21st century fate of the Mocho-Choshuenco ice cap in southern Chile

. Glaciers and ice caps are thinning and retreating along the entire Andes ridge, and drivers of this mass loss vary between the different climate zones. The southern part of the Andes (Wet Andes) has the highest abundance of glaciers in number and size, and a proper understanding of ice dynamics is important to assess their evolution. In this contribution, we apply the ice sheet model SICOPOLIS to the Mocho-Choshuenco ice cap in the Chilean Lake District (40 ◦ S, 72 ◦ W, Wet Andes) to reproduce its current state and to project its evolution until the end of the 21st century under different global 5 warming scenarios. First, we create a model spin-up using observed surface mass balance data on the south-eastern catchment, extrapolating them to the whole ice cap using an aspect-dependent parameterization. This spin-up is able to reproduce the most important present-day glacier features. Based on the spin-up, we then run the model 80 years into the future, forced by projected surface temperature anomalies from different global climate models under different radiative pathway scenarios to obtain estimates of the ice cap’s state by the end of the 21st century. The mean projected ice volume losses are 56 ± 16 % 10 (RCP2.6), 81 ± 6 % (RCP4.5) and 97 ± 2 % (RCP8.5) with respect to the ice volume estimated by radio-echo sounding data from 2013. We estimate the uncertainty of our projections based on the spread of the results when forcing with different global climate models and on the uncertainty associated with the variation of the equilibrium line altitude with temperature change. Considering our results, we project a considerable deglaciation of the Chilean Lake District by the end of the 21st century

Icefield and Cordillera Darwin are located in this region. The specific mass losses observed or inferred for the glaciers of the 20 Wet Andes are the highest in the Andes (Dussaillant et al., 2019;Braun et al., 2019) and among the highest of all glacier regions worldwide (Zemp et al., 2019).
The maritime climate of the Wet Andes is characterized by high precipitation rates of up to 10000 mm yr −1 on the windward side and rather mild temperatures with freezing levels generally above 1 km above mean sea level (Garreaud et al., 2013). This leads to an exceptionally high mass turnover (Schaefer et al., 2013(Schaefer et al., , 2015(Schaefer et al., , 2017 and high flow speeds for the glaciers in the 25 region (Sakakibara and Sugiyama, 2014;Mouginot and Rignot, 2015). Apart from climate forcings, ice dynamics and frontal ablation are important contributors to glacier changes in the region. As these are best represented in ice-flow models, they are appropriate tools to project future behavior of the glaciers of the Wet Andes.
Only few studies have tried to project future behavior of Andean glaciers. Réveillet et al. (2015) modelled Zongo Glacier (16 • S) in the tropical Andes using the 3-D full-Stokes model Elmer/Ice (developed by Gagliardini et al., 2013). They projected 30 volume losses between 40% and 89% until the end of this century under the RCP2.6 and RCP8.5 scenarios, respectively.
In the Wet Andes, Möller and Schneider (2010) projected an area loss of 35% of Glaciar Noroeste, an outlet glacier of the Gran Campo Nevado ice cap (53 • S), until the end of the 21st century using a degree-day model and volume-area scaling relationships. Schaefer et al. (2013)  In this contribution, our first objective is to reproduce the present-day behaviour of the Mocho-Choshuenco ice cap in the northern part of the Wet Andes (40 • S) using the ice-sheet model SICOPOLIS (Greve, 1997a, b). To this end, we make use 40 of a newly developed SMB parameterization scheme and glaciological data obtained on the ice cap to calibrate the model and reproduce its current state. Our second objective is to project the behaviour of the Mocho-Choshuenco ice cap through the course of the 21st century to provide one of the first constraints on future glacier dynamics in the Wet Andes. For this aim, we make use of temperature projections from 23 Global Climate Models (GCMs) participating in the Coupled Model Intercomparison Project phase 5 (CMIP5) (Taylor et al., 2012) under low (RCP2.6), medium (RCP4.5) and high emission 45 (RCP8.5) scenarios as input to SICOPOLIS.
We begin this paper by describing the observational data and methods (Section 2). In Section 3, we present the results: first, we validate the model spin-up using observed glacier outlines, ice thickness and flow speed. We then present the evolution of ice cap extension and volume during the 21st century as obtained through different emission scenarios. Then, in Section 4, we discuss our results, compare them to previous studies and analyse the limitations of our approach. We conclude the paper by 50 summarizing the main findings in Section 5.

