Hydrological response of a headwater catchment in the semi-arid Andes (30°S) to climate change

In many arid and semi-arid mountainous regions, the cryosphere is an important hydrological source (Jansson et al. 2003; Viviroli et al. 2011; Jones et al. 2018), in that the resultant meltwater can represent between 50% and 95% of the total streamflow during the summer (Viviroli & Weingartner 2004). This contribution is essential to meet the demand of ∼ 1.9 billion people that live in these regions, as well as the local ecosystems and other economic interests (Mills-Novoa et al. 2017; Immerzeel et al. 2020; Viviroli et al. 2020; Kellner & Brunner 2021). However, in the periods 1980–2018, air temperatures in mountain chains increased by ∼0.4 °C per decade (Yao et al. 2022), which has altered the timing, amount and type of winter precipitation as well as melt, thereby directly impacting streamflow resulting from glaciers and snow accumulation (Barnett et al. 2005; Immerzeel et al. 2010; Huss et al. 2017). Globally, most glaciers are experiencing accelerated retreat and permafrost degradation has progressively increased (Milner et al. 2017; Biskaborn et al. 2019; Zemp et al. 2019; Dimri et al. 2021). Added to this is the impact of changing land use on the quality and quantity of available water (e.g. Wamucii et al. 2021). The combination of these processes directly affects all sectors of primary industry, especially agriculture and horticulture, and has caused an increase in food availability concern both for the present and future (Mancosu et al. 2015; Qin et al. 2020).

As changes are occurring relatively rapidly, there is a need to use a range of tools to project possible outcomes that allow for the differences of data availability and process understanding across the globe (Clark et al. 2016). For instance, while modelling is a practical tool to study the hydrological processes of basins (Ragettli et al. 2016), glacio-hydrological simulation has become a major challenge. This is due to difficulties in installing and maintaining hydrometeorological and glaciological monitoring networks in isolated regions (Pellicciotti et al. 2012; Ragettli et al. 2014, 2016; Masiokas et al. 2020). Another factor is the difficulty of accurately representing spatial and temporal variability of hydrological processes (Ruelland et al. 2011; Ragettli et al. 2014; Scaff et al. 2017), which means that often models do not reproduce hydrological processes with a significant degree of realism (Dussaillant et al. 2012; Pellicciotti et al. 2012). These challenges have generally limited the understanding of the physical processes that control runoff generation in mountainous areas (Hewitt 2011; Stehr & Aguayo 2017).

For the semi-arid Andes, several studies have focused on melting and sublimation of glaciers (e.g. MacDonell et al. 2013; Ayala et al. 2016) and snow (e.g. Gascoin et al. 2013; Réveillet et al. 2020), with few considering the hydrological implications of these processes (Stehr et al. 2009; Ragettli et al. 2014; Omani et al. 2017; Navarro et al. 2023a). These studies have been instrumental in understanding specific controls on climate–cryosphere interactions, but of limited value for quantifying streamflow expected downstream both at present and into the future.

Some applications of hydrological models developed for the semi-arid Andes have tried to reproduce and project runoff behaviour (e.g. Araya & Hunt 2003; Trigos & Munizaga 2006; Souvignet et al. 2008; Ruelland et al. 2011; Hublart et al. 2013, 2014, 2015). However, some of the corresponding results have a significant degree of uncertainty, in some instances due to not considering the influence of accumulated snow in winter (e.g. Araya & Hunt 2003) or by only using a limited time period (e.g. Trigos & Munizaga 2006; Souvignet et al. 2008). On the other hand, hydrological models that have considered the effect of snow and a broader temporal range (∼30 years), have managed to reproduce and project flows with a significant degree of realism (e.g. Ruelland et al. 2011; Hublart et al. 2014) and have made it possible to visualise the influence of snow on groundwater table dynamics (Ruelland et al. 2011). Undoubtedly, the contribution made by these works has had a significant impact on the knowledge and characterisation of runoff in their studied basins. However, these studies have not considered recent hydroclimatic variability, including the mega-drought of the last 13 years.

