.s.l.), z_s is the elevation of the surface (m a.s.l.) and z_ice is the elevation of the firn–ice transition (m a.s.l.). From mass conservation, z_ice has to satisfy

(15) ∫ z ice z s ρ 0 + ρ ice - ρ 0 ( z s - z ice ) z s - z d z = ρ w F ref ,

where ρ_w is the water density (1000 kg m⁻³). Combining Eqs. (14) and (15) gives

(16) ρ z = min ρ 0 + ρ ice 2 - ρ 0 2 2 ρ w F z s - z ; ρ ice .

Figure 2The 30 MHz radar profile measured in 2015 along the red dashed line in Fig. 3a. This profile shows the presence of both cold ice (no reflection) and temperate ice (strong scattering).

Download

The density profile and the surface elevation are updated only if F>F_ref by adding the corresponding amount of snow at density ρ₀.

The surface enthalpy field obtained using the distributed 1D model is used as a surface Dirichlet condition to solve the enthalpy equation in the 3D model. The 3D thermo-mechanical steady state of the glacier is then computed by following an iterative procedure:

Compute the steady-state enthalpy field from Eq. (7) with Qlat=v=tr(σϵ˙)=0 forced by the surface enthalpy field given by the distributed 1D model.

1. Compute Q_lat from Eqs. (10) and (11).

2. Solve the ice velocity and deformational heat from the Stokes equation.

3. Compute the steady-state enthalpy field from Eq. (7) with Q_lat determined in Step 2.

4. Steps 2, 3 and 4 are repeated until a steady state is reached.

Table 1Parameters of the surface model for enthalpy boundary conditions.

^a Described in Gilbert et al. (2014b). ^b Formulation proposed by Calonne et al. (2011).

Download Print Version | Download XLSX

3.5 Transient evolution

Transient simulations were performed at yearly time step using the steady-state glacier as the initial condition. Surface mass balance and enthalpy are updated each year from the surface model described previously and forced by daily temperature reanalysis. We assumed a constant basal friction parameter through time.

4.1 Thermal regime and ice thickness measured with the GPR

GPR data show that the Rikha Samba Glacier is a polythermal glacier consisting mainly of cold ice but with significant temperate areas (Figs. 2 and 3). The measured maximum thickness was 178±2 m in the middle part of the glacier where the surface slope is relatively gentle. Ice temperature measurements by the thermistor chain support the GPR interpretation of an upper cold ice layer with sub-zero temperatures at the depth of 10 m (annual mean −2 ^∘C) in the ablation area of the glacier (Fig. 1). A notable characteristic of the GPR-based thermal regime is that the occurrence of temperate ice seems to be associated with the presence of surface crevasses (Figs. 2 and 3). The way that scattering is advected by the ice flow shows that the crevassed areas behave as a source term of scattering. It seems unlikely that the observed scattering would be produced by the crevasse field itself since the scatter occurs in the deeper part of the glacier where crevasses are not expected to persist. It would be rather more compatible with the formation of temperate ice which is then advected by ice flow.

Figure 3(a) Measured temperate ice thickness (dots) and observed crevassed areas (red continuous lines). The red dashed line shows the location of the radar profile presented in Fig. 2. (b) Measured total ice thickness.

4.2 Surface mass balance

We run the mass balance model using the 2014 surface topography over the period 1980–2016 using calibrated temperature reanalysis data (ERA-Interim) (Dee et al., 2011) while assuming a constant precipitation rate. The parameters were constrained by the stake measurements in 2012/2013 (Gurung et al., 2016), meteorological observations (Gurung et al., 2016), and geodetic mass balance over the periods 1974–1994 and 1998–2010 (Fujita and Nuimura, 2011) (Fig. 4). The parameters are summarised in Table 2.

Figure 4Modelled mass balance using ERA-Interim reanalysis data. (a) Surface mass balance at equilibrium. Black dots are stake locations from Gurung et al. (2016). (b) Mean surface mass balance during the period 1980–2016. (c) Corrected ERA-Interim annual temperature (d) Modelled annual surface mass balance compared to geodetic data (Fujita and Nuimura, 2011). (e) Modelled surface mass balance as a function of elevation at equilibrium (black line), in 2012 (red line) and 2013 (green line). Dots are stake measurements for the years 2012 and 2013 (Gurung et al., 2016).

Table 2Parameters of the surface mass balance model.

^a Gurung et al. (2016). ^b Calibrated in this study. ^c Fujita and Nuimura (2011). ^d Gilbert et al. (2016).

Download Print Version | Download XLSX

Balanced conditions for the 2014 geometry are reached for a climate that is 0.7 ^∘C colder than the 1980–2016 climate with an equilibrium line altitude (ELA) of 5770 m a.s.l. (the 1980–2016 ELA is 5880 m a.s.l.; Fujita and Nuimura, 2011). This provides a mean surface mass balance and a melting rate to force the steady-state glacier simulation (Fig. 4a).