Observational data
In this study, we simulate the present and future state of the ice cap covering the Mocho-Choshuenco volcanic complex, which we refer to as Mocho-Choshuenco ice cap. It is located in the Chilean Lake District at 40 • S, 72 • W (see inset map in Figure   55 1). Over the last 20 years, climatological and glaciological observations have been made on the ice cap (Rivera et al., 2005;Schaefer et al., 2017). SMB data were obtained through the traditional glaciological method on a stake network on the southeastern part of the ice cap (red stars in Figure 1). These measurements yielded an average negative SMB of −0.9 m w.e. yr −1 (meter water equivalent per year) with an exceptionally high mass turnover of around 5 m w.e. yr −1 (Schaefer et al., 2017). This high mass turnover can be explained by climatological data: between 2006 and 2015, the annual mean temperature was 2.6 • C at 60 an automatic weather station (green circle in Figure 1) at an elevation of 2000 m, and therefore close to the typical equilibrium line altitude (ELA) (Schaefer et al., 2017). Mean annual precipitation over the same period was around 4000 mm yr −1 in Puerto Fuy at an elevation of 600 m to the north of the volcano.
3 https://doi.org/10.5194/tc-2020-296 Preprint. Discussion started: 11 November 2020 c Author(s) 2020. CC BY 4.0 License. At some of the mass balance stakes (red stars with inner black dots in Figure 1), GPS measurements were made to infer surface flow velocity, giving typical values of around 30 m yr −1 (Geoestudios, 2013). Further measurements include ground penetrating radar (GPR) transects (green lines in Figure 2a) over most parts of the ice cap (Geoestudios, 2014). Through inverse distance weighting interpolation over the whole ice cap, a total ice volume of 1.038 km 3 was obtained (Geoestudios, 2014). The interpolated ice thickness map was subtracted from a digital elevation model ( TanDEM WorldDEM, acquired between 2012 and 2014) to yield a bedrock topography (Flández, 2017). We use this topography as the base of the ice cap in the simulations we perform with SICOPOLIS.

SICOPOLIS
The three-dimensional, dynamic/thermodynamic model SICOPOLIS (SImulation COde for POLythermal Ice Sheets) was originally created in a version for the Greenland ice sheet (Greve, 1997a, b). Since then, the model has been developed continuously and applied to problems of past, present and future glaciation of Greenland, Antarctica, the entire northern hemisphere, the polar ice caps of the planet Mars and others, resulting in more than 120 publications in the peer-reviewed literature 75 (www.sicopolis.net). The model supports the shallow-ice approximation (SIA) for slow-flowing grounded ice, hybrid shallowice-shelfy stream dynamics for fast-flowing grounded ice and the shallow-shelf approximation for floating ice (Bernales et al., 2017), as well as several thermodynamics solvers (Blatter and Greve, 2015;Greve and Blatter, 2016).
Mainly developed for ice sheets, the smallest ice body to which SICOPOLIS has been applied so far is the Austfonna Ice Cap, for which Dunse et al. (2011) reproduced the observed cyclic surge behaviour under constant, present-day climate conditions.
where v b is the basal sliding velocity, τ b the basal drag and C b the sliding coefficient. The value of the latter is determined 85 by the calibration procedure of the present-day spin-up (see Section 3.1). Since Mocho-Choshuenco is a temperate ice cap, we replaced solving the energy balance equation by keeping the temperature at a constant value of 0 • C (precisely speaking, and for technical reasons only as SICOPOLIS does not allow an all-temperate ice body, −0.001 • C). The rate factor is set to the value recommended by Cuffey and Paterson (2010) for 0 • C, which is A = 2.4 × 10 −24 s −1 Pa −3 . To ensure proper mass conservation despite the steep slopes and rugged bed topography, we use an explicit solver for the ice thickness equation that 90 discretizes the advection term by a mass-conserving scheme in an upwind flux form (Calov et al., 2018).