As long-term meteorological and hydrological datasets are largely absent from this region, and impacts of extreme weather events largely ignored, it is necessary to consider flexible alternatives to model water availability that can be used to develop water management plans that are widely applicable (Delpla et al. 2009; Beniston et al. 2011; Hill 2013). It is with these considerations in mind that the Inter-American Development Bank (IDB, or BID in Spanish) has developed the Hydro-BID for the Latin America and Caribbean (LAC) region (Andres et al. 2018). The advantages of this hydrological management model include that it can be applied in areas with data deficits and also to project the hydrological status of basins under climate change scenarios (Mena et al. 2021). Hydro-BID has been implemented in several countries in the LAC region (e.g. Moreda et al. 2014a, 2017; Wyatt et al. 2014). However, most studies have focused on basins with relatively regular relief and rainfall patterns (Arbuet et al. 2021; Mena et al. 2021). Given the utility of the tool for LAC generally, and that water generated in the Andes supplies >100 million people (Mills-Novoa et al. 2017), there is a need to evaluate how well Hydro-BID represents processes in Andean catchments. This paper represents the first step in this process by implementing the model in a relatively well-studied catchment in the semi-arid Andes of Chile. Here, the cryosphere plays an important role in water delivery (Navarro et al. 2023a), and so it is imperative that general models such as Hydro-BID are tested to evaluate their ability to represent these processes.

The main aims of this study are to analyse the hydrological contribution of snow in a semi-arid mountain system and how this may be impacted in future due to climate change and to evaluate the use of Hydro-BID in a high mountain environment. To address these aims, we use the Hydro-BID hydrological model to simulate daily, monthly and annual runoff in the upper Elqui River basin.

This section of the semi-arid Chilean Andes is characterised by a strong altitudinal gradient between the Pacific Ocean and the Cordillera, such that 6,000 m high peaks are located within 150 km of the coast (Figure 1). Above 3,000 m asl, glacial landforms dominate the landscape (e.g. moraines, U-shaped valleys and rock glaciers) (Monnier et al. 2014). These landforms are the result of changing fluvial, glacial and slope processes caused by glacial–interglacial oscillations during the Quaternary period (Zech et al. 2008; Riquelme et al. 2011; Aguilar et al. 2013). The underlying geology is characterised by granite and fractured volcanic outcrops from the Palaeozoic and Cenozoic periods, carbonates and sedimentary–volcanic material from the Jurassic and Inferior Cretaceous periods, and alluvial deposits from the Quaternary (Velásquez et al. 2021).

Between 1970 and 2009, the mean annual precipitation rate was 167 mm at the La Laguna station (3,200 m asl) (Monnier et al. 2014) which is concentrated in 5–10 events per year during the winter months (Rabatel et al. 2011). During precipitation events, the 0 °C isotherm is generally between 1,800 and 2,500 m asl, and the snowline lies approximately 280 m below this level (Schauwecker et al. 2022), which means that in the headwaters, precipitation is mainly in the form of snow. In recent years, the position of the snow line has increased at a rate of 10–30 m a⁻¹ (Saavedra et al. 2018) and during the last 50 years, the area where snow commonly falls has been reduced at an average annual rate of 12% per decade (Cordero et al. 2019). In the wider region (30–38°S), the last decade has been characterised by below-average precipitation rates and has been classified as a mega-drought (Garreaud et al. 2017, 2020).

Snow cover explains a large part of the annual variations in river flows in the semi-arid Andes (Masiokas et al. 2020). On an annual scale, snowmelt contributes between 66% and 93% of runoff in central Chile (Burger et al. 2019). Comparatively, during periods of drought, and the summer months, glaciers become more important for maintaining streamflow, and in central Chile, their contribution is equivalent to 67% of runoff in summer (Ayala et al. 2016). Sublimation from glaciers and snow is reasonably significant, and ranges from 36% to 86% (MacDonell et al. 2013; Voordendag et al. 2021), however, the ratio of sublimation to melt is dependent on elevation and seasonal conditions Réveillet et al. (2020); Voordendag et al. 2021).