The model is in good agreement with the observations but is not able to reproduce the same mass balance distribution as observed in 2013 (Fig. 4b–d). Probably, the inter-annual variability of the mass balance produced with our mass balance model is probably not very well-represented since we assume a constant precipitation rate. Furthermore, the Rikha Samba Glacier is a summer-accumulation-type glacier in which precipitation events in summer can significantly affect the mass balance through albedo feedback (Fujita and Ageta, 2000; Fujita, 2008). However, the long-term trend and the mass balance gradient agree with the observations, which is satisfactory for the purposes of this study focusing on the thermal regime.

4.3 Enthalpy surface boundary condition

The modelled upper boundary condition field (Fig. 5) reveals a mainly cold surface condition with a temperate band between 5800 and 5900 m a.s.l. where both melting and firn thickness are sufficient to maintain temperate conditions. In the higher part of the accumulation area, the water percolation occurs only in the first 2 m due to the limited amount of meltwater resulting in cold temperature at 10 m depth (see Site 1 in Fig. 5), whereas lower down at Site 2, meltwater percolates deep enough to keep temperate conditions all year round at 10 m depth. In the ablation zone (Site 3), water percolation is limited to the seasonal snow thickness, resulting in a cold boundary condition.

Figure 5One-dimensional temperature, water content and density modelled in the first 10 m of depth during 3 years and at three locations on the glacier. The one-dimensional model is forced by a reference temperature–precipitation annual cycle until reaching a stationary periodic response at 10 m depth. Mean annual temperature at 10 m depth is then used as an upper boundary condition for the thermo-mechanical model (mapped in the right panel).

4.4 Modelled steady-state glacier

The modelled steady-state glacier is in good accordance with measured ice thickness (Fig. 6) and measured horizontal velocities (Fig. 7a). We selected a value for the damage threshold of σ_th=0.05 MPa that allows the best match between observed and modelled crevassed areas (Fig. 7c). This value is within the recommended range found by Krug et al. (2014). The correction made on the bedrock topography following the method described in Sect. 3.4.2 greatly improves the quality of the modelling in areas without radar measurements (Fig. 6). A simple interpolation (Fig. 6a) would have led to non-physical ice thickness with unrealistic flux divergence, which is avoided by our method. The bedrock correction in place where there is radar measurement rarely exceeds 20 m (Fig. 6c) apart from a few exceptions (i.e. five points in the 2010 measurements). We preferred to allow the corrected bedrock to differ from radar measurement in order to favour the mass conservation associated with the prescribed surface mass balance. Moreover, radar measurements are uncertain, especially in temperate areas where the bedrock interface is weakly determined. The inversion of the basal friction coefficient (Fig. 7b) provides a final steady state where ice flow is in accordance with steady-state emergence velocities. The good agreement with horizontal velocities measured at stakes (Fig. 7a) shows that our estimated emergence velocities (from surface mass balance) are consistent with the observed ice flow.

Figure 6(a) Interpolated ice thickness from GPR measurements (dots in the three panels). (b) Ice thickness after bedrock topography smoothing and correction using free surface relaxation. (c) Difference between interpolated and corrected ice thickness and positions of the GPR measurements (black dots).

Figure 7(a) Steady-state surface velocities compared with the measurements (coloured dots and inset). Contour lines are surface topography with 50 m intervals (b) Friction coefficient inferred from emergence velocities (assumed to be opposite of surface mass balance). Contour lines are bedrock topography with 50 m intervals. (c) Modelled crevassed areas from maximal Cauchy stress anomaly (colour scale) compared with observations (red lines).

4.5 Thermal regime: influence of meltwater percolation

Modelling results show that water percolation in crevasses strongly affects the steady-state thermal structure of the Rikha Samba Glacier, leading to large temperate zones even at the glacier bed (Figs. 8b and 9b). It significantly extends the temperate-based parts, which cover almost the entire ablation area. Although we adopted a simple approach for water percolation through crevasses, modelled temperate ice thickness is in fairly good agreement with the GPR data (Fig. 8b). If water percolation through crevasses is neglected, the thermal regime of the glacier forced by mainly cold upper boundary conditions (Fig. 5) would result in a mainly cold-based glacier (Fig. 9a). In such a case, cold ice advection from the higher part of the glacier is able to compensate for the temperate surface conditions of the lower accumulation zone (Fig. 5), and only two bands of temperate ice are able to reach the bed on both sides of the flow line of the glacier (Fig. 9a). Such a thermal regime is in disagreement with the observed amount of temperate ice from the GPR data (Fig. 8a). This indicates the significant role of deep water percolation through cracks in cold ice as suggested by the GPR observations. The observed (Figs. 8b and 9b) and modelled (Figs. 8c and 9c) crevassed area leads to a similar result, which validates our approach in modelling the location of the crevasses.