Aspect-dependent SMB parameterization
SICOPOLIS incorporates a linear altitude-dependent SMB parameterization which is visualised in Figure 3a and can be described by the following formula: Here, ELA is the equilibrium line altitude, z(C) is the altitude of a specific grid cell C, M 0 denotes the mass balance gradient and S 0 is maximum snowfall.  These values can be summarized in a cosine function in ϕ with the direction of maximum ELA ϕ 0 , the amplitude A ELA and an offset of the average ELA B ELA : B ELA is used to shift the ELA to the desired mean altitude. ϕ is the cardinal direction of a point with respect to the summit and can be calculated by where arctan2 denotes the two-argument arctangent and x and y are the distances in the two directions from a grid point to the 110 summit location (x sum , y sum ).

Temperature projections
The main goal of this study is to project the future evolution of the Mocho-Choshuenco ice cap. We use future temperature simulations from 23 climate models participating in CMIP5 (see Appendix A). To ease the calculations, all the models were interpolated onto a common grid of 1.5 • × 1.5 • using bilinear interpolation. Then the time series of each model were extracted

Glacier sensitivity to temperature change
To link the projected 21st century temperature rise to ice dynamics, it is necessary to relate the temperature anomalies to 135 changes in SMB which is determined by the ELA in our case. For the Mocho-Choshuenco ice cap, there are four years (2009-2013) of availability of both the ELA and annual mean temperature at a similar altitude (Schaefer et al., 2017). These data are shown in Figure 5 together with the ELA error estimates.
9 https://doi.org/10.5194/tc-2020-296 Preprint. Discussion started: 11 November 2020 c Author(s) 2020. CC BY 4.0 License. In order to predict the ELA for any temperature, we first assume a linear relationship between both and solve a weighted least squares problem to find the slope and intercept (i.e. ELA gradient and ELA for 0 • C). ELA predictions for any temperature
Inserting the observed data, we identify the forward operatorĜ, the data covariance matrix Σ d and the vector of observed Predictions for a general temperature T can be made through Figure 5 shows the mean of ELA predictions against temperature together with the ELA predictions of mean plus and minus one standard deviation.
Since the temperature projections give anomalies with respect to the period 2006-2020, we only rely on relative rather 155 than absolute temperatures. Therefore, we convert the temperature changes into changes of ELA with the parameter µ 1 = 88 m K −1 , which means that the ELA increases by 88 m per • C temperature increase. We assess the uncertainty propagation of this parameterization through the ice flow simulation code by performing additional experiments with upper and lower ELA gradients µ 1 ± √ Σ 11 = (88 ± 37) m K −1 which corresponds to the one-sigma confidence interval.

Spin-up and model calibration
First, we reproduce the present-day ice cap behaviour as a steady-state. We calculate the root mean square error for ice thickness difference between the observed and modelled ice caps. We update the mean ELA (B ELA ), ELA amplitude (A ELA ), maximum snowfall (S 0 ) and sliding coefficient (C b ) according to a Nelder-Mead algorithm (Nelder and Mead, 1965) to create an ensemble of parameter sets with similarly low root mean square error in ice thickness. Out of this ensemble, the  Overall, a good agreement can be observed. Apart from local inaccuracies in ice thickness, there are two areas where the simulation does not reproduce well the observations: there is no ice between the two summits in the northern part of the ice cap, and a small ice tongue in the south-eastern catchment develops, with a maximum ice thickness of over 100 m.