Hydro-BID calculates runoff originating from precipitation, actual evapotranspiration, percolation, baseflow, deep percolation and stored water volumes in the saturated and unsaturated zones. These calculations are based on the generalised basin loading function (GWLF) (Haith et al. 1992), which integrates processes associated with snowmelt (SM), potential evapotranspiration (PET), runoff (R) and percolation (P) (Nalesso & Coli 2017). R is calculated through the curve number (NC). PET is estimated with daily temperature data and a coverage factor that depends on land use and vegetation cover. Baseflow and P are obtained from the groundwater recession and percolation constants, respectively. The volume of stored water is calculated from the daily hydrological balance, which enables runoff volume to be calculated at the basin and sub-basin scales (Moreda et al. 2014b). Limiting factors for the use of this model for applications in the semi-arid Andes is that it does not use an energy balance approach to model snowmelt, and ignores sublimation, the direct contribution of glaciers and the effect of snow transport by wind.

Hydro-BID uses hydroclimatic forcing data derived from daily time series of temperature (°C), precipitation (mm or cm) and discharge (m³/s) (Figure 2). In the present study, these datasets were obtained from nine meteorological stations and seven flow-monitoring stations, from the CAMELS-CL database (Alvarez-Garreton et al. 2018; Barría et al. 2021; Figure 2).

Both mean air temperature (MAT) and daily precipitation have continuous records over a 41-year period (1979–2020), however, the daily flow time series contains some gaps over that timeframe (see Supplementary Table S2). To avoid periods of poor data availability, the study period was shortened to 1986–2018, as it corresponded to the time period where discharge values were available from at least one station within the wider network (see Supplementary Table S2). The problem of temporal discontinuity in discharge measurements was mainly confined to the Derecho, Claro, Turbio and Elqui river stations (Figure 1). To replace missing data, linear regression models were used to fill missing discharge values based on stations that recorded values (see Supplementary Figure S1). Only regression models with R² values ​​ > 0.6, adjusted R² > 0.6, and standard error <10% were considered valid.

Shared Socioeconomic Pathways (SSPs) are climate change scenarios proposed by the IPCC that describe possible social, demographic and economic projections to the end of the century (Chen et al. 2020). The IPCC (2021) has proposed five potential scenarios based on the level of greenhouse gas emissions, mitigation measures and socioeconomic elements (denominated SSP1 through SSP5). SSP1 relates to a scenario where the use of green energy is encouraged (van Vuuren et al. 2014). Under SSP1, the air temperature will increase by less than 1.5 °C by 2100 (IPCC 2021). SSP2 is an intermediate pathway between SSP1 and SSP3 (Fricko et al. 2017). SSP3 is the result of the lack of global cooperation to face the climate crisis (Fujimori et al. 2017). In SSP3, a lower amount of greenhouse gas emissions is expected compared with SSP5. However, under SSP3, it is projected that by 2100 greenhouse gas emissions will be twice as high as today (IPCC 2021). Comparatively, the SSP4 scenario relates to the increase in inequality within, and between, countries (Calvin et al. 2017). Finally, SSP5 is the opposite extreme of SSP1 and is characterised by the indiscriminate use of fossil fuels, a rapidly growing economy and no implementation of mitigation measures (Kriegler et al. 2017).

The Hydro-BID climate change module (Figure 2) enables the simulation of hydrological response to changes in temperature and precipitation associated with the SSP proposed in the most recent IPCC report (IPCC 2021). For this study, precipitation data (1990–2020), monthly and annual temperature and precipitation projections were downloaded for SSP2-4.5 and SSP5-8.5, for the Coquimbo region from the World Bank database (Figure 2) (World Bank Group, Climate Change Knowledge Portal). The selected projection periods corresponded to 2040–2059, 2060–2079 and 2080–2099. Since the conditions projected by SSP1-1.9 and SSP1-2.6 are highly unlikely to occur, only scenarios SSP2-4.5 and SSP5-8.5 were used. SSP2-4.5 represents the best of the worst scenarios that could be achieved based on current conditions, whereas SSP5-8.5 is the worst possible future climate scenario.

