Rhonegletscher and the Gletsch catchment (Fig. , Table ), are situated in the central part of Switzerland within the Canton of Valais. The Gletsch catchment (39.4 km²) is the headwater of the Rhone River. The term “Rhonegletscher” in this study encompasses the main glacier (14.9 km² in 2016) along with 10 smaller glaciers (cumulative 1.8 km² in 2016) within the Gletsch catchment, all contributing to the hydrological dynamics of the region (Fig. ). This glacier has been the subject of extensive research on e.g. glacier mass balance, hydrology, glacier dynamics, and the impacts of climate change . The availability of extensive datasets allows us to develop, calibrate and validate our glacio-hydrological model and explore its performance. A summary of the main characteristics of the catchment are found in Table .

GERM is driven by a point time series of temperature and precipitation, either near or within the catchment area, which are subsequently distributed across the catchment using a monthly-averaged temperature lapse rate (cf. Table S1) and a constant precipitation lapse rate to every grid cell at the specified model resolution. For each meteorological product, temperature lapse rates were computed as monthly constants by performing a linear regression of air temperature against elevation of grid cells that fall within the catchment. For the ERA5 Reanalysis, neighbouring grid cells outside the catchment were also included, since the catchment itself is covered by only a single grid cell. For the Grimsel station data, we used observations from surrounding stations within a 50 km radius to perform the linear regression. These monthly lapse rates were then used to downscale the temperature time series across the model domain. Precipitation is distributed across the catchment by applying an overall correction factor (Cprec) and an annually fixed precipitation lapse rate (dP / dz) based on regional literature values e.g. and validated by in situ snow accumulation data over the glacier’s elevation range. For capturing the small-scale spatial variability of snow accumulation, a distribution matrix derived from terrain characteristics (slope and curvature) is superimposed on spatialized precipitation . In this setup, the spatial distribution of precipitation within the original product has a limited effect on the catchment-averaged time series applied in the model. This was tested by upscaling the high-resolution products to a coarser resolution prior to extracting the catchment-averaged precipitation time series (cf. Supplement Figs. S1 and S2). Consequently, in our model configuration, the ability of the precipitation product to accurately capture total amounts and temporal variability is of greater importance than its spatial resolution.

The annual glacier surface mass balance is quantified as the sum of solid precipitation (accumulation, A) and snow/ice melt (ablation, M) and only requires temperature and precipitation data as forcing. Accumulation A is estimated, in every grid cell (x,y) and day (d) as solid precipitation (Psolid), calculated as the amount of precipitation falling below a temperature threshold (Tthr) of +1.5 °C, with a linear transition range between +0.5 and +2.5 °C : 1A(x,y,d)=Psolid(d)⋅Cprec⋅D(x,y) The parameter Cprec allows for the adjustment of measured precipitation sums to the catchment . The spatial distribution (D, at every grid cell x,y) of accumulation on the glacier surface is modelled by a simplified parametrization of snow redistribution processes, including snow drift and avalanches. This is achieved using curvature and slope assessments derived from the input DEM and the specified model resolution at every grid cell . The snow distribution is then normalized across the catchment to a value of 1, ensuring that only the spatial distribution is affected, without altering the total amount of solid precipitation. Ablation is computed by using the distributed temperature-index model proposed by that incorporates potential solar radiation. The surface melt rates M in every grid cell (x,y) and day (d) is computed by : 2M(x,y,d)=FM+rice/snowI(x,y,d)T(x,y,d):T(x,y,d)>0°C0:T(x,y,d)≤0°C In the equation, FM is a melt factor, rice/snow are two radiation factors for ice and snow, I is the potential clear-sky solar radiation at every grid cell and day calculated based the topography and solar angle based on , and T is the local air temperature.

Glacier geometry and area are updated annually using the Δh-parametrization . It approximates changes in glacier surface elevation and glacier area in response to annual mass balance. This empirical approach redistributes net mass changes across the glacier based on a normalized elevation-dependent function (Δh) derived from observed surface elevation changes in the past. The parameterization is mass-conserving and reflects typical glacier behavior, producing the largest and smallest elevation changes in the ablation and accumulation area, respectively. It adjusts the glacier extent by removing glacier sections where the surface elevation falls below the bedrock. Albeit the Δh-parameterization does not explicitly simulate dynamic processes, it has been shown to closely replicate the results of a 3-D finite element flow model in terms of glacier volume, length, and area evolution over decadal scales . Since the Δh-approach allows the glacier to transiently adjust to the imposed climate forcing, no spin-up time was applied in our simulations.