Projected ice thickness and area evolution
When forcing the ice cap with higher temperatures under the RCP2.6, RCP4.5 and RCP8.5 scenarios, the ice cap responds with an increase in its ELA. This increase, in turn, leads to thinning and retreat, i.e. to a lower ice thickness and a smaller ice area, respectively, towards the end of the century. Figure 7 shows the 21st century ice thickness evolution for the different scenarios, 185 obtained after averaging through all 23 climate models for each of the three scenarios.
In the RCP2.6 scenario, thinning is more dominant than retreat. The ice retreats mostly in the northern part of the ice cap around the Choshuenco peak, and after 2060 in the outlet tongues of the southern part. Thinning, however, is ubiquitous and occurs over the whole domain during the 21st century.
The ice cap evolution under the RCP4.5 scenario shows a similar retreat pattern as that in RCP2.6, yet more areas in the

Projected ice volume evolution
Projected thinning and retreat of the ice cap illustrated in the previous Section yield the loss in total ice volume. The evolution of the total ice volume under different scenarios is shown in Figure 8. The projections for the 23 individual climate models (thin lines) can be summarized by the multimodel ensemble mean (thick solid lines) and one-sigma confidence interval (thick 205 dashed lines). The three scenarios begin to diverge in the 2040s, and in the 2050s the glacier projections obtained with the RCP8.5 scenario start to clearly diverge from the other two scenarios. The ice volume evolution under the RCP2.6 and RCP4.5 scenarios shows similar behaviour until the mid century, whereas notable differences exist between them towards the end of the century, particularly after the 2080s.
There is a large agreement among the ensemble members of the RCP8.5 scenario that generally show the same declining 210 patterns, which are even more prominent by the end of the century. On the other hand, RCP4.5 and RCP2.6 scenarios show increase in uncertainty with the increase in time, in particular, the projections under the RCP2.6 scenario illustrate a large uncertainty bound spanning from the mid century to the end of the century. These contrasts in the projections under different emission scenarios reflect higher signal-to-noise ratio for the RCP8.5 scenario, as this scenario has a more prominent temperature increase (also see Figure 4). It is worth noting that ice volume evolution tends to have similar changes with a small 215 uncertainty bound under each scenario during the first two decades (i.e., until 2040s) which can be explained by the dominant role of the internal variability in the near future. Projected ice volumes for different scenarios and years are summarized in Table 3.

Uncertainty of ELA gradient
In this Section, we analyse the impact that the ELA dependence on temperature has on glacier projections. We average the 23 interval (51 m K −1 and 125 m K −1 ). The resulting ice volume evolutions are shown in Figure 9, with the mean in solid lines and the one-sigma confidence interval indicated by dashed lines.
225 Figure 9. Volume evolution for different gradients between ELA and temperature.

Present-day simulations
The first part of our study consists of the creation of a present-day steady-state of the Mocho-Choshuenco ice cap. Since drivers of the SMB such as solar radiation and snow redistribution are strongly aspect-dependent, we developed a new SMB parameterization accounting for aspect-dependent SMB variations (see Section 2.3). As shown in Figure 6a, the steady-state 230 now reaches a volume very similar to the observed one, and the ice thickness distribution and ice outline (Figure 6b) match the observations much better than previously, without the new parameterization (Flández, 2017). Also, the simulated velocities are mostly in good agreement with observed values (see Figure 6d and Table 1).
The difference between observed and simulated thickness (Figure 6c) shows over-and underestimations of up to 100 m, with two notable features. In the south-eastern part of the ice cap, the simulations create a new ice tongue. This feature can be 235 explained by high ice velocities in this area (Figure 6d), leading to a fast redistribution of ice before the highly negative SMB melts it away. However, close observations of satellite images show that there is a debris-covered body of dead ice in the valley below the glacier (Scheiter, 2019). This shows that the persistence of ice in this area is not unrealistic, even though the high observed velocities demonstrate that ice dynamics are not well represented by our model (Figure 6d).
The other notable feature is the area between the two summits where no ice is present in the simulations, contradicting the 240 observations. One possible explanation is that the SMB variations underlying our aspect-dependent SMB parameterization also hold for the Choshuenco peak, but were only applied to the Mocho peak. Furthermore, there are no GPR data available in this area (see Figure 2b), making an interpretation difficult in the context of this study.
The velocities presented in Table 1 show simulated values at the stake locations that are lower than the observed ones. However, a direct comparison is not meaningful as the observed values were taken in October, while the simulated velocities are