The model was calibrated in the Turbio River sub-basin for the 1986–2012 period, which included part of the mega-drought (Garreaud et al. 2017, 2020). During calibration, several parameters were adjusted to account for regional processes (Figure 2). These included: curve number (NC), which characterises the use and hydrology of the soil; available water content (AWC), which refers to water available to plants in the soil; recession coefficient (CR), which determines groundwater contribution to streamflow following precipitation events in which discharge increases; losses (S), which is the exchange of water between groundwater units; evapotranspiration factors during periods with and without cultivation; temperature threshold (TT) and the snow melting factor (SMF). The validation of the calibrated model was carried out in the same sub-basin of the Turbio River and for the entire upper part of the Elqui River basin for 2012–2018.

The evaluation of model effectiveness included statistical and graphical techniques. These included analysing correlation coefficients (r), modified correlation coefficients (r_mod), Nash–Sutcliffe coefficients (NSE), the general volume error and a sensitivity analysis. For the coefficients r, r_mod and NSE (Figure 2), when their values are close to 1, they indicate that the degree of adjustment between the modelled discharge and the observed discharge is high. Comparatively, the general volume error indicates the mean error percentage between the simulated and observed values. In terms of error, an overall error of ≤±15% was considered acceptable (Moriasi et al. 2015). Note that the Hydro-BID model framework calculates these indices for daily and monthly runoff calculations, but to calculate annual values of the performance indices, the hydroGOF R package was used (Zambrano-Bigiarini 2020). Graphical model evaluation included evaluating the degree of fit between the simulated and observed water volume curve (Figure 2). The sensitivity analysis considered the value of the root mean square error (RMSE). For this, the minimum and maximum values of the range of values for each parameter as suggested by Nalesso & Coli (2017) were evaluated.

The calibration resulted in new values of several parameters (see Supplementary Table S3), such that all except for NC, S and TT were adjusted within the range recommended by Nalesso & Coli (2017). Of those outside the normal range, the adjusted value of TT was the most notable, such that a value of 4 °C was used which reflects the fact that precipitation in the mountainous region of Elqui is predominantly snow. While outside the suggested range, this adjustment is in line with other studies in similar environments (e.g. Ruelland et al. 2011; Kraaijenbrink et al. 2021).

The calibrated model results for the Turbio River sub-basin show that Hydro-BID accurately simulates measured streamflow (see Supplementary Figure S2), as seen in the R² of ∼0.7 for the daily and monthly flows, while for the annual flow it is ∼0.8 (see Supplementary Table S4). This means there is a high degree of fit between the values of the simulated daily, monthly and annual hydrographs with the observed data (see Supplementary Figure S2(a)–S2(c)). However, the maximum flows related to periods with higher precipitation are underestimated by the model (see Supplementary Figure S2(a) and S2(b)). Regarding r_mod, the values ​​are >0.5, which indicates that the model explains relatively well a large part of the variance of the simulated discharge with respect to the observed discharge. Comparatively, the NSE value is ∼0.7 for the simulated daily, monthly and annual flow data and the results of the calibrated model show an overall error of ∼− 2%, ∼− 1% and ∼2% for the simulated daily, monthly and annual flows, respectively. Finally, Supplementary Figure S2(d) shows the degree of fit between the simulated and adjusted water volume curves is high.

Moriasi et al. (2015) showed that the daily, monthly or annual results of a basin-scale hydrological model can be defined as ‘satisfactory’ when the value of R² > 0.6, NSE > 0.5 and the general error is ≤± 15%. According to this standard, the calibrated and validated model results can be classified as satisfactory. This is supported by the low RMSE value, and the high degree of fit shown in Figure 3 and Supplementary Figure S2.