GERM uses a runoff routing scheme that integrates meltwater and rainfall, with evapotranspiration subtracted at each time step seefor a detailed description of this model component. The scheme is structured around the concept of linear reservoirs and simulates the water balance of every grid cell and time step across diverse surface types – including ice, snow, rock, vegetation, and groundwater – by routing water through type-specific reservoirs with fixed retention constants. Each land surface type is assigned to a reservoir and associated with specific fixed retention and storage parameters, originally described in and . These parameters are not calibrated in this study but are based on validated applications of GERM to similar catchments, including the Gletsch basin e.g.. A detailed list of the parameter values used is provided in Table S3. This representation captures both rapid surface runoff and delayed subsurface flow components, which are particularly relevant during summer rainfall events and low-flow conditions. The total discharge is obtained by summing the outflows from all reservoirs at the catchment level, enabling a fully distributed, partitioned hydrograph simulation . 3Qd=Pliq,d+Md-ETd-∑rΔSr,d. Here, d is the time step (days), Qd total runoff, Pliq,d liquid precipitation, Md snow/ice melt, ET_d evapotranspiration and ΔSr,d the storage change of the reservoir r.

Besides the two Experiments 1 and 2, we also explore two calibration procedures to assess model performance under varying data availability scenarios. In both procedures, GERM’s calibration focuses on two main parameter groups: accumulation and ablation parameters. Geodetic glacier mass change serves as the primary constraint, and additional constraints can include measured runoff data. During the calibration process, the model adjusts the ablation parameter, which includes the melt factor (FM) and the radiation factors for ice and snow (rice/snow) in an automated procedure. FM and rice/snow have a fixed relation to each other (rice/FM=0.024; rsnow/rice=0.66). The ratio between the parameters was adopted from earlier applications of the same model, which demonstrated their suitability for glacierized catchments in the Swiss Alps . Thus, they can be handled as one parameter, which is optimized without setting a fixed parameter range. At the same time, the precipitation correction factor (Cprec) is optimized within bounds of [0.6, 1.5]. Cprec is a constant parameter, adjusting daily catchment precipitation – both liquid and solid – by a fixed percentage, thereby increasing or decreasing it uniformly over the modelling period. Since accumulation in GERM is entirely determined by solid precipitation, and Cprec directly scales this input, it effectively also controls the magnitude of accumulation in the model.

The two calibration procedures tested here differ in their use of constraints and the scope of parameter optimization. In the single-data calibration, only the ablation parameter (FM, rice/snow) is optimized. In contrast, the accumulation parameter (Cprec) remains fixed at 1.0, and no measured runoff data is used as a constraint, only geodetic glacier ice volume change (Table , left column). This approach ensures that the total precipitation input remains unchanged. Thus, avoiding increases in runoff that might result solely from precipitation adjustments, overshadowing the effect of each forcing product on the model outcome. It simulates a data-scarce scenario where runoff data is unavailable for calibration. In contrast, the multi-data calibration involves optimizing both the ablation and accumulation parameters (Table , right column). In this case, geodetic volume change and measured annual runoff sums are constraints, representing a best-case scenario with additional data availability, allowing for more precise model tuning.

In this study, model calibration was performed over the period 2013–2021, which aligns with the availability of high-resolution geodetic glacier volume change data. This period serves as the calibration window for both the single- and multi-data calibration approaches. Although it is becoming increasingly common to also include snow cover or snow depth observations as additional constraints during model calibration e.g., we deliberately refrained from doing so in this study, as our aim is to replicate data-scarce conditions. In such settings, consistent and spatially representative snow observations are rarely available, whereas geodetic glacier volume change data are more accessible both locally and globally e.g.. To maintain methodological consistency with these conditions, we based the model calibration solely on geodetic glacier volume change in the single-data calibration, complemented by runoff observations in the multi-data calibration setup. Model evaluation (Section 3.6) was then conducted over the simulation period (2001–2022), allowing assessment of long-term model performance, seasonal variability, and year-to-year consistency. This fixed calibration–evaluation approach was selected to maintain consistency across experiments.