21st century projections
For the RCP2.6 scenario, we project a steady decay in ice volume until the end of the 21st century. This ice loss is predominantly driven by thinning, and to a lesser degree by retreat. A reason for this might be that the ice cap is in an overall stable position at the moment and slow temperature increase leads first to thinning and then to retreat due to dynamical glacier response. We assess the uncertainty of the RCP2.6 scenario based on two influences: different climate models and the uncertainty of the 255 ELA against temperature parameterization. Both uncertainties increase linearly over the 21st century, but the climate ensemble uncertainty has about twice the magnitude. A possible explanation for this is that the moderate average increase in temperature of only 0.33 • C does not emphasize differences in the ELA gradient as much as it is the case for more drastic scenarios.
In the RCP4.5 scenario, our projections show a higher retreat rate compared to the RCP2.6 scenario, and together with a relatively high thinning rate in the end of the 21st century leads, the ice loss almost doubles. Similar to the previous scenario, 260 the RCP4.5 projections also shows a linear, but slightly slower increase in climate model uncertainty, separating significantly from the RCP2.6 scenario in the 2080s. The uncertainty introduced by the ELA gradient variations is much higher than in the RCP2.6 scenario. This underpins the conclusion that the higher rate in temperature increase emphasizes the ELA gradient differences more.
The high-end atmospheric warming scenario RCP8.5 causes a highly accelerated ice loss from the 2040s to the 2080s with 265 high retreat rates, before becoming more subtle from 2080 to 2100. This behaviour illustrates a highly unstable ice cap during most of the 21st century, and the more subtle retreat in the 2090s can be explained by the fact that most ice has already melted away. This saturation in volume loss can also be observed in the climate model uncertainty: while it increases strongly between 2050 and 2080, it becomes very small towards the end of the century, showing an overall agreement between the climate models regarding the point in time where the ice cap almost vanishes. The ELA uncertainty shows a similar behaviour, with an 270 increase until 2080 and subsequent decrease, which is more pronounced for the lower bound. The ice loss for the upper bound remains lower than that in the lower bound of the RCP4.5 scenario, which highlights the high uncertainty introduced by the ELA parameterization.

Limitations of our approach
In this study, the principal uncertainties we assign to our results are based on the spread of the temperature projections of the 275 global circulation models and on the uncertainty of the temperature-ELA parameterization. In this section, we discuss possible further sources of uncertainty, and make suggestions on how future work could encounter these challenges.
Our approach is based on the shallow ice approximation (SIA), with assumptions including almost parallel and horizontal glacier bed and surface, significantly larger horizontal than vertical dimensions and simple-shear ice deformation. While these assumptions hold well for the large Greenlandic and Antarctic ice sheets, it is less obvious that the SIA can be employed on 280 such a small study object as the Mocho-Choshuenco ice cap. The SIA assumptions are violated especially in the steep regions around the two summits and towards the boundaries of the present-day ice cap. However, they hold true for large parts of the plateau on the south-eastern part which accounts for over a third of the ice cap and most important area in our future projections.
Previous studies have suggested that low-order assumptions such as the SIA hold well for glaciers whose behaviours are mostly driven by SMB (Adhikari and Marshall, 2013), which is the case for the Mocho-Choshuenco ice cap. However, it would be a 285 valuable experiment to reproduce our results with a full-Stokes model such as Elmer/Ice to verify the applicability of the SIA.
Knowledge about the bed of the ice cap is essential to perform ice flow simulations. We created a bed map based on presentday topography and a number of ground-penetrating radar profiles published by Geoestudios (2014). Even though these profiles cover a significant portion of the ice cap, there are large gaps in data coverage, especially in the north-eastern part of the ice cap. More observations could help to reduce the uncertainty introduced by these gaps.