In the context of the hydrological year, the observed data (April to March 1990–2020) show that the maximum flow is reached during December (Figure 5). However, at different time periods under the SSP2-4.5 trajectory, peak flow will be reached during September.

Following SSP5-8.5, the mean annual temperature is projected to increase by 1.5, 2.6 and 3.7 °C for the periods 2040–2059, 2060–2079 and 2080–2099, respectively, whereas precipitation will decrease by 28% (2040–2059), 39% (2060–2079) and 44% (2080–2099). In the case of SSP5-8.5, the projected results also show significant annual, monthly and daily changes in streamflow in the study area. In the context of SSP5-8.5 for 2040–2059, 2060–2079 and 2080–2099, the annual average discharge is projected to decrease by 57%, 70% and 77%, respectively (Table 1). In the case of monthly average flows, a decrease of 57% is estimated for the period 2040–2059 (Figure 5(d)), a decrease of 70% for 2060–2079 (Figure 5(e)) and a reduction of 76% for 2080–2099 (Figure 5(f)). For daily average discharge, the period 2040–2059 experiences a reduction of 58% (Table 1), while for 2060–2079, the reduction is 71% and for 2080–2099 it decreases by 77%.

In the case of the hydrological year, the same trend is observed as modelled in SSP2-4.5 (Figure 5). That is, in the different periods analysed for the SSP5-8.5 trajectory, the maximum flow during the hydrological year will be reached during September.

Hydrological modelling using Hydro-BID for a site with abrupt morphology such as the semi-arid Andes of Chile is not a simple task to undertake, mainly because morphological parameters cannot be adjusted in the model set-up. However, Hydro-BID has other elements that make it possible to simulate runoff for a mountain context, which enables the implementation of the model in Andean catchments. In this study, that meant adjusting some parameters outside the standard ranges and made it possible to improve model performance (see Supplementary Table S2). However, it was not until the modification of TT that the model was able to satisfactorily model the hydrology of the study region. The TT parameter defines the temperature at which precipitation is liquid or solid, and during the calibration stage, it was increased iteratively from 1 °C (default) to 4 °C. The modification of TT was based on two considerations. The first was related to the recognition that the hydrological system is driven predominantly by a snow regime in the Elqui River basin (Favier et al. 2009; Hublart et al. 2014; Balocchi et al. 2017). Secondly, most of the meteorological stations used in this study are located at relatively low elevations, and record events below the snow–rain transition zone (Schauwecker et al. 2022). By increasing TT, Hydro-BID was forced to incorporate a higher proportion of solid precipitation. This meant that the hydrological model reproduced with a greater degree of reality the natural precipitation process that dominates in the headwaters of Chilean semi-arid Andean basins. By allowing for this process, Hydro-BID simulates reasonably accurately the effect of snowmelt on the outflow of the upper part of the Elqui River (Figure 4). This coincides with what has been proposed by other studies (Ruelland et al. 2011; Hublart et al. 2013) and demonstrates the influence of snow on runoff from the upper part of the Elqui River basin. However, unlike previous studies in the area, Hydro-BID allows this process to be reproduced considering solid precipitation, not only liquid precipitation (Figure 4).

The results of the discharge projections show that for all scenarios, streamflow decreases, which is consistent with the trend projected for South America by the IPCC (2021). However, Figure 5 also shows an evident change in the timing of annual peak flow. At present (1990–2020), the maximum annual flow is achieved in December (Figures 4(b) and 5); however, in future, the annual maximum will be reached during September (Figure 5). It is likely that this change is due not only to a decrease in precipitation rates, but also to a switch in the fraction of precipitation falling as snow and as rain. Given that snowmelt is likely the dominant control in semi-arid Andean basins (e.g. Ayala et al. 2016; Burger et al. 2019; Masiokas et al. 2020), any modification to snow processes will have an immediate impact on discharge (Barnett et al. 2005; Huss et al. 2017). This is consistent with work carried out in the semi-arid Andes, where the reduction in snow-covered area and permanence has already been observed (Saavedra et al. 2018; Cordero et al. 2019). In fact, the model results already show indications of these changes. In Figure 4(c) and 4(d), it is observed that between 2012 and 2018, the maximum snow accumulation was lower and was reached in July. This has meant that, during this period, the maximum flow is lower and often occurs in September, which is three months earlier than what was observed between 1986 and 2012 (Figure 4(a) and 4(b)). This change is likely due to the mega-drought that has affected the wider region (Garreaud et al. 2017, 2020). The local effect of the mega-drought has been compounded by the increase in mean annual temperature related to climate change (Souvignet et al. 2012). The results of Hydro-BID in the Elqui River basin are evidence that a lower amount of snowfall, both now and in the future, significantly influences the amount of water available for the population, ecosystem and economic activities.