Figure 8Modelled vs. measured temperate ice thicknesses (upper panels) and comparison between model (colour background) and measurements (dots) (lower panels): (a) without crevasse influence, (b) with water percolation in observed crevassed areas and (c) with water percolation in modelled crevassed areas.

Figure 9Modelled steady-state basal temperature. (a) Without crevasse influence. (b) With water percolation in observed crevassed areas. (c) With water percolation in modelled crevassed areas. Dashed lines delimit the temperate areas.

4.6 Transient evolution

Despite the good agreement between the GPR data and the steady-state model, a significant difference exists at the highest crevassed field. A temperate area is clearly visible in the GPR data (Fig. 3a) whereas the steady-state thermal regime model predicts cold ice (Fig. 8). Mass balance measurements show that the Rikha Samba Glacier has not been in an equilibrium state for at least 40 years, with an almost constantly negative rate of −0.5 m w.e. a⁻¹ (Fig. 4d). This temperate area could therefore be the signature of a transient response to air temperature warming. In order to investigate the potential impact of 40 years of unbalanced state on the glacier thermal regime, we performed a transient simulation starting in 1975 from the modelled steady state (with crevasse locations imposed from the modern observations showed in Fig. 3a) and forced by the ERA-Interim reanalysis time series (Fig. 4c). It shows that the upper boundary conditions changed significantly with a cooling of the former accumulation zone in response to firn disappearance and a warming of the highest elevation due to melting increase (Fig. 10b). After a 40-year run, a temperate zone, which do not exist at the steady state, had developed in the highest crevassed area (Fig. 10c–d) as observed in the 2015 GPR measurements (Fig. 10d). This result strongly suggests that the presence of temperate ice in this zone is a result of a transient response to the air temperature warming and increasing surface melting at the higher elevations. The results also agree better with the observations, including the thermistor data (Fig. 10a).

Figure 10Modelled thermal regime after a 40-year transient run forced by mean climatic condition over the period 1981–2016 (see Fig. 3b). (a) Modelled and measured mean temperature profile (2014/2015) in the ablation zone (black circle in b). The inset shows the radar measurement next to the thermistor (black circle in b) with the dashed line showing the modelled cold–temperate transition surface (CTS). (b) Mean surface boundary condition over the period 1981–2016. (c) Modelled basal temperature. (d) Distribution of modelled and measured temperate ice thickness. (e) Modelled temperature along the radar longitudinal section presented in Fig. 2.

To assess the sensitivity of the thermal regime to future temperature change, we performed a future simulation of the glacier retreat and thermal change until 2100 with (Fig. 11) and without (Fig. 12) water percolation through crevasses for a linearly increasing temperature of +1 ^∘C between 2014 and 2100 (+1.7 ^∘C in comparison with the steady-state climatic condition). This shows a much faster development of a new temperate area when water percolation in the crevasses is taken into account. In such a scenario, the glacier becomes almost entirely temperate-based by 2050 (Fig. 11), whereas it would have remained almost entirely cold had water percolation through the crevasses been neglected (Fig. 12). This highlights the crucial role of deep water percolation through cracks when modelling the past and future changes of glaciers' thermal regimes.

Figure 11Future evolution of Rikha Samba Glacier assuming a linear temperature increase of +1 ^∘C between 2014 and 2100 (+1.7 ^∘C in comparison with the steady-state climatic condition). Panels (a)–(d) represent basal temperature evolution. Panels (e)–(h) are temperature evolution along the middle longitudinal section. Black dashed lines delimit the cold–temperate transition surface.

Figure 12Same as Fig. 11 but without water percolation in crevasses.

5.1 Uncertainty

The sensitivity experiment performed for the steady-state simulation shows that the amount of modelled basal temperate ice can vary significantly depending on the choice of parameters associated with firn thickness and basal heat flux, which are poorly constrained. However, the thickness of the modelled temperate ice always remains much less than indicated by the GPR observations if the crevasse influence is not taken into account. This means that uncertainty on those parameters alone does not explain the disagreement between data and model when the role of crevasses is neglected.

The mass balance model we used is simplified since we did not take into account the seasonality and time variation in precipitation, which can be sources of uncertainties in our study. However, the main focus of this study is the thermal regime of the glacier, and, hence, we do not claim that our simulation is accurate regarding the Rikha Samba Glacier past and future evolution. Our study relies on a long-term average of the surface mass balance and our simple calibrated model should be able to adequately capture it.