In our study, model accuracy refers to how well simulated glacier mass balance (annual and seasonal) and catchment runoff match corresponding observations over the historical evaluation period. Thus, to assess the influence of Experiment 1 and 2, and the effect of single versus multi-data calibration on the accuracy of the model results, we evaluate the simulated glacier mass balance and runoff against observational data for both. The runoff simulations are assessed against measured daily catchment runoff at Gletsch over the period 2001–2022, while glacier mass balance is evaluated using annual and seasonal measurements spanning 2007–2022. Note that, due to the spin-up of the hydrological storages (up to one year for groundwater in this catchment), the first simulation year was discarded from the runoff evaluation. The monthly and annual Nash–Sutcliffe efficiency (NSE), annual Kling–Gupta efficiency (KGE) and monthly relative difference (%) are used to quantify the agreement between observed and simulated runoff and capture seasonal variations. This model evaluation focuses on the melt season (April–September), when snow and glacier melt dominate the hydrological response. This period is most relevant to our study objectives, which centre on glacier-influenced hydrology. Winter runoff is excluded due to its limited relevance and higher associated uncertainty from low flows. Additionally, the partitioning of runoff into snow and ice melt is evaluated. Here we use the ice and snow runoff simulations forced with the Grimsel meteorological data as baseline to compare the other simulations with, as no measurements of these components are available. For glacier mass balance, the Root Mean Square Error (RMSE, in m w.e.) is calculated to evaluate the accuracy of simulated annual and seasonal mass balance relative to observations. We also evaluate the impact of single- versus multi-data calibration by comparing the model results from both calibration procedures.

The analysis of single-data simulations reveals distinct patterns in glacier mass balance across various scales and seasonal periods, with both the choice of forcing dataset and model resolution influencing the model results (Fig. A, B). In Experiment 1, the model runs utilizing ERA5-Land and MS_grid demonstrate the highest agreement for annual glacier mass balance (Fig. A), closely followed by the Grimsel dataset and ERA5-Reanalysis. All of them indicating an overall good model performance on the annual scale. However, winter glacier mass balance (Fig. B) is consistently underestimated relative to observational data, especially in simulations using gridded forcing products. This underestimation implies a compensatory underestimation effect on summer mass loss. For glacier area evolution (Fig. C), all datasets show a comparable rate of glacier retreat, although ERA5-Land forcing results in a lower glacier retreat rate.

Experiment 2 illustrates that simulations conducted at coarser model resolutions computed a more positive annual and winter mass balance than observed. Although the discrepancies are relatively limited (Fig. D, E), significant differences are noted regarding glacier area evolution (Fig. F). Model runs with a higher spatial resolution exhibit a gradual decline in glacier area, whereas coarser resolutions display, as expected, a more abrupt retreat, as much of the area is lost as soon as a glacier grid cell is removed. Furthermore, the initial glacier area in coarse-resolution simulations diverges from observed values by approximately ±2 km² (for the 1000 and 3000 m resolutions), while the glacier area for the 2000 m resolution remains constant.

Applying the multi-data calibration for Experiment 1 demonstrates no substantial changes at the annual scale, but significantly better agreement in winter for all simulations with different forcing products (Fig. ). This improvement indicates that a more accurate (positive) winter mass balance leads to a correspondingly more negative summer mass balance (closer to observed levels) to ensure consistency with the annual mass balance. When analysing the glacier area evolution, the simulations performed with ERA5-Land still produce the slowest glacier retreat, with the retreat being even more limited than with the single-data calibration. For simulations performed with ERA5-Reanalysis the glacier retreat rate also slows down, compared to before with the single-data calibration.

In the simulations for Experiment 2, model agreement with observations slightly increases at the annual scale but no relevant improvement was found for winter mass balance (see Fig. ). Specifically, at the 3000 m resolution, the agreement with winter mass balance is reduced, showing notably more negative values than the observations. This decline is not reflected in the evolution of glacier area. Whether single- or multi-data calibration is used, the outcomes remain the same. Thus, this emphasizes that regardless of the calibration method used here, the annual glacier mass balance does not change substantially. This consistency is primarily controlled by the calibration constrained with the geodetic ice volume change applied in both calibration procedures.