Due to ongoing water scarcity in the region, the Watershed Sustainability Index (WSI) categorises the Elqui River basin as having an intermediate degree of sustainability (Cortés et al. 2012). Following Hublart et al. (2013), ongoing scarcity will also be extended to annual flows in the future. For the period 2041–2060, Hublart et al. (2013) estimated that the maximum flow will be reached in November and the annual volume of water will be reduced by between 30% and 70%. This range is relatively wide, and less constrained than the results of Hydro-BID for the period 2040–2059, where a reduction in the annual flow is projected to be 54% and 57% for SSP2-4.5 and SSP5-8.5, respectively (Table 1). In contrast to the work of Hublart et al. (2013), the Hydro-BID results show that the maximum flow will be reached in September (Figure 5). These differences between both models respond to two main drivers. The first is related to the fact that Hydro-BID considered part of the effect of the mega-drought in its calibration period, a period not yet experienced in the dataset used in Hublart et al. (2013). Secondly, the input forcing data used in Hydro-BID are derived from climate projection models that have a lower degree of uncertainty than those used by Hublart et al. (2013). However, both models project significant decreases in streamflow in the Elqui River basin at daily, monthly and annual time scales (Figure 5 and Table 1). With such a result, it is also necessary to reflect on the contribution of water derived from other sources in the catchment headwaters.

This study has focused on seasonal snow contribution to streamflow, but snow also relates to the hydrological role of glaciers, rock glaciers, snow and aquifers over a variety of timeframes. For instance, without snowfall events, it is not possible for some of these mountain components to be generated or preserved in the long term (Gascoin et al. 2013; Kinnard et al. 2020). Therefore, precipitation is the fundamental water source and the engine of the hydrological system in a catchment (Levizzani & Cattani 2019). Mountain water resources fulfil the function of redistributing precipitation accumulated in winter to runoff during spring and summer (Viviroli et al. 2011). Therefore, glaciers, rock glaciers, snow and aquifers fall into the category of freshwater reserves (Immerzeel et al. 2020; Viviroli et al. 2020), which can be categorised as medium or long term. The former has the potential to cover the demand for water on a daily, monthly and annual scale (e.g. snow) (Jansson et al. 2003). Long-term reserves have the capacity to contribute water for a period of time that varies from months to centuries (e.g. glaciers) (Jansson et al. 2003). The role of rock glaciers in this context is largely unclear (Jones et al. 2018, 2019).

Hydro-BID projects a decrease in the medium-term reserve (snow) in the study area. Therefore, the long-term freshwater reserves of the Elqui basin will become essential in facing a future with less precipitation and snow accumulation. The presence of glaciers in the Elqui basin is limited and most ice is contained within one glacier, the Tapado Glacier, which is located in the upper part of the Elqui basin (Robson et al. 2022). Pourrier et al. (2014) identified that the hydrological contribution from the Tapado Glacier complex is important for meeting the demand in the lower areas of the Elqui River basin, especially during the summer and periods of drought. However, on a global scale, glaciers have been progressively losing mass (Zemp et al. 2019) and glaciers in the semi-arid Andes are no exception (e.g. Kinnard et al. 2020; Masiokas et al. 2020). In the case of the Tapado Glacier, during the last ∼ 64 years, its surface has been reduced by 20%–30% and the mega-drought has increased the frequency and size of negative mass-balances (Robson et al. 2022). As precipitation is the main driver of mass-balance change in these systems (Kinnard et al. 2020), the reduction in the amount of accumulated snow projected by Hydro-BID implies greater pressure on permanent ice masses such as Tapado. This increased pressure will further contribute to the loss of glacier mass, and increase the probability that a long-term water reserve such as the Tapado Glacier will become a medium-term water reserve in the future.