290
Regarding the ELA gradient we use to relate temperature increase to glacier response, it is important to note that we have a few data points given for this relationship (Schaefer et al., 2017). With more years of ELA-temperature pairs and a thorough 20 https://doi.org/10.5194/tc-2020-296 Preprint. Discussion started: 11 November 2020 c Author(s) 2020. CC BY 4.0 License. uncertainty estimation, we could achieve a higher confidence in our ELA gradient. However, by performing the simulations for the mean gradient and a lower and upper bound, we are within the range of most previous studies (e.g. Six and Vincent, 2014;Sagredo et al., 2014;Wang et al., 2019).

295
Another significant limitation lies in the SMB parameterization. While the new aspect-dependent parameterization was able to improve the reproduction of the present-day ice cap significantly, there is still space for improvement. Especially the northern part is still not well reproduced by SICOPOLIS, and it might be of advantage to extend the new parameterization to the Choshuenco peak. In order to verify our parameterization, it would be helpful to obtain SMB measurements in the northwest, i.e. between both summits, and thus extend the stake network that is currently focused on the main catchment in the 300 south-east of Mocho summit. This could provide more observational constraints on the ELA difference between the north-west and south-east.
Another way of producing more realistic SMB maps for the ice cap would be using explicit models that try to quantify the physical processes which determine glacier mass balance, e.g. the COSIPY Model (Sauter et al., 2020). Drawbacks of these complex models is that they need many input parameters (such as precipitation, relative humidity or wind speed) with 305 a high spatial resolution. These can be obtained by regional climate model simulations (e.g. Bozkurt et al., 2019). However considerable uncertainties are associated to these simulations and a careful validation of the results is necessary before using them as drivers of SMB simulations. Additionally, only few high resolution regional climate simulations are available in the moment which is why we prefer our simple temperature dependent SMB parameterization combined with a multi-model approach using 23 different GCMs as drivers of our simulations.

Global context of glacier decline
To our knowledge, there are only few previous studies that have projected the future evolution of glaciers in the Andes. The nearest study object to the Mocho-Choshuenco ice cap is the Northern Patagonian Icefield for which until 2100 an ice mass loss of 592 Gt has been projected under the A1B scenario which is comparable to the RCP6.0 scenario, and therefore between our results for RCP4.5 and RCP8.5 (Schaefer et al., 2013). Relating this ice loss to more recent estimates of total ice mass 315 (Carrivick et al., 2016;Millan et al., 2019), around 50% of the ice mass would have disappeared. However, these simulations were performed on a fixed geometry, and therefore considered only changes in SMB, making it difficult to compare their results to ours. Collao-Barrios et al. (2018) obtained a committed mass loss of approximately 10% for San Rafael Glacier under current climate. However, they also maintained glacier area constant during their simulations and therefore neglect glacier retreat, which could dramatically change rates of frontal ablation. 320 Möller and Schneider (2010) projected the future evolution of Glaciar Noroeste, an outlet glacier of the Gran Campo Nevado ice cap in southern Patagonia between 1984 and 2100. Their projections were made for the B1 scenario and yielded a volume loss of around 45% which is significantly less than the 61% volume loss that we project for the comparable RCP4.5 scenario between 2013 and 2100. Their results are based on a calibrated relationship between area and volume, and not on ice flow modelling as our study.