In Experiment 1 in combination with the single-data calibration, the summer catchment runoff is notably underestimated, no matter the applied forcing product, particularly in July and August (Fig. A, B). When forced with MS_grid and ERA5-Reanalysis, the model produces up to 20 % less ice melt than when forced with Grimsel (which yields the results that are most consistent with the observed total runoff). Simulation forced by ERA5-Land show the greatest underestimation at the onset of the melt season in May. It is important to note that, since these are relative values, even small differences during periods of low runoff (winter months) can result in high relative discrepancies. For snowmelt, simulations driven by ERA5-Land typically underestimate melt, while simulations driven by ERA5-Reanalysis tend to overestimate it when compared to Grimsel-forced results. The seasonal inconsistency in runoff totals is further underscored by the monthly NSE metric (Fig. ), focusing on April to October when runoff data is sufficiently reliable, as low winter flows introduce uncertainties and even small absolute differences between modelled and simulated can translate into large relative differences. From April to June, NSE values gradually increase, reflecting challenges at the onset of the melt season. During the main melt period, from June to October, NSE values remain between 0.6 and 0.8 for all simulations, indicative of moderate performance. However, a sharp decline in NSE is observed in September across all simulations, followed by an increase towards the winter months. Ultimately, across seasonal and annual scales, the single-data calibration consistently underestimated runoff (Fig. ). Similarly, the KGE values reflect a comparable fit, with the annual values indicating a bias primarily driven by the systematic underestimation of runoff. However, the year-to-year runoff variability was reasonably well captured, with simulated Coefficient of Variation (CV)-values slightly lower than the observed variability (Figure C). In the context of multi-data calibration, correcting the measured precipitation by +20 %–30 % (Table ) significantly reduced the runoff underestimation and improved summer runoff estimates. Here, a similar NSE evolution is observed with generally higher NSE values, particularly during April and May, indicating better alignment with observed runoff data at the melt season’s onset.

In Experiment 2, combined with single-data calibration, the findings indicate that as the model resolution coarsens, runoff becomes progressively too low and shifts temporally to later in the season (Fig. D). Analysing contributions from snow and ice melt underscores these patterns. At coarser resolutions (excluding 3000 m), ice melt is generally underestimated early in the season compared to high-resolution simulations (25 m). However, these coarser resolutions overestimate ice melt toward the end of the melt period, a trend mirrored in snowmelt behavior. The 3000 m resolution diverges significantly, consistently overestimating both total runoff and melt components, particularly during the early melt season, suggesting a temporal shift toward earlier melt timing. Monthly NSE-values from April to October further illustrate these seasonal trends (Fig. ). For most resolutions (except 3000 m), NSE declines from April to June, likely due to a seasonal shift in runoff timing (i.e., a delayed onset of melting), then improves markedly from June to August, before dropping again in September. In contrast, the 3000 m resolution exhibits stable NSE values from April to July but shows an earlier decline in August, likely reflecting a premature shift in runoff timing. On the annual scale, the overestimated runoff produced with the 3000 m resolution is also evident, though the year-to-year variability is well captured (Fig. ). In contrast, finer resolutions, up to 1000 m, align more closely with observed annual runoff in both magnitude and variability. When performing this Experiment in combination with the multi-data calibration, no notable improvement is observed for resolutions finer than 2000 m. However, for coarser resolutions (2000 m and above), simulated runoff increasingly underestimates annual totals, even with adjusted precipitation (Table  and Fig. ). This suggests that finer resolutions effectively capture annual runoff patterns. In contrast coarser resolutions struggle with runoff dynamics due to heightened sensitivity to melt timing and precipitation distribution.

The results of Experiment 1, combined with the single-data calibration, reveal a good agreement for annual glacier mass balance with all forcing products. However, consistent underestimation of winter snow accumulation on the glacier (Fig. ), particularly when forced with either of the Reanalysis products, lead to inaccuracies in capturing seasonal glacier mass balances. This is attributed to their lower estimates of winter and annual precipitation for this catchment (Fig. G, H), consistent with studies documenting precipitation biases in other alpine regions e.g.. similarly emphasized the importance of accurately capturing seasonal precipitation variability to represent snow accumulation processes in glacierized regions. The model compensates for winter precipitation deficits with a more positive summer glacier mass balance to maintain consistency with geodetic glacier ice volume change. This reflects findings by , who noted that errors in precipitation inputs in glacierized catchments are often offset by compensatory adjustments in glacier melt estimates. The single-data calibration achieves satisfactory annual glacier mass balance results. Seasonal dynamics, however, are poorly represented, particularly in seasonal glacier mass balance and in summer runoff, which is consistently underestimated (Figs. , ). This underlines the sensitivity of glacio-hydrological models to forcing data quality, as highlighted by . They identified precipitation inaccuracies as a primary driver of errors in hydrological simulations. Nonetheless, the year-to-year variability of runoff is captured well throughout the simulation period (Fig. ), which is mainly driven by temperature variability . In line with the finding that meteorological variables are the main source of uncertainty the parameter sensitivity analysis of GERM by in the Gletsch catchment, showed that constant retention and storage capacity parameters have a relatively minor impact compared to temperature lapse rate, precipitation correction, and ablation parameters. This justifies the decision not to calibrate reservoir-specific parameters individually. Instead, calibration efforts are best focused on accurately estimating temperature gradients and ablation dynamics, which contribute most significantly to uncertainty in runoff projections.