For rock glaciers, the outlook is different, as they are a form of periglacial landform composed of a mixture of water, ice and debris (Berthling 2011). Their categorisation as potential long-term freshwater reserves (Jones et al. 2018, 2019) responds more to their detrital coverage than necessarily their ice content. As sediment acts to insulate the massive or interstitial ice stored inside from positive air temperatures and the effects of radiation, it is likely that these landforms will be less affected by the effects of climate change (Jones et al. 2018; Schaffer & MacDonell 2022). Another advantage of the rock cover is that it causes the hydrological discharge of water to be gradual during the summer (Krainer & Mostler 2002; Harrington et al. 2018). In the case of the upper part of the Elqui River basin, rock glaciers are distributed in large numbers and, therefore, occupy a significant surface area (Barcaza et al. 2017; Schaffer et al. 2019). It has been estimated that the exposed glaciers and rock glaciers in the upper part of the Elqui basin represent 0.1 and 0.5 km³ of water equivalent, respectively (Schaffer et al. 2019). In terms of contribution, it has been estimated that rock glaciers can contribute between 9% and 20% of the annual discharge of the La Laguna sub-basin (Schaffer et al. 2019). Likewise, in the study area, geophysical data indicate that rock glaciers can have significant ice and liquid water content (de Pasquale et al. 2022; Navarro et al. 2023b). In fact, Monnier & Kinnard (2015) estimated an average ice fraction of 66% in a rock glacier located in the La Laguna sub-basin.

Processes related to glaciers and mountain permafrost are not directly considered by Hydro-BID. It is clear that the significant decrease in the Elqui River basin discharge (Table 1 and Figure 5) can be associated with reduced snow accumulation. Looking forward, it is possible that the projected decrease in flow (Figure 5) will not be as severe in the mid-term due to the presence of glaciers and rock glaciers. However, to substantiate this hypothesis, it is essential to develop more studies that allow quantification of the hydrological significance of rock glaciers and permafrost environments (Hilbich et al. 2022). A better understanding of permafrost processes and the interaction between snow and the underlying terrain will greatly improve the application of hydrological models in high mountain environments.

Hydro-BID is a hydrological model developed to support water management in low-lying catchments in LAC (e.g. Arbuet et al. 2021; Mena et al. 2021). However, the results of this work have demonstrated its applicability in a mountainous region such as the semi-arid Andes (Figures 3 and 4). This suggests that Hydro-BID has the capacity to estimate and project runoff in mountainous regions of LAC, which could help to refine water management plans in, and stemming from, these areas. The applicability of Hydro-BID is primarily related to the AHD, which enables Hydro-BID to be applied in areas where there is limited data availability (Mena et al. 2021). The AHD has made it possible to compensate for the data deficit of other models and integrate their results with Hydro-BID (e.g. Mena et al. 2021).

In general, in the Andes, semi-distributed and distributed hydrological models have been used to solve water management and research problems (Ragettli et al. 2014). The former are simple and only need temperature data to differentiate solid from liquid precipitation (Omani et al. 2017). However, their results may be altered as a result of compensation errors, since there is no total reproduction of high mountain processes (Pellicciotti et al. 2012; Ragettli et al. 2014). Distributed models guarantee accuracy and less dependency on calibration (Pellicciotti et al. 2012). However, they are subordinate to the availability of data (Ragettli & Pellicciotti 2012). The results of the present study show that, unlike distributed models that have been applied in the Andes (e.g. Ragettli & Pellicciotti 2012; Krogh et al. 2015; Ayala et al. 2016; Ragettli et al. 2016; Mernild et al. 2018; Burger et al. 2019), Hydro-BID cannot directly quantify the contribution derived from snow and glaciers. However, using a simplified approach, it has demonstrated the importance of snow for runoff in the region and has the ability to project flow based on future climate scenarios (Figure 5). In fact, Hydro-BID satisfactorily reproduces the flows observed in the Elqui River basin to the same degree as other semi-distributed hydrological models applied in the Andes (Stehr et al. 2009; Ruelland et al. 2011; Ragettli et al. 2014; Omani et al. 2017; Escanilla-Minchel et al. 2020).