325
The only ice sheet modelling approach to project future change under climate change scenarios in the Andes so far is the work of Réveillet et al. (2015) on Zongo glacier in Bolivia, using the full-Stokes model Elmer/Ice. They projected 40% and 89% volume loss for the RCP2.6 and RCP8.5 scenarios, respectively. The value for the high-end scenario is comparable to ours (94%), which might be expected as both glaciers are about to disappear by the end of the century, and therefore have already lost the majority of their ice mass. On the other hand, our projections for the RCP2.6 are lower (25%) than their RCP2.6 330 projections. The might be due to the climate differences between the tropics and the Wet Andes, and also due to the higher ELA gradient with temperature of 150 m K −1 used in their study in comparison to 88 m K −1 used in our study. Hock et al. (2019) projected future mass loss in 19 different glacier regions worldwide using six different glacier models of different complexity. Similar to our approach, different ensemble of the CMIP5 GCMs were used to drive the glacier model projections. For the Southern Andes, they projected regional ice mass losses of 18.7% (RCP2.6), 31.6% (RCP4.5) and 335 45.6% (RCP8.5). Their projected regional percental mass losses are clearly lower than our projected mass loss for the Mocho-Choshuenco ice cap, especially for RCP4.5 and RCP8.5. Here it is important to note that the regional mass loss in the Southern Andes is greatly determined by the mass loss of the large Patagonian ice fields. Here many glaciers are calving glaciers and the projections of Hock et al. (2019) need to be interpreted with care, since only one of the six models include a parameterization of frontal ablation.

340
Outside of South America, there are several studies that project a glacier shrinkage similar to our projections in the next 100 years. Kienholz et al. (2017) projected a mass loss of 73% for Black Rapids Glacier in Alaska between 1980 and 2100 under the RCP8.5 scenario with a SMB model coupled to a simple mass-conservation based retreat parameterization. They used a similar downscaling approach to obtain high-resolution temperature and precipitation fields than that applied by Schaefer et al. (2013).

345
Another study in North America is that of Adhikari and Marshall (2013) who performed ice flow simulations on Haig Glacier in the Rocky Mountains and projected the disappearance of the glacier by 2080 under the RCP4.5 and RCP8.5 scenarios. These results hold for both high-order and low-order mechanical models, and the authors conclude that low-order models are suitable for glaciers with low ice velocity and a glacier geometry controlled largely by SMB. As these conditions hold strongly for the Mocho-Choshuenco ice cap, their conclusions also underpin our assumption that a low-order model such as SICOPOLIS is https://doi.org/10.5194/tc-2020-296 Preprint. Discussion started: 11 November 2020 c Author(s) 2020. CC BY 4.0 License. many mountain glaciers in different parts of the world to disappear in the first half of the 22nd century, without reductions of greenhouse gases.

Conclusions
In this study, we applied the ice sheet model SICOPOLIS to reproduce the current state of the Mocho-Choshuenco ice cap and to project its future evolution under different emission scenarios. This is the first estimate of future glacier evolution obtained 365 from a flow model in the Wet Andes, and one of the first in the whole Andes. Using a linear temperature-ELA parameterization, we investigate the future of the ice cap using projected temperature changes from 23 GCMs as input. A considerable spread of the projected ice volume at the end of the 21st century is obtained, depending on the emission scenario and GCM. However, under the emission scenario RCP8.5, which does not consider a reduction of our emission of greenhouse gases, the spread of the results is less pronounced and in this scenario it is likely that the ice cap will loose more then 90% of it current volume by 370 2100. Since temperature projections are relatively uniform in the region and geometry of the surrounding ice caps are similar to Mocho-Choshuenco ice cap, we project similar high volume losses for other ice-caps in the Chilean Lake District (39-41.5 • S).
The Mocho-Choshuenco ice cap is the smallest ice body to which SICOPOLIS has been applied so far, justified a priori by the cap-like geometry (as opposed to, for example, valley glaciers), and a posteriori by the reasonably good performance of the model in replicating the present-day ice cap. Nevertheless, it would be valuable to check if the application of a full-Stokes 375 glacier flow model (as for example Elmer/Ice, Gagliardini et al. (2013)) affected the simulated state of the ice cap notably, or if the disagreements are mainly caused by our simplified surface mass balance parameterization. When projecting the future of the large Patagonian icefields in the southern part of the Wet Andes, the interaction of their outlet glaciers with the surrounding water bodies becomes crucial. Adequate parameterizations for frontal ablation are necessary, which allow the glaciers to adapt their frontal positions according to the glacier flow, which will be crucially determined by these parameterizations.