Multi-data calibration improves the simulation of seasonal glacier mass balance and runoff. Introducing a second constraint to the calibration – in this case, measured runoff – and adjusting the precipitation correction factor (Table ) results in better agreement with observations. Nevertheless, such corrections do not necessarily enhance the physical accuracy of modelled processes, as they rely on constant adjustments that fail to capture seasonal or spatial variability in precipitation patterns . Equifinality issues like these can obscure underlying deficiencies in the forcing data . Therefore, although the multi-data calibration improves the agreement with observations, it does not necessarily signify a more accurate representation of actual physical processes. It instead reflects the model's adjustment of forcing data and parameters to better align with the calibration data. However, the model’s representation of evapotranspiration provides a useful point of validation. While evapotranspiration plays a relatively small role in this high-alpine environment, it becomes relevant during summer in non-glacierized areas. Modelled annual evapotranspiration values (173–206 mm yr⁻¹) are consistent with the historical range of 131–240 mm yr⁻¹ reported by (Table S3), indicating that this process is well represented. This suggests that the main sources of uncertainty in summer runoff simulations are not due to evapotranspiration losses, but rather arise from reservoirs more directly affected by meteorological forcing – such as glacier and snow components – which are also more sensitive to calibration parameters. Similar to NSE, KGE values are highest for simulations forced with Grimsel data and decline when using gridded meteorological products (Fig. D). Multi-data calibration generally improves KGE values across all simulations. However, the ranking between forcing products remains consistent with that seen for NSE, reinforcing the conclusion that meteorological forcing quality impact the reliability of runoff simulations.

Furthermore, for all simulations a similar evolution of glacier area retreat between 2000 and 2022 is observed for all tested simulations. This is a gradual decrease in glacier area regardless of the calibration method. However, the magnitude of area change varies among simulations. For example, the model forced with ERA5-Land projects approximately 0.5 km² more glacier area remaining by the end of the simulation period compared to the average of other simulations, a difference that represents nearly one-third of the total glacier area change over the period. Variations stem primarily from differences in the calculated mean glacier mass balance, particularly before the calibration period. These are primarily driven by variations in the temperature and precipitation time series of each forcing product. These discrepancies in mean glacier mass balance affect the calculated glacier volume and result in divergent glacier area evolution across the various simulations. The spatial distribution of meteorological inputs over the glacier surface plays a critical role in driving the sensitivity of lower and higher-elevation areas to melt processes . Thus, simulated glacier area retreat can be disproportionally affected across extended periods by even small parameter combinations or meteorological forcing differences.

Capturing seasonal variability in precipitation inputs is one of the most important variables for accurately capturing glacier mass balance and runoff at various scales. While the application of the multi-data calibration procedure can improve seasonal accuracy, high-resolution and well-constrained forcing data are also needed to reduce unwanted parameter compensation. To further isolate the impact of the meteorological forcing, we conducted additional model runs without re-calibrating model parameters to each forcing product. The results (Figs. S3–S5) show that in the absence of calibration, the deviations between modelled and observed glacier area, mass balance, and runoff are even larger. Calibration reduces these differences but does not eliminate them, confirming that the choice of meteorological forcing product remains a primary driver of model performance.