Model simplicity enables Hydro-BID to provide an indication of the general behaviour of runoff from a mountain basin in a short timeframe. Hydrological characterisation using Hydro-BID can be used as an initial step, before addressing more complex high mountain hydrological problems with distributed hydrological models (e.g. TOPKAPI, SnowModel, CRHM). Similarly, due to its extensive database, Hydro-BID offers the opportunity to solve the problems that some models have presented regarding the availability of local data on land use and type (e.g. Stehr et al. 2009). Considering that Hydro-BID is a flexible model into which climate change, reservoirs, sediment transport and groundwater modules have been incorporated (Moreda et al. 2016), for mountainous regions, it would be good to develop a specific mountain module or set of parameterisations that incorporate glaciers and permafrost to make Hydro-BID an even more complete high mountain hydrological model.

This study has reinforced the importance of snow for runoff in the semi-arid Andes of Chile (32°S). Additionally, it has demonstrated that the Hydro-BID model can satisfactorily simulate and project runoff in a catchment that is highly dependent on the contribution of snowmelt over the course of a hydrological year. In addition, despite being a relatively simple model, Hydro-BID has the ability to reproduce both solid and liquid precipitation in a basin and the resultant hydrological implications. This suggests that Hydro-BID can be used as a simple tool to understand the general behaviour of runoff in high mountain, Latin American basins. Therefore, Hydro-BID can support water management planning in areas where water is generated in the mountains, and where projections of future inputs are required.

The modelled projections for both SSP2-4.5 and SSP5-8.5 scenarios indicate that for the next 40, 60 and 80 years, the daily, monthly and annual runoffs will decrease between 50% and 77%. However, the most significant change is the month in which the maximum flow is reached. In future, the peak flow will be reached three months earlier than in the present. This means that there will likely be greater hydrological pressure on other long-term water reserves present in the semi-arid Andes, such as glaciers and rock glaciers.

This study reinforces the need to ensure even simple hydrological modelling platforms such as Hydro-BID include modules or parameters that consider cryospheric processes. This consideration will strengthen the simulation of high mountains that are easily integrated into a water management framework. However, there is additionally an opportunity to improve the current performance of the model. At present, the meteorological station data used to drive Hydro-BID is primarily based on stations located in valley floors. These datasets are used to create interpolated climate datasets, therein creating additional uncertainty due to station location bias. Exploration of new ways to interpolate climate datasets should be explored, which may include the use of virtual stations in summit areas. The drawback of this solution is the need to use reanalysis data (e.g. ERA5). However, with a robust method of bias correction, the data from these potential virtual stations will have the potential to better represent the temperature and precipitation rates of headwater regions and therein decrease the uncertainty of the Hydro-BID results.

We thank ANID-FONDECYT (1181540) and ANID-CENTROS REGIONALES (R20F0008) (RL3) projects for supporting this work. E.Y. would like to thank María San Francisco y Waldo Yáñez for their infinite patience and unconditional support. We would also like to thank Dr Mauro Nalesso and Theresa Schutz for their willingness and time to teach E.Y. how to use Hydro-BID during the ‘Getting to know Hydro-BID’ technician course.

All relevant data are available from an online repository or repositories. These include the CAMELS repository: https://camels.cr2.cl/, and the World Bank repository: https://climateknowledgeportal.worldbank.org/download-data.

The authors declare there is no conflict.