Since the density of ice is well-constrained and there was no snow or firn over the surveyed part of the glacier at the time of the field measurement in 2015, the main uncertainties of the GPR measurements arise from the GPS positioning of the GPR measurements, the radar wavelength and scattering of the radar signal. For the point measurements and those parts of the GPR profiles along which the bedrock reflection was clearly identified, the accuracy of the horizontal coordinates is about ±20 m, especially on the steepest surface slopes. In addition, the vertical resolution of the GPR signal is usually considered to be approximately one-quarter of the radar signal wavelength, which is about 5.6 and 33.6 m for the 30 and 5 MHz antennae, respectively. In other words, the vertical resolution of the englacial scattering, interpreted as temperate ice and ice thickness along the continuous GPR profiles, is about 1.4 m, whereas the same for the ice thickness obtained from the point measurements is about 8.4 m. The manual picking of bed reflectors in the radargrams introduces extra uncertainties up to 20 m, especially in temperate areas where the bed reflectors are weaker (see Fig. 2). Finally, the limited coverage of the radar profiles on the glacier introduces uncertainty in the bedrock topography inferred from the GPR data even after correction using the model. Uncertainties in the surface DEM are expected to be lower than 1 m and do not introduce extra uncertainties in the bedrock topography reconstructed from the measured thicknesses. Our interpretation of the thermal regime based on the englacial radar scattering of the GPR 30 MHz profiles is supported by previously found close agreement between the observed scattering and borehole temperatures without significant difference in observed englacial scattering relative to the expected measurement error at 10, 35 and 50 MHz antenna frequencies (Wilson et al., 2013).

In the modelled bedrock topography, the difficulty arises from the fact that the friction coefficient is unknown and we had to assume a uniform value of basal friction coefficient to correct the bedrock topography from flux divergence anomalies. The friction coefficient we inferred in a second step to force the steady emergence velocity to match the balanced surface mass balance is therefore affected by ice thickness errors. The resulting velocity field is consistent with the surface mass balance, but uncertainties in the respective contribution of the basal sliding and ice deformation to surface velocities remain. This makes the interpretation of the modelled basal friction difficult (Fig. 7b), which thus has to be seen more as a tuning parameter than a parameter revealing physical processes.

5.2 Role of surface runoff through supraglacial streams

Our study shows that the thermal regime of the Rikha Samba Glacier can be modelled by taking into account the liquid water derived from local surface melt percolating into crevassed fields. Neglect of water inputs coming from surface runoff through supraglacial streams seems thus to be a valid hypothesis. For a polythermal glacier with a cold surface layer, which is a common feature in the ablation areas where firn heating is nonexistent, surface runoff occurs in well-marked and persistent streams at the surface (Ryser et al., 2013). These streams bring water into the crevasse fields through a few localised entry points only. Hence, as there is only a relatively small contact surface between the streams and cold ice, a limited amount of water will refreeze. This is why the influence of surface runoff on the thermal regime of the glacier is likely to remain limited. Similarly, Lüthi et al. (2015) have concluded that moulins have little influence on the thermal regime of the Greenland Ice Sheet since they can be represented as line sources which provide limited warming to the surrounding ice. On the contrary, surface melting occurring in crevassed fields is well-distributed and can release latent heat on a much larger area, thus having a stronger impact on the thermal regime.

5.3 Enhanced influence of climate change on glaciers' thermal regime and dynamics

The influence of deep latent heat release through meltwater percolation in crevasses has been already inferred for Greenland. Lüthi et al. (2015) and Seguinot et al. (2020) observed temperature anomalies in borehole measurements that can only be explained by latent heat released down to a 400 m depth in crevassed fields. Similar conclusions have been made by Hills et al. (2017), although they show that these effects remain localised and may not really influence the thermal regime of the Greenland Ice Sheet on a large scale. In the case of mountain valley glaciers, crevassed fields can cover a significant fraction of the total glacier area, which is combined with a generally much faster ice flow, leading to efficient advective processes that transport the heat produced in the crevassed areas. The results of these combined effects are significant, greatly influencing the thermal regime at the glacier scale as shown in this study. As already pointed out for ice sheets (Phillips et al., 2010, 2013), the timescale of the glacier thermal regime response to climate change is also greatly diminished in comparison with the case where only heat diffusion and advection of surface changes are involved in the warming process. However, we show here that deep water percolation is likely restricted to the crevassed areas and absent elsewhere in order to model the observed thermal structure. This restricts the spatial extent of the process on the glacier as it is dependent on the bedrock topography and its related crevasse positions. For instance, the Yala Glacier, another polythermal glacier in the Nepal Himalaya (Sugiyama et al., 2013), is entirely cold in its ablation area, likely due to thinner ice and a lack of crevassed areas. Nevertheless, it is likely that the meltwater percolation via crevasses has a significant impact on the thermal regime of numerous polythermal high-altitude glaciers. Future warming could also lead to a faster thermal regime response than previously thought (Gilbert et al., 2015), especially in the cold accumulation areas where the pre-industrial melting rates were not sufficient to form temperate ice.