In Experiment 2, the spatial resolution of the model does not significantly affect the computed annual glacier mass balance up to a resolution of 1000 m, while the winter glacier mass balance remains the same up to a resolution of 3000 m (Fig. ). Similarly, no substantial differences are observed in the annual catchment runoff up to a resolution of 1000 m (Fig. B). The observed strong decline in KGE (Fig. D) values with coarser spatial model resolution, larger than 100 m, supports this interpretation. While NSE primarily reflects timing and shape agreement, KGE additionally accounts for deviations in runoff magnitude and variability. Thus, the decreasing KGE at coarser resolutions emphasizes that errors in total runoff volumes increase as spatial detail is lost. Seasonal shifts occur, however, as the resolution becomes coarser, particularly at resolutions of 200 m and coarser (Fig. D, L). These shifts are linked to a delayed onset of snow and ice melt caused by the compression of glacier and catchment areas to higher elevations in the coarser resolution models. With coarser resolutions the model also considers elevations outside of the original catchment and glacier area (Fig. ), where colder temperatures prevail. Furthermore, the compression occurs, because lower glacier areas, which are typically glaciated to a smaller extent, tend to disappear as grid sizes become coarser, leaving only the extensively glaciated higher elevations. As a result, the mean glacier elevation shifts upward. One of the effects noticed from this, is the apparent glacier area stability observed for the 2000 m resolution. With an average higher glacier elevation, the low-lying ablation areas are under-represented, and the relative proportion of the accumulation area increases, leading to an unrealistically balanced mass budget and a stable glacier extent. At the coarsest resolution of 3000 m, the unexpected earlier onset of melt is likely a result of amplified ablation parameter adjustments necessary in the calibration procedure to compensate for these elevation biases, caused by the coarse spatial discretization, in both the single-data and multi-data calibration (Table ).

Furthermore, large grid cells aggregate varying elevations into a single value creating “steps” in the elevation distribution. As temperatures rise, these coarse cells abruptly contribute melt all at once, unlike smaller grid cells that allow a gradual melt progression as the 0 °C isotherm moves across elevation bands. observed similar shifts in melt dynamics. Larger grid cells smooth out rapid hydrological responses, making the timing of runoff less accurate .

Glacier extent and area changes further highlight the impact of model resolution on the model output. Coarser resolutions progressively lose fine-scale glaciological and topographic details. These include altitude, slope, and glacier hypsometry (Fig. ). At resolutions higher than 200 m, the model captures both the main glacier and ten smaller glaciers in the catchment. In contrast, at 1000 m, only one smaller glacier is resolved alongside the main glacier, and at 2000–3000 m resolution, only the main glacier remains, with smaller glaciers effectively excluded. This coarse representation limits the model’s ability to simulate the runoff contributions from smaller glaciated areas. Such limitations are particularly problematic in regions where small glaciers contribute significantly to seasonal runoff variability . Coarser spatial resolutions can sometimes provide a reasonable balance between model reliability and computational efficiency. However, the suitability of a model resolution depends heavily on the study objective and the glacier area within the catchment. Larger basins with more complex glacier dynamics may require higher resolutions or sub-grid parametrizations to ensure accurate projections of runoff and glacier volume changes . Furthermore, abrupt glacier area change at coarse resolutions reflects the stepwise retreat of large grid cells. This effect might average out over time though, as glacier melt dynamics at different altitudes interact with the coarse grid's smoothing effects . However, future projections neglecting finer spatial details could lead to underestimating glacier melt and runoff contributions .

The findings demonstrate that coarse model resolutions, while computationally efficient, can oversimplify critical glaciological processes, particularly for smaller glaciers. While the results are derived for a single, well-instrumented catchment, they hint at broader implications for modelling glacierized catchments under data-scarce conditions and with small glaciers.

The experiments presented in this study are subject to structural limitations of GERM. In the model, runoff is routed at the grid-cell level using a linear-reservoir approach . Each cell contains a set of reservoirs that represent different runoff components, depending on the local surface type. The runoff from each grid cell is computed individually and then aggregated at the catchment outlet. This structure allows for a spatially distributed simulation of runoff generation. At the same time, GERM uses some simplifications. Although spatial heterogeneity in the form of land-cover classes is explicitly represented, drainage networks and topographic flow pathways are not resolved . As a result, changes in spatial resolution primarily affect the mean catchment elevation and the distribution of land-cover classes, whereas flow directions, accumulation zones, and travel times remain unaffected. This is in contrast to fully distributed hydrological models, where resampling to coarser resolutions can significantly alter hydrological connectivity and runoff dynamics in complex terrain . In flat terrain, this resampling can affect flow directions . In contrast, in steep and spatially limited catchments such as Gletsch, this effect is expected to be minor – particularly at a daily temporal resolution – since most water reaches the outlet within a day. These simplifications do not diminish the value of the experiments, but they need to be considered when interpreting the results and when transferring the findings to other settings. In particular, applications in larger or more topographically complex basins may require model structures that explicitly resolve drainage pathways in order to capture the full sensitivity of the hydrological responses to spatial resolution.